Superconducting thin film nanoelectronics

by
Adam Nykoruk McCaughan
Submitted to the Department of Electrical Engineering and Computer
Science
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Electrical Engineering
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
September 2015
○
c Massachusetts Institute of Technology 2015. All rights reserved.

Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Department of Electrical Engineering and Computer Science
August 31, 2015

Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Karl K. Berggren
Professor of Electrical Engineering and Computer Science
Thesis Supervisor

Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Leslie A. Kolodziejski
Chairman, Department Committee on Graduate Theses
2
 Superconducting thin film nanoelectronics
by
Adam Nykoruk McCaughan

Submitted to the Department of Electrical Engineering and Computer Science

on August 31, 2015, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy in Electrical Engineering

Abstract
Superconducting devices have found application in a diverse set of fields due to their
unique properties which cannot be reproduced in normal materials. Although many
of these devices rely on the properties of bulk superconductors, superconducting de-
vices based on thin films are finding increasing application, especially in the realms of
sensing and amplification. With recent advances in electron-beam lithography, super-
conducting thin films can be patterned into geometries with feature sizes at or below
the characteristic length scales of the superconducting state. By patterning 2D ge-
ometries with features smaller than these characteristic length scales, we were able to
use nanoscale phenomena which occur in thin superconducting films to create super-
conducting devices which performed useful tasks such as sensor amplification, logical
processing, and fluxoid state sensing. In this thesis, I describe the development, char-
acterization, and application of three novel superconducting nanoelectronic devices:
the nTron, the yTron, and the current-controlled nanoSQUID. These devices derive
their functionality from the exploitation of nanoscale superconducting effects such
as kinetic inductance, electrothermal suppression, and current-crowding. Pattern-
ing these devices from superconducting thin-films has allowed them to be integrated
monolithically with each other and other thin-film superconducting devices such as
the superconducting nanowire single-photon detector.

Thesis Supervisor: Karl K. Berggren

Title: Professor of Electrical Engineering and Computer Science

3
4
Acknowledgments
The work performed during my graduate school career was only possible thanks to the
support, advice, and contributions of my colleagues, family, and friends. In particular,
I would like to thank:
My advisor Karl Berggren for his enthusiasm for science, his dedication to pro-
fessional and personal development, as well as his remarkable intuition. Under his
direction both my creative abilities and discipline thrived.
Professor Terry Orlando and Professor Rajeev Ram for agreeing to be on my thesis
committee, the excellent classes they taught, and the great advice they’ve given me
over the years.
Isaac Chuang, for guiding me during my early years of graduate school and showing
me how scientific inquisitiveness and technical discipline strongly reinforce each other.
Mark Mondol, Jim Daley, and Tim Savas for their expertise and support with
everything NSL and nanofabrication-related.
Faraz Najafi, Qingyuan Zhao, Andrew Dane, Francesco Bellei, Nate Abebe, Jake
Mower, Di Zhu, David Meyer, Nick Harris, Luca Alloatti, and Francesco Marsili for
their many helpful discussions, collaborations, and experimental assistance.
Yachin Ivry and Richard Hobbes for their advice and insight into chemistry and
other topics well outside the realm of my own research.
My colleagues Michael Gutierrez, Arolyn Conwill, Stephan Schulz, Anders Mortensen,
Amira Eltony, Yufei Ge, and Paul Antohi.
My mother, father, and sister, who have always inspired me and have always
supported my intellectual development. A large part of where I am today is due to
their support, and I am overwhelmingly grateful. My father especially, for sharing
with me his enthusiasm for research and never hesitating to support my scientific
curiosity.
My wonderful fiancée (and creative muse) Cammy, who I am very excited to marry
this October. Her support kept the world turning even when the demands of graduate
school seemed overwhelming.

5
6
Contents

1 Introduction to superconducting devices 23

1.1 Phase- and magnitude-based devices . . . . . . . . . . . . . . . . . . 24
1.1.1 Phase-based devices . . . . . . . . . . . . . . . . . . . . . . . 25
1.1.2 Magnitude-based devices . . . . . . . . . . . . . . . . . . . . . 26
1.2 This thesis: Thin-film nanoelectronic devices . . . . . . . . . . . . . . 28

2 Nanoscale superconducting devices 31

2.1 The effects of scaling down superconductors . . . . . . . . . . . . . . 32
2.1.1 Characteristics of bulk superconductors . . . . . . . . . . . . . 32
2.1.2 Superconducting thin films and kinetic inductance . . . . . . . 34
2.2 Applications of thin superconducting films . . . . . . . . . . . . . . . 36

3 The 2D electrothermal model 39

3.1 Electrothermal basics . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.1 The heat equation . . . . . . . . . . . . . . . . . . . . . . . . 40
3.1.2 Modifying the basic heat equation . . . . . . . . . . . . . . . . 41
3.2 Adapting the 1D model to 2D . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Variables, equations, and parameters . . . . . . . . . . . . . . . . . . 45
3.3.1 Electrothermal variables . . . . . . . . . . . . . . . . . . . . . 45
3.3.2 Boundary conditions and initial values . . . . . . . . . . . . . 50
3.4 Conclusions and future work . . . . . . . . . . . . . . . . . . . . . . . 51

4 The current-biased nanoSQUID 53

7
 4.1 Device characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2 Analysis of the nanoSQUID switching current . . . . . . . . . . . . . 56
4.2.1 Analysis of the nanoSQUID switching current . . . . . . . . . 56
4.2.2 Visualizing the analysis . . . . . . . . . . . . . . . . . . . . . . 57
4.3 Measurements and results . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4 Results analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.5 Minimizing Lk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.6 Variations in the behavior of the nanoSQUID . . . . . . . . . . . . . 64
4.6.1 Material dependency . . . . . . . . . . . . . . . . . . . . . . . 64
4.7 Use of the nanoSQUID geometry as an Lk metrology tool . . . . . . . 67

5 The nanocryotron (nTron) 69

5.1 Challenges of superconducting circuitry . . . . . . . . . . . . . . . . . 70
5.2 nTron device description . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2.1 Fabrication of the nTron . . . . . . . . . . . . . . . . . . . . . 72
5.3 nTron operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.4 Simulation and design parameters . . . . . . . . . . . . . . . . . . . . 75
5.5 Digital applications and characterization . . . . . . . . . . . . . . . . 76
5.5.1 nTron logic gates . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.5.2 The nTron half-adder . . . . . . . . . . . . . . . . . . . . . . . 78
5.5.3 Measuring the threshold sensitivity of the nTron . . . . . . . . 79
5.5.4 Measurement of a pseudo-eye-diagram at 10 MHz . . . . . . . 79
5.5.5 Integration with a superconducting nanowire single-photon de-
tector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5.6 Power dissipation and clock rate . . . . . . . . . . . . . . . . . 82
5.6 High-Tc YBCO nTrons . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.6.1 Cryostat experimental setup . . . . . . . . . . . . . . . . . . . 83
5.6.2 YBCO nTron results . . . . . . . . . . . . . . . . . . . . . . . 84
5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

8
6 The current-crowding cryotron (yTron) 93
6.1 Device description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Device operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.1 Current crowding and the channel critical current . . . . . . . 95
6.2.2 Output characterization . . . . . . . . . . . . . . . . . . . . . 97
6.3 Device design considerations . . . . . . . . . . . . . . . . . . . . . . . 100
6.3.1 Material considerations . . . . . . . . . . . . . . . . . . . . . . 100
6.3.2 Geometric considerations . . . . . . . . . . . . . . . . . . . . . 100
6.4 Operating modes of the yTron . . . . . . . . . . . . . . . . . . . . . . 101
6.4.1 Isolation of the gate from the channel . . . . . . . . . . . . . . 102
6.5 Measurement details . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.6 Inline, nondestructive measurements of a quantized superconducting
loop current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.7 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

7 Experimental techniques 109

7.1 Measurement and automation . . . . . . . . . . . . . . . . . . . . . . 109
7.1.1 Critical current measurements . . . . . . . . . . . . . . . . . . 109
7.1.2 Measurement automation with Python . . . . . . . . . . . . . 118
7.2 Sample holder design . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.2.1 Sample holder construction . . . . . . . . . . . . . . . . . . . 120
7.2.2 In-house PCB fabrication . . . . . . . . . . . . . . . . . . . . 124

8 Conclusion and outlook 131

A Python equipment automation code 133

9
10
List of Figures

3-1 Diagram showing the production and flow of energy between the dif-
ferent coupled systems in the electrothermal model. . . . . . . . . . . 42

3-2 A graph of Eq. 3.5, plotting the superconducting bandgap Δ versus

temperature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3-3 A graph of Eq. 3.6, showing the electron specific heat ce versus electron
temperature Te . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3-4 A graph of Eq. 3.7, showing the electron thermal conductivity 𝜅e versus
temperature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3-5 (above) A graph of Eq. 3.8, showing the resistivity 𝜌 versus tempera-
ture. (below) A graph of the inverse of Eq. 3.8, showing the conduc-
tivity 1/𝜌 versus temperature. . . . . . . . . . . . . . . . . . . . . . . 48

3-6 A graph of Eq. 3.9, showing the critical current density Jc versus tem-
perature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3-7 A graph of Eq. 3.10, showing the electron-phonon interaction time 𝜏e-ph
versus temperature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3-8 A graph of Eq. 3.11, showing the phonon specific heat cph versus
phonon temperature Tph . . . . . . . . . . . . . . . . . . . . . . . . . . 50

11
4-1 (left) Scanning-electron micrograph of a current-controlled nanoSQUID
device, fabricated from a thin niobium film. Inset shows a closeup
of one of the nanoSQUID constrictions, which were measured to be
105 nm wide at their narrowest point. (right) Equivalent circuit of the
nanoSQUID device. Shown are the four terminals of the device and
their inputs. I bias , which is used to measure the switching current
of the device, flows in from terminal 1 at the top and is carried out
through terminal 4 at the bottom. The modulation current Imod enters
and leaves through the terminals 2 and 3 on the right. Isym and Iloop
are the symmetric and circulating components of Imod , respectively. . 55

4-2 Graph plotting the solution to the nanoSQUID inequalities for 𝑟 = 0.5,
𝐼0 = 0.5, 𝐼𝑠𝑤
𝑅
= 1, and 𝑛 = 0. (left) Graph of the boundaries generated
by the inequalities in Eq. 4.5. (right) Graph of the area which solves
all four inequalities in Eq. 4.5 . . . . . . . . . . . . . . . . . . . . . . 57

4-3 Graph plotting the solution to the nanoSQUID inequalities for 𝑟 = 0.5,
𝐼0 = 0.5, 𝐼𝑠𝑤
𝑅
= 1 (left) Graph of the valid regions for 𝑛 = 0 and 𝑛 = 1.
(right) Graph of the valid regions for all integer values of 𝑛 . . . . . . 58

4-4 Results of the analysis of the nanoSQUID for 𝑟 = 0.5, 𝐼0 = 0.5, 𝐼𝑠𝑤
𝑅
=
1 and all integer values of 𝑛, plotted with the total I gate + I bias which
better corresponds to the experimentally measured switching current. 58

4-5 Results of the analysis of the nanoSQUID based on different parameter

inputs. (left) Graph generated when the splitting ratio is 𝑟 = 0.8.
(middle) Graph generated when the flux induced current 𝐼0 was set to
= 0.2. (right) Graph generated when the constriction critical currents
were asymmetric, such that 𝐼𝑠𝑤
𝑅
= 3. . . . . . . . . . . . . . . . . . . . 59

12
4-6 Experimental results of the nanoSQUID being modulated by injected
current. Shown is the distribution of the nanoSQUID switching current
(Isw ) varying as a function of the injected modulation current (Imod ).
Each vertical slice of the graph corresponds to a a measurement of the
Isw distribution for that value of Imod . (inset) Two slices showing the
distribution of Isw when maximally and minimally modulated by Imod . 60

4-7 Hourglass nanoSQUID geometry designed to be as low-inductance as

possible. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4-8 Figure showing how a bridge which nominally comprises only a few
squares actually has more squares due to the path of the current flow.
On the left is a simulation of current flowing across a narrow constric-
tion. It appears to be about 1 square in total, but the simulation
reveals it is more than 3 squares total. This is due a majority of the
current taking an hourglass path, shown by the streamlines on the left,
and represented geometrically on the right. . . . . . . . . . . . . . . . 63

4-9 Measurements of the switching distributions for a NbN nanoSQUID.

This nanoSQUID was biased using an induced magnetic field from a
solenoid instead of current-biased. . . . . . . . . . . . . . . . . . . . . 65

4-10 Diagram showing the nSQUID states as I bias is increased. First, I gate
is set and I bias is zero (point A) – the device begins in the state 𝑛 = 2.
Next, I bias increased until it reaches the boundary of the 𝑛 = 2 state
(point B). At this time, if the device is hysteretic, it may create a
hotspot and latch. Otherwise, the device will transition to the 𝑛 = 1
state by ejecting a fluxoid, and as I bias is further increased it exits the
valid region and forms a hotspot (point C). . . . . . . . . . . . . . . . 66

5-1 (A) Three-terminal circuit symbol. The position of the gate arrow
denotes the location of the choke relative to the narrowing of the chan-
nel. (B) SEM of a fabricated nTron, the inset depicts a close-up of the
choke, the area in which the resistive hotspot is first formed. . . . . . 73

13
5-2 Circuit schematic and output characteristics for an nTron in a non-
inverting amplifier configuration. I gate was fixed and I bias was swept
from 0 to 120 µA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5-3 Numerical simulation of the nTron depicting the three states of oper-
ation. OFF state: The device is fully superconducting, bias current
is drained through the channel to ground. Transition state: Current
is added to the gate input, forming a resistive hotspot which locally
suppresses superconductivity. (inset, upper) Closeup of the resistive
hotspot forming in the choke. (inset, lower) Contour map of Jc sup-
pression extending from the hotspot. From inner to outer, the bands
represent reductions in Jc by 0 % (blue), 25 % (light blue), 50 % (green),
75 % (orange), and ≥99 % (magenta). ON state: The critical current
of the channel is reduced sufficiently that the bias current triggers the
formation of a resistive hotspot in the channel. . . . . . . . . . . . . . 75

5-4 Digital gates based on the nanocryotron. (A) Schematic of a set of

universal logical gates from the basic three-terminal nTron. The AND
gate and OR gate are topologically identical, and are only differen-
tiated by their bias conditions. AND/OR/COPY were constructed
purely from nTrons, while the NOT gate required a shunt impedance
for the bias (in this case a resistor). (B) AND-gate timing diagram for
pipelined logic propagation. Once gates A and B have valid inputs,
the bias current is enabled and the resulting output can be used as an
input for the next stage. 𝜏𝑏𝑖𝑎𝑠 denotes the propagation delay due to
the low-rate bias electronics. . . . . . . . . . . . . . . . . . . . . . . . 77

14
5-5 Experimental demonstration of an nTron half-adder. (A) Half-adder
circuit schematic constructed from logical gates. Single inputs were
provided into the initial (yellow) COPY gates, which acted as buffers
for the signals Input A and Input B, each with a fanout of three. Con-
nections to ground and between gates were made with low resistance,
non-superconducting links. (B) Per-channel output for the half-adder
for computation of 0+0, 0+1, 1+0, and 1+1, repeated twice. HIGH
(1) and LOW (0) current values were input to Input A and Input B,
and after a bias electronics delay 𝜏𝑏𝑖𝑎𝑠 , the lower bit and carry (upper)
bit outputs represented the resulting sum of the inputs. The red text
overlay of ones and zeros corresponds to HIGH and LOW values. . . . 86

5-6 The current comparator experiment used to test the input sensitivity
of the nTron. (A) Circuit diagram for the nTron current comparator.
The channel was biased at a fixed value, and the gate was ramped
until output appeared at the scope. (B) Histogram of I gate values for
the gate current at which the comparator switched and produced an
output voltage at the scope. . . . . . . . . . . . . . . . . . . . . . . . 87

5-7 Circuit schematic for 10 MHz eye diagram experiment. (A) Circuit
diagram for the nTron 10 MHz eye-diagram experiment. The area in
blue represents the portion of on the sample holder and submerged in
liquid helium at 4.2 K. Placing the resistors close to the device allowed
us to convert the incoming voltage square waves to a low-amplitude
current square waves. The resistors 𝑅𝐿 , 𝑅𝑏𝑖𝑎𝑠 , and 𝑅𝑔𝑎𝑡𝑒 were 1.46 kΩ,
20.8 kΩ, and 42.0 kΩ, respectively (as measured at 4.2 K). (B) 10 MHz
modified eye diagram output taken directly from oscilloscope . . . . . 88

15
5-8 Jitter measurements for an nTron integrated as an amplifier for a su-
perconducting nanowire single-photon detector (SNSPD) pulses. De-
tection of laser photons from a sub-ps laser by the detector (inset,
purple ‘S’ box) generated an electrical pulse on Port 1 (inset, red) and
also triggered a concurrent, amplified pulse from the nTron on Port
2 (inset, blue). Plotted is a histogram of the relative delay between
the laser sync edge and the resulting electrical pulse edges of the un-
amplified SNSPD (red dots) and nTron-amplified output (blue dots).
Gaussian fits to each data set are shown as solid lines. The reduced jit-
ter in the amplified signal is due to increased signal amplitude. (upper
right) Device schematic of the integrated SNSPD-nTron pulse amplifier. 89

5-9 Circuit schematic for the SNSPD and nTron pulse amplifier experi-
ment. (A), Device circuit schematic. The inductors were made by
patterning long nanowires, which intrinsically produce kinetic induc-
tance. The length of the inductor nanowires (and thus their total in-
ductance) were scaled against the SNSPD, which had an approximate
kinetic inductance of Lk ≈ 25 nH. (B) Room-temperature readout and
bias electronics. Pulses generated from the device and output to the
coax in (a) arrived at the other end of the coax, shown in (b), where
they were amplified with three 20-3000 MHz amplifiers in series before
being input to the scope. . . . . . . . . . . . . . . . . . . . . . . . . 90

5-10 Microscope image of the YBCO chip created for us by Lombardi group
at Chalmers University. The material is 50 nm of YBCO capped with
50 nm of gold on an MgO substrate. . . . . . . . . . . . . . . . . . . . 90

5-11 YBCO sample wirebonded to the custom PCB and mounted in the
vacuum cryostat. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

5-12 I-V curve shown for the YBCO nTron, with I gate = 0. . . . . . . . . . 91

5-13 I-V curves of the YBCO nTron versus as a function of varying I gate .
Shown are values of I gate in increments of 200 µA. . . . . . . . . . . . 92

16
6-1 Scanning electron micrograph of a yTron with a 200 nm gate and
100 nm channel. The low contrast of the edges that form the intersec-
tion are due to the tapering of the e-beam resist in that region. . . . 94

6-2 Fabrication steps for patterning the yTron out of a thin NbN film. (a)
NbN is deposited on an SiO2 substrate. (b) Titanium-gold contact
pads are added by a photolithographic liftoff process. (c) The e-beam
resist HSQ is spun on the sample. (d) The HSQ is patterned by an
e-beam tool and developed. (e) The sample is etched, leaving NbN
only in the areas protected by the HSQ and contact pads. . . . . . . . 95

6-3 Simulation of current flowing around a sharp corner. Current-streamlines

are shown, and the coloration indicates the current density, which is
at a maximum around sharp corner feature. . . . . . . . . . . . . . . 96

6-4 Current flow streamlines in the current-crowding cryotron for various

gate biases. (a) The gate is biased at the same current density as the
channel, and there is minimal current crowding at the intersection. (b)
The gate is at half the channel current density. (c) The gate carries no
current, and as a result the streamlines from the channel curve sharply
around the intersection, causing significant current crowding. . . . . . 97

6-5 Simulation of two yTron bias points showing the summation of horizon-
tal currents. (a) Current flowing in from the upper left arm and current
flowing from the upper right arm produce horizontal current compo-
nents which mostly cancel each other out, reducing current crowding
at the intersection point. (b) Current only flowing in from the upper
right arm. In this scenario there is no cancellation of horizontal current
components, and so there is a large amount of current crowding at the
intersection point. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

6-6 Channel switching current modulation versus gate current. . . . . . . 98

17
6-7 The two operating modes of the yTron, which match the operating
modes of a typical nanowire [1] based on whether or not the nanowire is
hysteretic. (a) The channel is shunted by a small resistance in parallel.
Flux flows across the channel, but the small resistance shunts the bias
current and prevents a stable Joule-heated hotspot from forming. (b)
The channel has a large shunt resistance. Significantly more power
is dissipated in the channel, allowing a self-sustaining Joule-heated
hotspot (normal region) to form. . . . . . . . . . . . . . . . . . . . . 102

6-8 IV curves of the yTron channel for different values of I gate . Each IV
curve looks approximately like a nanowire with a different Ic value. . 103

6-9 Readout procedure for the inline nondestructive measurement of the

superconducting loop. The gate-source loop started out at rest (left),
and then the Ic of the channel was measured by ramping 𝐼𝑟𝑒𝑎𝑑 until
a hotspot formed in the channel (right). 𝐼𝑟𝑒𝑎𝑑 was then turned off
and the system returned to rest (left). This process was able to be
repeated several thousand times without changing 𝑛, the number of
fluxons trapped in the gate-source loop. . . . . . . . . . . . . . . . . . 105

6-10 Procedure to change the number of fluxons 𝑛 in the superconducting

gate-source loop. (a) The entire device starts out unbiased, completely
superconducting. 𝑛 fluxons are stored in the gate-source loop. (b)
An applied electrical current from an external wire creates a hotspot
in part of the gate-source loop, breaking the superconductivity and
allowing flux to enter or leave the loop randomly. (c) The number of
fluxons in the loop has changed from 𝑛 to 𝑚. . . . . . . . . . . . . . . 106

18
6-11 Sequential trials of measurements of a quantized superconducting loop
using the yTron as an inline readout. Each dot corresponds to the
median value of 100 measurements of the Ic of the yTron channel.
The bars around each dot indicate the standard deviation of the Ic
measurements for that trial. Between each trial, the loop was heated
and cooled to allow fluxons to enter and leave. The step-like, evenly-
spaced division of Ic values indicate that the yTron was able to read
out the quantized current stored in the superconducting loop. . . . . 107

7-1 Circuit diagram of the Ic sweeping setup. DUT stands for “Device
Under Test” and refers to the nanowire being measured. The arbitrary
waveform generator (AWG) was a Agilent 33250A, the 2 MHz lowpass
filter was a Mini-Circuits BLP-1.9+, and the 80 MHz lowpass filter
was a high-rejection Mini-Circuits VLFX-80+. . . . . . . . . . . . . . 113

7-2 Oscilloscope voltage traces for an Ic sweeping measurement. Shown

in yellow is the sine-wave reference voltage produced by the AWG. In
pink is the voltage of the device under test (in this case an SNSPD). . 115

7-3 Ic sweep distribution measurement, shown in terms of the raw volt-

age (to convert to the Ic current divide by 10 kΩ). The Ic was swept
300,000 times to build this histogram. (inset) Zoomed portion of the
distribution showing the unexpected striations (periodic Gaussian-like
shapes) of the measured distribution. These striations were due to the
digital nature of the digital oscilloscope. . . . . . . . . . . . . . . . . 116

7-4 Ic sweep distribution measurement with the improved measurement

technique, eliminating the striations in the distribution (to convert to
the Ic current divide by 10 kΩ). The Ic was swept 58,000 times to build
this histogram. (inset) Zoomed portion of the distribution showing the
corrected distribution which does not have periodic behavior. . . . . . 117

7-5 Photograph of a GPIB cable connector and a GPIB-USB interface. . 119

19
7-6 Photograph of three different custom-built sample holders. The two
sample holders on the left were constructed from PCBs that were fab-
ricated in-house using the etching process described in this thesis. The
other sample holder (green PCB) had a PCB which was designed in-
house but purchased from a commercial PCB company. . . . . . . . . 122
7-7 Finished sample holder with cover on top. . . . . . . . . . . . . . . . 123
7-8 Vector drawing for an acrylic sample holder drawn in Inkscape. All
of the solid lines represent cuts performed by the laser cutter. The
smallest holes are holes meant for 4-40 tapping . . . . . . . . . . . . . 123
7-9 Photograph of gold-coated components which have been cut into pieces
and soldered to the PCB to act as a wirebonding targets. . . . . . . . 124
7-10 (left) Bare copper PCB board used as the blank substrate for PCB
patterning. (right) The same copper PCB board, covered with two
coats of black spraypaint. . . . . . . . . . . . . . . . . . . . . . . . . . 126
7-11 Laser cutter in the process of patterning the spraypaint on the surface
of the PCB board. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7-12 PCB design made in Inkscape. Areas which are black will be “printed”
by the laser cutter, exposing the bare copper and allowing those areas
to be etched. Areas in white will be copper in the finished PCB. . . . 127
7-13 The laser-patterned PCB, parts of which have been cleaned with a
Q-tip soaked in isopropyl alcohol. Before etching, the entire pattern
should be as shiny as the original bare copper was before spraypaint
application. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7-14 The final product from the in-house PCB fabrication process using the
laser cutter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

20
List of Tables

21
22
Chapter 1

Introduction to superconducting
devices

Superconducting devices are being investigated and applied to address critical needs
in the areas of computing [2], communications [3], and sensing [4]. They are vitally
important to diverse research and industrial fields such as magnetic-field sensing [5],
quantum and classical computing [6], photon sensing in communications [7], and as-
tronomy [8] [9]. These devices fall approximately into two classes of operation: (1) de-
vices which track and manipulate superconducting phase, and (2) devices which take
advantage of non-equilibrium states of the superconducting material. Devices in the
first class rely on the phase angle difference of the Ginzburg-Landau complex order
parameter between bulk superconducting electrodes. Devices in the second class gen-
erally manipulate the magnitude of the Ginzburg-Landau complex order parameter,
and so typically rely on weakened superconducting states which can be perturbed eas-
ily. The wide array of functionality provided by superconducting devices is sourced
from these two types of operation. However, there is a large unexplored territory for
new devices which can take advantage of both types of operation simultaneously.
Superconducting devices that are built out of thin films are intrinsically well-
suited to the manipulation of both the phase and magnitude of the order parameter,
making thin-films an ideal platform the exploration of new device functionality. In
films with thicknesses on the order of the superconducting coherence length, the

23
superconducting state is weak but stable. In these films, the magnitude of the order
parameter is tied to the phase due to the low current density Jc –the superconducting
state can be broken down by over-winding the superconducting phase (equivalent to
exceeding Jc ). Additionally, due to the presence of kinetic inductance in films of this
dimension, the phase is strongly tied to the flow of current in the device.

These relationships enable functionality to emerge from thin films just by pattern-
ing them into 2D shapes and passing current through them. Ultimately, the shape of
the pattern defines the spatial evolution of the phase. When current flows through a
patterned thin film the phase can take on complex patterns based on where the cur-
rent injection points are and the shape of the patterned film. For example, current
flowing around a notch in a thin-film nanowire will generate a large phase gradient
around the sharp features of the notch. If more current is added and the phase gra-
dient is increased, this superconducting state will actually break down at the notch,
allowing vortices to enter the wire as a means of relaxing the phase

The work done in this thesis describes nanoelectronic devices which take advantage
of the phase-magnitude relationship of thin superconducting films. By patterning 2D
geometries into thin superconducting films, these devices utilize these relationships
as well as the nanoscale effects caused by the controlled breakdown of the supercon-
ducting state [10] [11] [12] in order to produce useful functionality such as sensing
and amplification. The three devices described in this thesis are the result of the
exploration of this rich realm. In this introduction I describe the basic operation of
each class of devices, and give examples of devices from each category.

1.1 Phase- and magnitude-based devices

Here I describe a number of devices which either use superconducting phase, or non-
equilibrium dynamics of the superconducting order parameter magnitude for func-
tionality. This list is by no means comprehensive, but serves to provide an overview
of the various types of the devices in each of these categories.

24
1.1.1 Phase-based devices

The majority of devices which use the superconducting phase for operation are based
on the Josephson junction [13] [14] [15]. The Josephson junction is a two-terminal
device which is typically composed of two superconducting electrodes separated by
a thin insulating layer [16]. As long as this insulating layer is on the order of (or
smaller than) the superconducting coherence length 𝜉, the quantum states of the two
electrodes overlap enough to allow electrical current to tunnel through the insulating
layer without resistance. The magnitude of this tunneling current is described by the
Josephson relation 𝐼 = Ic sin 𝜑, where Ic is the critical current of the junction and 𝜑
is the phase difference between the superconducting electrodes.

The most commonly used device which incorporates the Josephson junction is the
superconducting quantum interference device (SQUID). A SQUID is formed by elec-
trically connecting two Josephson junctions in parallel with superconducting wires.
By connecting them with superconducting wires, the phase difference across the junc-
tions becomes related and as a result anything which perturbs this phase – such as
a magnetic field threading the loop formed by the SQUID – can be detected by the
resulting change in the combined junction tunneling currents. This phase sensitivity
gives SQUIDs their functionality as the world’s most sensitive magnetometers.

Manipulation of the superconducting phase can also be used to create digital

logic. One such technology is rapid-single-flux-quantum (RSFQ) logic, which tracks
the phase in a SQUID loop as a way to represent the digital values of 0 and 1 [17]
[18]. Since the phase in a SQUID can evolve extremely rapidly (changing by 𝜋 in
under a picosecond), this technology has demonstrated logical clock rates as high as
770 GHz [19].

Unfortunately, systems based on SQUIDs (including RSFQ electronics) suffer from

major disadvantages which render them impractical for a variety of applications and
environments. These disadvantages include low gain, high sensitivity to magnetic
fields, difficulty in driving large-impedance loads, and challenges in fabrication [20].
Devices based on the SQUID must be biased with substantial amounts of current, but

25
are limited in how much output current they can source. The result is a device with
low gain. In addition, the requirement that these devices include Josephson junctions
– ultra-thin tunneling barriers – renders them notoriously sensitive to fabrication
imperfections. A variation of an atomic layer in barrier thickness can radically change
the operating point of a device. Finally, SQUIDs are intrinsically the most sensitive
magnetic field sensors available. This feature is a blessing and a curse, as SQUID-
based computing devices must be heavily shielded in order to operate.
The Andreev interferometer is one of the few devices which uses superconducting
phase to produce functionality, but does not require Josephson junctions [21]. This
type of interferometer can measure long-range correlations across a non-superconducting
metal conductor [22]. These correlations mediate the apparent resistance of the nor-
mal metal conductor based on the superconducting phase difference between the two
ends of the conductor. The Andreev interferometer can be used to spectroscopically
probe the quantum state of superconducting magnetic-flux-based qubits [23].

1.1.2 Magnitude-based devices

There are a diverse set of of devices based on the various implementations of non-
equilibrium superconductivity, but for the most part these devices fall into two cate-
gories: sensors, and digital devices.

Sensors Superconducting sensors have been highly successful, and have found appli-
cations in a number of fields. These devices mostly rely on the shifts in superconduct-
ing equilibrium incurred by incident radiation. One example is the superconducting-
nanowire single-photon detector (SNSPD), which is used as an optical sensor for
low-power classical and quantum optical processes [24] [3]. Due to its narrow cross-
section, the SNSPD can detect single photons by means of a breakdown of supercon-
ductivity in the area where a photon lands. The fast response time of the SNSPD
and low timing-jitter make it a leading candidate for readout integration with quan-
tum photonic processors [25]. The SNSPD has found broad application in a diverse
set of fields such as biological sensing [26], circuit thermal analysis [27], and space

26
communications [28].
Another example of a successful superconducting sensor is the microwave kinetic-
inductance detector (MKID) [9] [4], which uses high-quality-factor superconducting
resonators to detect the perturbation of the superconducting state by incident pho-
tons. Since each MKID pixel is based on a high-Q superconducting resonator, thou-
sands of pixels can be read out on a single transmission line [29]. Due to its inherent
multiplexing capabilities, this device was rapidly developed as a useful astronomi-
cal tool. In the last 20 years MKID implementations have grown dramatically, and
several present-day telescopes use MKID-based cameras with over 1,000 pixels [30].

Digital devices The implementation of magnitude-based superconducting devices

for logic and other digital readout applications has been less successful than their
sensor counterparts [31]. These devices have the potential to use the advantages of
superconducting electronics (e.g., low noise, low dissipation, high speed, etc) while
avoiding the disadvantages of the Josephson junction (magnetic field sensitivity, low
gain, etc), but have thus far failed to find common application because they typically
require multi-layer structures for fabrication, have low input-output amplification,
and low-impedance outputs.
Superconducting digital devices have been pursued ever since the 1950s (well-prior
to the invention of the Josephson junction) when Dudley Buck invented the cryotron,
a four-terminal logic element which was significantly more compact than the then-
dominant vacuum tube technology [32]. However, these four-terminal devices were
abandoned after the development of the Josephson junction and the SQUID. In the in-
tervening years, SQUID-based technologies have dominated the literature, but a num-
ber of other devices based on non-equilibrium states have also been introduced. These
include tunable-supercurrent SNS junctions [33], resistive heaters stacked on top of
superconducting films [34], quasiparticle-injection links [35] [36] [37], and Josephson
FETs [38]. Despite their diversity, all of these devices have required two or more active
layers, and none have been demonstrated beyond the characterization of their basic
three- or four-terminal unit. Additionally, none of them except the cryotron has been

27
able to demonstrate a high-gain (>10), high-impedance output (≥50 Ω) which has
limited their abilities to be integrated with other non-superconducting technologies.
As a result, the practical implementation of any kind of superconducting logic device
which avoids the disadvantages of the Josephson junction has eluded researchers for
the last 40 years.

1.2 This thesis: Thin-film nanoelectronic devices

Advances in nanofabrication have enabled researchers to miniaturize devices smaller
with every passing year. Currently, electron-beam lithography has enabled routine
patterning of features as small as 10 nm [39] [40], which is less than nearly every
intrinsic length scale of the superconducting material NbN – a material which was
used extensively in the work done for this thesis. As device dimensions approach these
intrinsic length scales, the resulting changes in the superconducting behavior can
produce new behavior and correspondingly, new functionality. This thesis describes
new devices which have been developed through the exploration of this behavior.

Current-biased nanoSQUID The first device described in this thesis is a new

type of nanoSQUID which can be modulated by current biasing instead of magnetic
field biasing. A description of this current-biased nanoSQUID, its output character-
istics, and its development appear in Chapter 4. This device takes advantage of the
kinetic inductance as a means to asymmetrically bias an otherwise-symmetric super-
conducting loop. Since it does not require a large magnetic inductance to operate, the
method of biasing can apply to arbitrarily small nanoSQUIDs. This biasing method
reduces the need for large magnetic fields to be applied at a pickup loop, which may
be inadvertently coupled to the sample area.

nTron Presented next in this thesis is the nanocryotron (nTron), which is detailed
in Chapter 5. The nanocryotron is based on the cryotron, which used magnetic fields
induced by one superconducting wire to switch another nearby superconducting wire.
The cryotron was a promising superconducting digital logic element, but was never

28
successfully shrunk to nanoscale sizes because its magnetic field requirements did not
scale well to dimensions below several micrometers of size. The nanocryotron, how-
ever, is able to achieve similar operation to the cryotron at the nanoscale by changing
the mechanism of operation from magnetic field suppression to electrothermal sup-
pression. The key to the effectiveness of this electrothermal suppression is that the
device must be on the order of – or smaller than – the quasiparticle diffusion length,
which is about 100 nm in NbN.

yTron In Chapter 6, both electrothermal effects and kinetic inductance are utilized
to create a novel three-terminal device called the yTron. By utilizing kinetic induc-
tance to create current crowding, the yTron is able to act as an inline current sensor
for superconducting currents. Previously, sensing superconducting currents without
significantly perturbing them was only possible by weakly magnetically coupling to
them. Using a Y-shaped geometry, the yTron can be used to infer the magnitude of
current passing through one of the upper arms of the “Y” by measuring the critical
current of the other arm. Because of the excellent film-substrate thermal coupling,
this device can actually sustain a voltage state in one arm without disturbing the
current in the other arm. This feature has enabled the yTron to read out quan-
tized superconducting currents from a superconducting loop without perturbing the
number of quanta in the loop.

Experimental methods In Chapter 7, I discuss a number of experimental tech-

niques which enabled the rapid development of the devices described here. Among
these include a robust way to perform Ic measurements, along with a discussion why
it is important to choose a consistent Ic measurement method. Also included is a
primer on automating test measurements using Python and GPIB, and details re-
garding the fabrication of the sample holders that were used for submerging samples
in liquid helium.

29
30
Chapter 2

Nanoscale superconducting devices

While superconductivity has been a topic of research since 1911, only recently have
superconducting device dimensions reliably approached the nanoscale. In many cases,
shrinking these superconducting devices has the potential to yield improve metrics.
For example, in superconducting nanowire single photon detectors (SNSPDs), nar-
rower nanowires have enabled the detection of longer wavelengths with better effi-
ciency [41]. For superconducting magnetometers such as the nanoSQUID [42] [43],
reducing device sizes down to the nanoscale has enabled them to be sensitive to
the point of detecting a single electron spin [44]. For large-scale integrated super-
conducting electronics such as rapid-single-flux-quantum logic (RSFQ) [17], scaling
down the basic element of computation – the Josephson junction – has reduced power
consumption and increased their density of integration [45].

However, even as device sizes shrink, the fundamental length scales which govern
a superconducting material remain approximately constant. This scaling leads to a
variety of effects which can impact the operation of miniaturized devices. Although
these effects can adversely impact the operation of existing devices, these phenomena
can also be exploited to create new devices as well.

31
2.1 The effects of scaling down superconductors

Although superconductivity is best known as the transition of a material from being

electrically resistive to having zero resistance, there are several other effects which are
relevant to superconducting device construction. Additionally, there are additional
phenomena which begin to manifest as the length scales of the superconducting object
are reduced. Here we describe some of the basic principles of bulk superconductors,
and analyze how these principles are affected by reduced dimensions.

2.1.1 Characteristics of bulk superconductors

Superconductivity is characterized by the transition of a material from a diffusive

“sea” of electrons to the condensation of a macroscopic quantum state [46] [47]. This
quantum state is characterized by the superconducting energy gap Δ, which repre-
sents the energy difference between the quasiparticle sea and the superconducting
ground state. This energy can also be interpreted as a frequency through the relation
𝑓 = Δ/ℎ, where ℎ is Planck’s constant equal to 6.626 × 10−34 J s. The frequency 𝑓 ,
which is typically in the hundreds of gigahertz, is called the gap frequency and can
be thought of as the minimum frequency that incoming radiation would need to have
to perturb the superconducting state.
The onset of superconductivity occurs when certain materials drop below what
is called the transition temperature or critical temperature Tc . When the material
temperature drops below Tc , the electrons in the material become correlated and
binding together to form Cooper pairs. This attraction is actually present at all
temperatures in the material, but at temperatures above Tc the spectrum of phonon
noise in the system includes frequencies above 𝑓 (with non-negligible amplitudes),
meaning the electron-electron attraction is drowned out and correlations are destroyed
as quickly as they are formed.
These correlated electrons, or Cooper pairs, are able to move through the material
while avoiding several types of scattering that would otherwise affect a single elec-
tron. Specifically, the scattering sources which are avoided are those due which are

32
symmetric to time-reversal, e.g. phonons and non-magnetic impurities. This lack of
scattering means that current can be carried through the material without resistance.
Unlike normal electrons, the Cooper pairs do not have their momentum randomized
as they scatter off atoms in the material lattice, and so are able to carry current
without dissipation, as well as store energy in their motion.

Additionally, superconductors can be subjected to a certain magnitude of magnetic

field before the magnetic field is allowed to penetrate or destroy the superconducting
state. This magnitude is called the critical field, and its level is dependent on the
superconducting material (𝐻𝑐1 or 𝐻𝑐2 ). Associated with this magnetic field is a metric
called the critical current density Jc . Although a superconductor can carry current
without dissipation, the total amount of current it can carry before breaking down
is finite. As the result of this critical current, superconductors can exhibit strongly
nonlinear responses by being biased near Jc . This nonlinear response was made use
of extensively in the production of the devices described in this thesis.

As a result of the ability to carry current without resistance, superconductors act

similarly to perfect conductors and are able to repel applied magnetic fields. If, for
example, a magnetic field is applied to a perfectly conducting sphere with an infinite
charge carrier density, magnetic induction produces a surface current on the sphere
which generates an equal and opposite field inside the sphere. Thus, the superposition
of the applied field and the induced field cancel each other. The result is that the
sphere gains a surface current. This surface current does not dissipate, and so never
allows the magnetic field into the sphere’s core; the magnetic field inside the sphere
remains constant under all conditions.

Although its surface currents can be carried without dissipation, a superconduct-

ing sphere’s interior is not perfectly shielded from magnetic fields. Due to the finite
density and non-zero mass of the charge carriers, magnetic fields are imperfectly
screened on the surface of a superconductor. As a result, an applied magnetic field
penetrates into a superconductor’s surface to a finite depth – this depth is generally
known as the “penetration depth” 𝜆. In the superconducting clean limit (e.g. when
looking at bulk niobium), this depth is called the “London penetration depth” 𝜆L .

33
The penetration depth refers to the fact that in the surface of the superconductor,
the magnetic field falls off exponentially as 𝐵(𝑥) = 𝐵0 exp − 𝜆𝑥 , where 𝑥 is the depth
(︀ )︀

into the material. This thesis deals primarily with thin-film niobium and niobium ni-
tride and so we will deal only with the dirty-limit form of 𝜆. The equation describing
𝜆 is [46]

√︃
ℎ
𝜆= (2.1)
2𝜋 2 Δ𝜇 0 𝜎𝑛

where 𝜇0 is the magnetic permeability of vacuum, and 𝜎𝑛 is the normal-state

conductivity of the material (as measured just above Tc ).

2.1.2 Superconducting thin films and kinetic inductance

If the superconducting sphere is scaled down so that its radius is on the order of 𝜆,
however, the magnetic field will be able to penetrate the entire sphere to some degree.
As a result, current will flow throughout the entirety of the sphere, trying to repel as
much of the magnetic field as possible. However, due to its small dimension, there
are a limited number of electrons available for carrying the surface currents in the
superconductor. In these circumstances, the Cooper pairs act as an inertial energy
storage mechanism, since each Cooper pair (composed of two correlated electrons)
carries kinetic energy 𝐸𝑘 = 2( 12 𝑚𝑒 𝑣 2 ), where 𝑚𝑒 is the mass of the electron and 𝑣 is its
velocity. This kinetic inductance is not typically present in normal conductors (except
under very high frequency radiation), because this energy is continually dissipated due
to scattering.
This inertial energy storage is called kinetic inductance because it is an analogous
energy storage mechanism to the more familiar magnetic inductance. In both super-
conductors and normal conductors, energy is stored in the magnetic field surrounding
the current-carrying conductor. The amount of energy stored in the magnetic field
is dependent solely on the magnitude of current, the geometry of the conductor, and
the magnetic permeability of the materials surrounding it. In superconductors, the
energy stored in the kinetic motion of the Cooper pairs also depends on magnitude

34
of current, but since the energy storage is taking place inside the conductor – instead
of an external magnetic field – it does not depend on the surrounding geometry. The
total amount of energy carried by the Cooper pairs in a superconducting wire can be
equated to the inductive energy storage term Lk by

1 1
2( 𝑚𝑒 𝑣 2 )(𝑛𝑠 𝑙𝐴) = 𝐿𝐾 𝐼 2 (2.2)
2 2

where 𝑣 is the velocity of the Cooper pairs, 𝑛𝑠 is the Cooper pair density per unit
volume, 𝑙 is the length of the wire, 𝐴 is the cross-sectional area of the wire, and 𝐼 is
the total current carried by the wire.

For a fixed amount of current circulating in a superconducting object, the dispar-

ity between the energy stored in the magnetic field versus the energy stored in the
kinetic motion increases as the size of the superconducting object shrinks. Take for
example a very thin superconducting disc with a magnetic field applied perpendicu-
lar to the disc plane. If the thickness of the disc is such that 𝑡 << 𝜆, the magnetic
field inside the superconducting material will be almost completely uniform. The
circulating current induced by the applied magnetic field will be carried by all the
Cooper pairs in the entire volume of the device. If the disc is thinned by a factor of
two, there will be half the number of total charge-carrying Cooper pairs available to
carry current. So to carry a given amount of current, each of the Cooper pairs will
have approximately double the velocity. Along with this doubled velocity, the kinetic
energy storage per unit current the kinetic inductance will also approximately double
since Lk ∝ 𝐴. However, reducing the disc thickness does not significantly change the
magnetic inductance of the disc, and so as the length scales of the superconducting
object shrink below 𝜆, the kinetic inductance can overtake the magnetic inductance.

A useful approximation for kinetic inductance in a dirty thin film can be produced
using only the superconducting gap energy and the London penetration depth. This
approximation assumes that the superconducting material obeys the BCS relation
2Δ0 = 3.528kB Tc , which may not be true for extremely thin films or exotic mate-
rials, but is a reasonable estimate for the dirty-limit films used in this thesis. The

35
approximation is [4]

~Rs R
Lk = 𝜇0 𝜆2 /𝑡 = = 1.38 s pH/ (2.3)
𝜋Δ0 Tc
where Rs is the thin-film sheet resistance (measured just above Tc ), and Lk is
measured in terms of the sheet inductance of the thin film. The sheet inductance is
similar to the sheet resistivity, and is measured by counting the number of squares
in a current path. For instance, a 1 µm long wire that is 100 nm wide would be ten
squares long, regardless of thickness. To complete the example and calculate the sheet
inductance for a realistic film, if the film had a sheet resistance of Rs of 300 Ω/ and
a Tc of 12 K, the kinetic inductance per square of the film would be 34.5 pH/, and
the total inductance of the ten-square-long wire would be approximately 345 pH.

2.2 Applications of thin superconducting films

Patterning devices out of thin superconducting films is a useful approach for making
devices which have at least one nanometer-scale dimension. For instance, there are
several devices which can be produced from few-nanometer-thick superconducting
thin films, such as the superconducting nanowire single photon detector (SNSPD) [48].
The SNSPD is a nanowire typically about 4 nm thick and 100 nm wide, have lengths
on the order of 100 µm to 1000 µm, and are made from niobium nitride (NbN) or
tungsten silicide (WSi) [49] [50]. Due to their thin nature they have tens to hundreds
of nanohenries of kinetic inductance–typically each square (100 nm × 100 nm segment)
of the nanowire contributes 50 pH–100 pH of inductance. In operation, these devices
are current biased such that the wire current density is just below Jc , the critical
current density of the superconducting material. When an optical photon lands on
the nanowire, it deposits its energy onto the wire and perturbs the superconductivity
in a small area. Due to the extremely small dimensions of the nanowire, this <1 eV
perturbation is enough to suppress Jc at cross-section of the wire where the photon
lands. The result is that the superconductivity breaks down at the photon arrival
location, forming a resistive region known as a hotspot. This resistive region expels

36
the current from the nanowire, and diverts it out into a load impedance such as an
amplifier or transmission line.

Aside from its need for a small cross-section, the SNSPD derives much of its
functionality from its thin-film nature. First, by virtue of its large total kinetic
inductance, the current which is diverted into the external load takes some time to
recover back into the nanowire. This recovery time is critical to the free-running
operation of the SNSPD, as the expelled current must remain diverted from the
nanowire long enough for the nanowire resistive region to cool and heal back into
the superconducting state [51]. The second advantage of fabricating an SNSPD from
a thin film also ties into this cooldown time. Because the nanowire cross section
is wide and thin, the superconducting material making up the SNSPD has a large
surface area which is in intimate contact with the substrate [52]. For the SNSPD,
the substrate acts as a thermal tank which cools the resistive hotspot. A wide and
thin cross-section allows for excellent thermal coupling between the device and the
substrate, and speeds up this cooling process. This increased thermal coupling allows
the device to recover more quickly and operate at a higher count rate.

Another device which is typically fabricated from a superconducting thin film is

the kinetic inductance detector (KID) [9]. The MKID consists of an LC resonator
where the inductive component is provided by the kinetic inductance of a thin film.
Hundreds or thousands of these resonators can be capacitively coupled to a trans-
√
mission line, such that each resonator has a unique resonant frequency 𝑓𝑖 = 1/ 𝐿𝐶.
By measuring the spectrum of the transmission line, each resonator can be uniquely
identified by its frequency. When a photon lands on one of the resonators, some of
the Cooper pairs in that resonator are briefly broken apart. Since the total number
of available Cooper pairs is reduced in the resonator, the kinetic inductance of the
resonator increases and correspondingly, the resonant frequency of that resonator is
reduced. That frequency shift can then be read out by looking at the spectrum of
the transmission line. Thus, using a single transmission line several thousand KIDs
can be read out simultaneously.

Both the SNSPD and KID are typically created by patterning and etching a su-

37
perconducting thin film. This fabrication method has natural advantages for devices
whose functionality relies on thin film properties such as kinetic inductance or excel-
lent thermal cooling. The growth of thin films can be controlled to the sub-nanometer
level by sputter deposition, and can be deposited in thickness ranging from a few
nanometers to hundreds of nanometers [53]. Due to this range and precision, devices
which rely on kinetic inductance can have their thicknesses tuned precisely. By grow-
ing a thin film, the smallest dimension of the device can be characterized in advance,
and only 2D patterning is needed to complete the fabrication process.

38
Chapter 3

The 2D electrothermal model

To better understand the operation of a new device, it is important to build a sim-

ulation or analysis framework with which to test observations. In the case of the
SNSPDs, the development of a 1D electrothermal model [54] was crucial to the devel-
opment and understanding of the nanowire physics. However, this 1D model was not
applicable to devices that have a fundamentally 2D geometry such as those described
in this thesis. This chapter describes the development and application of the 2D
electrothermal model, which was used with great success to simulate the operation of
the nTron and yTron.

3.1 Electrothermal basics

The electrothermal model described here is based on the two-temperature model

described by Ref. [55]. In this work, the authors analyze a superconducting film in
terms of two effective temperatures: the electron temperature Te , and the phonon
temperature Tph . These temperatures represent the non-equilibrium distributions
of the electron and phonon systems, respectively. These temperatures are spatially
dependent, for example, the local value of Te corresponds approximately to the local
density of Cooper pairs and may vary spatially across the device. This formulation
lends itself well to simulating device operation, as it is inherently time-dependent,
couples the systems together in a straightforward fashion, applies to a wide range

39
of temperatures (e.g. not only 𝑇 ≪ Tc or 𝑇 ≈ Tc ), and can smoothly transition
between the superconducting and normal states.

To implement the two-temperature model then, the first step was to choose a
numerical solver capable of working with the partial differential equations (PDE)
of interest: an electrical PDE and two thermal PDEs. For this implementation, we
used the COMSOL numerical simulation software, which had a built-in heat equation
PDE, current flow PDE, and allows for multiple PDEs to be coupled easily. After
that, it was a matter of implementing the relevant equations and parameters to suit
the material of interest, in this case, thin-film niobium nitride (NbN).

3.1.1 The heat equation

The two-temperature model desribed in Ref. [55] used two coupled heat equation
PDEs to describe the effect electron and phonon temperatures of the superconducting
material. The heat equation PDE fundamentally describes the inflow and outflow of
thermal energy versus time for each point in space. Its most basic form is

𝜕(𝑐𝑇 )
= ∇2 (𝜅𝑇 ) (3.1)
𝜕𝑡

where 𝑇 is the temperature, 𝑐 is the specific heat of the material, 𝜅 is the in-plane
thermal conductivity of the material. In the implementation described here, all of
these variables are functions of both space and time (e.g. 𝜅 is 𝜅(⃗𝑟, 𝑡)). For every
point in space, what Eq. 3.1 describes is the conservation of thermal energy during
a diffusive process, neglecting any external couplings. For any point in space, the
rate of thermal energy loss (the left term of Eq. 3.1) is equal to the curvature of the
thermal energy (the right term of Eq. 3.1). This characterizes a diffusive thermal
process: as time passes the system smooths any non-uniformity in 𝑇 such that the
whole geometry tends to a uniform temperature.

40
3.1.2 Modifying the basic heat equation

The basic form of the heat equation shown in Eq. 3.1 neglects any outside sources of
thermal heating or cooling, and so only describes the passive diffusion of heat from a
given starting condition. In order to fully describe the operation of a current-biased
thin film on a substrate, this equation must add three additional terms: one term
to represent electrical Joule heating from current flow across resistive regions, one
term to represent the thermal coupling of the film to the substrate, and one term to
describe the coupling between the electron and phonon systems. Since the substrate
subtracts energy only from the phonon system, and the Joule heating adds energy to
only the electron system, it is now prudent to explicitly write out the two coupled
heat equation PDEs and the electrical PDE.

Electron temperature PDE

Eq. 3.2 describes the electron system effective temperature, and is represented by the
basic heat equation with an additional term for Joule heating, and another term to
represent the electron-phonon coupling.

𝜕(ce Te ) c
= ∇2 (𝜅e Te ) + |J|2 𝜌 − e (Te − Tph ) (3.2)
𝜕𝑡 𝜏e-ph
In Eq. 3.2, ce is the electron specific heat, 𝜅e is the electron thermal conductivity,
|J|2 is the norm of the current density, 𝜌 is the resistivity, and 𝜏e-ph is the electron-
phonon interaction time. Since experimental measurements often measure De and ce
instead of 𝜅e directly, it should be noted that 𝜅e and ce are related by the standard
thermal diffusivity relation De = 𝜅e /ce .

Phonon temperature PDE

To describe the phonon system, we use the basic heat equation with an additional
term for cooling into substrate, as well as another term to represent the electron-
phonon coupling. The term for electron-phonon coupling is the negative of the term
in Eq. 3.2) so that the sum of the two coupling terms is zero, because the coupling

41
only describes energy transfer and not net gain or loss. The equation for the phonon
system then becomes

𝜕(cph Tph ) cph c

= ∇2 (𝜅ph Tph ) − (Tph − Tsub ) + e (Te − Tph ) (3.3)
𝜕𝑡 𝜏esc 𝜏e-ph

where cph was the phonon specific heat, 𝜅e was the phonon thermal conductivity,
Tsub is the temperature of the substrate, and 𝜏esc was the escape time constant which
determined the rate of cooling into the substrate. Note that the term with the escape
time constant has several different forms–for instance, it can instead be replaced by
the Kapitza resistance which has the form ((Tph )𝜂 − (Tsub )𝜂 ) where 𝜂 is typically 4,
but can range between 3 and 6 depending on the substrate-film interface [56]. The
precise form was never adequately measured in our NbN films, and so the form given
by Ref. [24] above was used. Literature on transition-edge sensors has a wealth of
information on this topic for further investigation [56]. As described before, all of the
variables in Eq. 3.2 and Eq. 3.3 are spatially dependent and time dependent, with
the exception of the parameters Tsub which is a constant.

Figure 3-1: Diagram showing the production and flow of energy between the different
coupled systems in the electrothermal model.

42
Electrical currents PDE

The electron temperature heat equation now contains a term corresponding to resis-
tive heating caused by current flow. This means that there must be a third PDE
coupled to the two heat equation PDEs, which couples to the electron system heat
equation by the Joule heating term |J|2 𝜌. This PDE is based on Ohm’s law:

1
J= E (3.4)
𝜌

3.2 Adapting the 1D model to 2D

Much of the groundwork for this implementation was laid out in Ref. [54], which out-
lines an electrothermal model in one dimension. While it would have been convenient
to simply impose the correct boundary conditions to adapt the 1D model to 2D, it
was not that simple. The greatest difference between the 1D implementation and the
one outlined here was the 1D model’s discrete separation of the superconducting and
normal states. In the 1D model, when 𝑇 > Tc or 𝐽 > Jc for any segment of the wire,
the internal logic of the simulation instantaneously changed that segment of the wire
from being completely superconducting (zero resistance) to completely normal.
The discrete division of the superconducting state from the resistive state was
not possible in a 2D topology, due to the effect of current crowding [57]. Consider
the example of a current-biased superconducting nanowire which has a photon land
on it. The simulation begins with a nanowire which has a constant current flowing
through it such that 𝐽 is well below Jc . If a photon lands and a resistive island is
formed in the center of the nanowire, physically one would expect the resistive spot
to dissipate since there is not enough current to support the creation of a full hotspot.
However, in this numerical implementation, in the time step following the creation
of the resistive island, the solver will compute the current flow around that resistive
island. In this time step, current crowding guarantees the solver will see a greatly
increased current density at the boundary between the normal and superconducting
sections–especially if the mesh is coarse. Then, because 𝐽 > Jc for the mesh elements

43
near the boundary of the island, the solver will set those elements to the normal state,
increasing the size of the island. This process will continue until every mesh element
in the entire wire is resistive. Additionally, this erroneous process occurs no matter
how small the timesteps are because the boolean check of 𝐽 > Jc . For instance, if
the solver steps with 1 fs increments, the entire wire can become resistive in under a
picosecond, which represents a non-physical response.

The non-physical expansion of resistive regions was amended in the 2D model

by more fully implementing the two-temperature model and eliminating the discrete
division of the superconducting and resistive states. In the two-temperature model,
the resistivity becomes a function dependent only on Te . Since the Te heat equation
is fundamentally diffusive, this helps ensure a spatially-smooth solution for 𝜌. Instead
of adjacent mesh elements having different states and causing current-crowding, the
value of Te and thus 𝜌 instead varies smoothly over several mesh elements. This adap-
tation also requires a modification of the Joule heating term of the two-temperature
model. In the 1D model, the discrete superconducting and resistive states make cal-
culation of the Joule heating simple: superconducting sections do not produce Joule
heating, and resistive sections produce Joule heating according to 𝐼 2 Rn , where 𝐼 is the
current flowing through the 1D wire and Rn is the normal resistivity of that segment.
In the 2D model this distinction is not quite so clear since the superconducting and
normal state are not as well-defined. This 2D model implements the Joule heating
by adding the equation for Jc (Te ), so that everywhere Te < Tc has a nonzero Jc .
We then assumed that the Joule heating for any area where Te < Tc was equal to
min (|J| − 𝐽𝑐(𝑇 𝑒))2 , 0 . This form had a phenomenological purpose: it prevented
(︀ )︀

the model from producing heating in areas where the current density was smaller
than Jc , even if the resistivity was nonzero. It is unclear whether this formulation
was microscopically accurate, but subtracting the Jc (Te ) term from the current den-
sity norm reduced the time to numerical convergence, and did not appear to affect
the operational results when the simulation was used on actual device geometries.
from

44
3.3 Variables, equations, and parameters

This section lists all the physical variables and parameters used in constructing the
2D electrothermal model, the references in which they appeared, and notes specific
to their implementation in COMSOL. Most of the values compiled here were specific
to thin-film NbN .

3.3.1 Electrothermal variables

The superconducting gap energy Δ

The equation for the superconducting gap Δ can typically only be solved by numerical
integration [46]. However, for standard BCS superconductors there is a form that is
accurate to within a few percent, which can be found in Ref. [58]. This form is

(︃ √︃ (︂ )︂)︃
𝜋 2 Tc
Δ(Te ) = 1.76kB Tc tanh (1.43) −1 (3.5)
1.76 3 Te

where kB is Boltzmann’s constant, and Tc is the thin-film superconducting critical

temperature which was measured to be 12.6 K. A plot using these parameters is
shown in Fig. 3-2.

Figure 3-2: A graph of Eq. 3.5, plotting the superconducting bandgap Δ versus
temperature.

45
The electron specific heat ce

The equation for the electron specific heat ce is state dependent [54]. In the super-
conducting state (Te < Tc ), it decays exponentially with temperature. In the normal
state (Te ≥ Tc ), it increases linearly like most metals. Its form is
⎧
⎨𝐴 exp (−Δ(Te )/(kB * Te )) 0 < Te < Tc
⎪
ce (Te ) = (3.6)
⎩c T Te ≥ Tc
⎪
e0 e

where ce0 was 240 J/(m3 K2 ) for thin-film NbN [24], and 𝐴 was the proportional-
ity constant 2.43ce0 Tc . The electron specific heat had a discontinuity at Te = Tc ,
corresponding to the specific heat gained by the onset of superconductivity. This
discontinuity caused problems in the simulation, because when the solver made small
changes to Te around Tc , the specific heat changed drastically. To remedy this issue,
the COMSOL function was smoothed by enforcing a continuous second derivative,
sacrificing some accuracy for the sake of ensuring the solution converged successfully.
A plot using these parameters is shown in Fig. 3-3.

Figure 3-3: A graph of Eq. 3.6, showing the electron specific heat ce versus electron
temperature Te .

The electron thermal conductivity 𝜅e

The thermal conductivity of the electron system was also state-dependent. Its equa-
tion for Te > Tc was well defined, but in the superconducting state there are several
formulae and representations that can be used. One solution, used in Ref. [59] and

46
subsequently in Ref. [54], was to linearly interpolate 𝜅e (Te ) between zero and its value
at Tc . This is expressed by the piecewise equation
⎧ (︁ )︁
⎨ 𝐿Te Te
⎪
0 < Te < Tc
𝜌 T
𝜅e (Te ) = 20K c
(3.7)
⎩ 𝐿Te
⎪
Te ≥ Tc
𝜌20K

where L is the Lorenz number 𝐿 = 2.44 × 10−8 WΩ/K2 , and 𝜌20K was the nor-
mal resistivity of the NbN film (measured just above Tc ) which had a value of
2.85 × 10−6 Ω m. A plot using these parameters is shown in Fig. 3-4.

Figure 3-4: A graph of Eq. 3.7, showing the electron thermal conductivity 𝜅e versus
temperature.

The resistivity 𝜌

For simulation purposes, it was convenient to not distinguish 𝜌 expressly between the
superconducting and normal states, and instead express the onset of superconductiv-
ity as a smooth reduction of 𝜌(Te ) as Te approached zero. This was accomplished
using modified version of the form specified in Ref. [24]:

(︂ (︂ )︂)︂−1 (︂ )︂
Te − Tc Te
𝜌(Te ) = 𝜌20K 1 + exp −4 + 𝜌sc 1 + (3.8)
𝑑Tc Tc
where 𝜌sc was an artificial constant set to 1 × 10−8 Ω m, and 𝑑Tc was the width of
the resistive-to-superconducting transition, set to 1.1 K. This modified form includes
(︁ )︁
the 𝜌sc 1 + T T
e
term as a way to assist the solver in converging to a solution. Without
c

this term, near Te = 0 the resistivity becomes exponentially close to zero. This

47
approach to zero causes problems for the solver because the values in the electrical
PDE has a 1/𝜌 term which diverges rapidly. By adding the extra term, 𝜌 was given
a minimum value of 𝜌sc which limited the size of the 1/𝜌 component of the electrical
(︁ )︁
PDE solution. The 1 + T part of the extra term additionally enforced a minimum
Te
c

value to the first derivative of 𝜌, which also helped with convergence. A plot showing
both 𝜌 and 𝜎 = 1/𝜌 is shown in Fig. 3-5 to help illustrate this point.

Figure 3-5: (above) A graph of Eq. 3.8, showing the resistivity 𝜌 versus tempera-
ture. (below) A graph of the inverse of Eq. 3.8, showing the conductivity 1/𝜌 versus
temperature.

The critical current density Jc

The form of the critical current density is taken directly from Ref. [60]. The equation
is

(︃ (︂ )︂2 )︃3/2 (︃ (︂ )︂2 )︃1/2

Te Te
Jc (Te ) = Jc0 1− 1+ (3.9)
Tc Tc

48
 where Jc0 was the measured critical current density of the NbN film. This value
was 6.6 × 1010 A/m2 for thin-film NbN, and was determined by measuring the critical
current of a relatively wide (1 µm) wire, and dividing it by the cross-sectional area of
the wire. A plot using these parameters is shown in Fig. 3-6.

Figure 3-6: A graph of Eq. 3.9, showing the critical current density Jc versus temper-
ature.

The electron-phonon interaction time 𝜏e-ph

The electron-phonon interaction time described the rate at which the phonon and
electron effective temperatures reached equilibrium with each other. The form used
in this implementation was taken from Ref. [24], and was

(︂ )︂−1.6
Te
𝜏e-ph (Te ) = 𝜏e-ph0 (3.10)
Tc

where 𝜏e-ph0 was 17 ps. A plot using these parameters is shown in Fig. 3-7.

The phonon specific heat cph

Only one of the variables in the electrothermal model depended on on the phonon
temperature, and that variable was the phonon specific heat. The phonon specific
heat was state independent, and scaled like the cube of the temperature Tph :

cph (Tph ) = cph0 Tph 3 (3.11)

49
Figure 3-7: A graph of Eq. 3.10, showing the electron-phonon interaction time 𝜏e-ph
versus temperature.

The value cph0 was taken from Ref. [24] and was set to 9.8 J/m3 K4 . A plot using
these parameters is shown in Fig. 3-8.

Figure 3-8: A graph of Eq. 3.11, showing the phonon specific heat cph versus phonon
temperature Tph .

3.3.2 Boundary conditions and initial values

The boundary conditions used in the 2D electrothermal modal were similar to those
used in Ref. [54]. Electrical terminals were placed where desired on the 2D geometry,
and those boundaries had their temperatures fixed to Te = Tph = Tsub . This bound-
ary condition represented the wire extending off to infinity, where it was in equilibrium
with the substrate temperature. All other boundaries were set to be both thermally
insulating as well as electrically insulating, ensuring no heat or current passed through
them.

50
 In many situations, the NbN nanowire structures being simulated were hysteretic.
Due to this hysteresis, it was required to provide the solver with sensible initial values,
since at a given bias point there could be multiple solutions and the solver would not
know which ones to choose. The most robust method that was found for setting initial
values was to begin with bias conditions where the entire device was superconducting.
The initial values could then be set such that everywhere Te = Tph = Tsub for the
heat PDEs, and 𝑉 = 0 everywhere for the electrical PDE. Once these initial values
were set, the time dependent simulation could begin, during which the bias conditions
could be changed to create resistive regions or simulation other types of operation.

3.4 Conclusions and future work

This chapter described the implementation of a 2D electrothermal model which was

based off the two temperature model described by Ref. [55] and 1D implementa-
tion done in Ref. [54]. The model was critical to predicting and understanding the
operation of the nTron, which is described in Chapter 3.
There are still areas of the model which could benefit from further work. The
first is the inclusion of an inertial term to the electrical PDE which would model the
effects of kinetic inductance. In the implementation described here, current flow is
determined independently in each time step, and so may change freely from one time
step to the next. In the actual physical thin film, however, kinetic inductance plays a
role in ensuring current flow cannot change its magnitude or direction instantaneously.
Adding a kinetic inductance term would better model the physical system, as well
as likely help the implementation converge faster, since the electrical PDE would not
be able to change drastically between adjacent time steps. Currently the COMSOL
implementation described here was not extremely efficient, as the simulation of about
100 ps of time required about an hour on a Intel Core i7 3 GHz processor.
Another improvement that could be implemented in the 2D model would be to add
the physics from a microscopic quantum theory, such as the time-dependent Ginzburg-
Landau [61] equation or Usadel [62] equations. The two-temperature model described

51
here currently lacks an important component of current-carrying superconducting
thin films: the motion of vortices. Vortices have been suggested as a mechanism
for detection events in SNSPDs [63], and play an important role in other nanoscale
superconducting devices [64].

52
Chapter 4

The current-biased nanoSQUID

The nanoSQUID is a nanoscale superconducting quantum interference device (SQUID)

in which the weak-link elements are Dayem bridges [65] instead of Josephson junc-
tions (JJs) [66]. These devices have have been fabricated from a number of materials,
including aluminum [43], niobium [67] [68] [69], and lead [44], and have demonstrated
√
flux sensitivities of 50 nΦ0 / Hz [44], sufficient to resolve a single electron spin. The
use of Dayem bridges instead of multilayer JJs allows the nanoSQUID to be patterned
from a single-layer thin film, to reach diameters below 100 nm, and to be realized in
high-Tc superconductors [70] [71]. At these scales, the geometric inductance of the
nanoSQUID superconducting ring is small, on the order of 100 fH for a ring of size
100 nm. However, the kinetic inductance of the thin film can be significant even at
these sizes, ranging from comparable to the geometric inductance to several orders of
magnitude larger.
To date, nanoSQUIDs have been probed using an applied magnetic field; but in
many applications, the presence of this field will perturb the sample being studied.
Like a conventional SQUID, tuning the nanoSQUID to its most sensitive region typ-
ically involves coupling an external magnetic bias field to the device. In the case of
scanning SQUID microscopy (SSM) the bias field must be capable of coupling several
Φ0 of flux in the nanoSQUID loop to provide closed-loop feedback. This amount of
flux corresponds to a large local field that must be applied to the sensing area of
the nanoSQUID, potentially perturbing the nearby sample. Here we report a new

53
method of modulating a nanoSQUID by coupling to the device kinetic inductance
instead of its geometric inductance. We have been able to demonstrate nanoSQUID
modulation without the application of any external field, using kinetic inductance
coupling of currents injected asymmetrically into the nanoSQUID. More generally,
this result demonstrates current-based modulation of a fluxoid inside a supercon-
ducting mesoscopic loop, which was previously performed by applying magnetic flux
to the loop [72] [66] [73] [74].
The text that follows constitutes a preliminary writeup of work which will be
condensed and submitted to a journal for publication.

4.1 Device characteristics

The nanoSQUID device geometry is shown in Fig. 4-1 and is composed of a supercon-
ducting ring, with four connecting terminals and two constrictions, all patterned on
a 10-nm-thick niobium film. The terminals at the top and bottom of the loop were
used to measure the switching current of the nanoSQUID, while the two terminals
coming from the right of Fig. 4-1 were used to inject modulation current. The fabri-
cated constrictions were 105 nm wide, several times larger than the coherence length
of thin-film niobium, but significantly smaller than the thin-film penetration depth
𝜆thin = 𝜆2 /𝑑 where 𝑑 is the thickness of the film and 𝜆 is the London penetration
depth [75].
The device was fabricated from ∼10 nm niobium deposited on sapphire by DC
magnetron sputtering. The film had a Tc of 8.2 K, a room-temperature sheet resis-
tance of 30.4 Ω/, and a residual resistance ratio (RRR) of 3.3. Contact pads were
created by evaporating titanium and gold onto the surface using a liftoff process.
The nanoSQUID geometry was then patterned by electron-beam lithography, using
∼50 nm HSQ as a resist. The pattern was transferred into the film by reactive-ion
etching at 0.15 W/cm2 for 3 min in 1.3 Pa (10 mTorr) CF4 .
These constrictions shown in Fig. 4-1 are a type of superconducting weak link,
the Dayem bridge [65]. Depending on its dimensions, a Dayem bridge may have a

54
Figure 4-1: (left) Scanning-electron micrograph of a current-controlled nanoSQUID
device, fabricated from a thin niobium film. Inset shows a closeup of one of the
nanoSQUID constrictions, which were measured to be 105 nm wide at their narrowest
point. (right) Equivalent circuit of the nanoSQUID device. Shown are the four
terminals of the device and their inputs. I bias , which is used to measure the switching
current of the device, flows in from terminal 1 at the top and is carried out through
terminal 4 at the bottom. The modulation current Imod enters and leaves through
the terminals 2 and 3 on the right. Isym and Iloop are the symmetric and circulating
components of Imod , respectively.

significantly different current-phase relationship (CPR) than the typical Josephson

relation 𝐼 = Ic sin(𝜑). Our constrictions are wider than the coherence length, and
so likely have a different CPR [76]. However, as long as the bridge cross section is
significantly smaller than the thin-film magnetic penetration depth (𝜆thin is ∼60 µm
for our 10-nm-thick Nb), the CPR of the bridge is expected to be 2𝜋-periodic and
allow phase slippage [77]. The exact nature of the CPR determines the method of
phase slippage. In wider bridges like the ones used here, phase slippage occurs by the
passage of vortices across the wire. However, the modulation technique reported here
is not dependent on the form of the bridge CPR, and so should extend to narrower,
more Josephson-like bridges as well.

55
4.2 Analysis of the nanoSQUID switching current

With some basic analysis, it was possible to plot what the expected the nanoSQUID
switching current versus the gate current should look like. This was a useful way to
check the results of the experiment, as well as do some basic analysis to extract values
of interest from the measured device.

4.2.1 Analysis of the nanoSQUID switching current

The analysis began by quantifying the currents traveling through the left and right
constrictions. These two equations were

𝑖𝐿 = I bias /2 + I gate 𝑟 − 𝑛𝐼0 (4.1)

𝑖𝑅 = I bias /2 + I gate (1 − 𝑟) + 𝑛𝐼0 (4.2)

where 𝑖𝐿 is the current through the left constriction, 𝑖𝑅 is the current through the
right constriction, I bias was the applied bias current, I gate is the applied gate current,
𝑅
𝐼𝑠𝑤 is the current in the right constriction, 𝐼𝑠𝑤
𝐿
is the switching current in the left
constriction, 𝐼0 is the current induced in the loop by a single fluxoid, and 𝑟 is the
splitting ratio of the gate current between the left and right constrictions (0 < 𝑟 < 1).
For simplicity, all currents are normalized such that 𝐼𝑠𝑤
𝐿
= 1.
The next step was to define the conditions that must be met in order for a given
combination of I bias and I gate inputs to be superconducting. The condition imposed
by this requirement is simply that 𝑖𝐿 and 𝑖𝑅 must be less than their respective critical
currents. This resulted in the following inequalities:

|𝑖𝐿 | =| I bias /2 + I gate 𝑟 − 𝑛𝐼0 |< 1 (4.3)

|𝑖𝑅 | =| I bias /2 + I gate (1 − 𝑟) + 𝑛𝐼0 |< 𝐼𝑠𝑤

𝑅
(4.4)

56
 Rearranging these to eliminate the absolute value terms yielded a set of four
inequalities that indicated, for a given I gate , the range values of I bias which were valid
superconducting states:

I bias <2(1 − I gate 𝑟 + 𝑛𝐼0 ) (4.5)

I bias <2(𝐼𝑠𝑤
𝑅
− I gate (1 − 𝑟) + 𝑛𝐼0 ) (4.6)

I bias >2(−1 − I gate 𝑟 + 𝑛𝐼0 ) (4.7)

I bias >2(−𝐼𝑠𝑤
𝑅
− I gate (1 − 𝑟) + 𝑛𝐼0 ) (4.8)

4.2.2 Visualizing the analysis

In the 2D plane of I gate versus I bias , each of the inequalities in Eq. 4.5 forms a boundary
for the valid superconducting region as shown on the left in Fig. 4-2. By solving for
all four of them, an area of this 2D plane can be solved for, as shown on the right in
Fig. 4-2.

Figure 4-2: Graph plotting the solution to the nanoSQUID inequalities for 𝑟 = 0.5, 𝐼0
= 0.5, 𝐼𝑠𝑤
𝑅
= 1, and 𝑛 = 0. (left) Graph of the boundaries generated by the inequalities
in Eq. 4.5. (right) Graph of the area which solves all four inequalities in Eq. 4.5

The graphs shown in Fig. 4-2 represent the solutions for all values of I gate and
I bias , but only for the fluxoid state 𝑛 = 0. In the actual device, the fluxoid state was
allowed to change through phase slips, so therefore the inequalities had to be solved
for any integer value of 𝑛, which is shown in Fig. 4-3.

57
Figure 4-3: Graph plotting the solution to the nanoSQUID inequalities for 𝑟 = 0.5,
𝐼0 = 0.5, 𝐼𝑠𝑤
𝑅
= 1 (left) Graph of the valid regions for 𝑛 = 0 and 𝑛 = 1. (right) Graph
of the valid regions for all integer values of 𝑛

At this point, the solutions still do like the standard SQUID diagram, where ap-
plied magnetic flux creates a sawtooth modulation in the critical current. This was
because the measurement that actually takes place in the experiment is a measure-
ment of Isw , which is equal to I gate + I bias . By instead plotting the total current
I gate + I bias versus I gate as shown in Fig. 4-4, the plots better represented what the
experimental results measured.

Figure 4-4: Results of the analysis of the nanoSQUID for 𝑟 = 0.5, 𝐼0 = 0.5, 𝐼𝑠𝑤 𝑅
=1
and all integer values of 𝑛, plotted with the total I gate
+I bias
which better corresponds
to the experimentally measured switching current.

By tuning the parameters and replotting the results, such as those in Fig. 4-5,
this analysis enabled us to quickly develop a qualitative understanding of what each
of the parameters did to the results. It additionally assisted us with the final analysis
of the physical device.

58
Figure 4-5: Results of the analysis of the nanoSQUID based on different parameter
inputs. (left) Graph generated when the splitting ratio is 𝑟 = 0.8. (middle) Graph
generated when the flux induced current 𝐼0 was set to = 0.2. (right) Graph generated
when the constriction critical currents were asymmetric, such that 𝐼𝑠𝑤
𝑅
= 3.

4.3 Measurements and results

To measure the nanoSQUID, we injected a fixed modulation current Imod into the
device, as shown in the circuit schematic of Fig. 4-1. We then measured the switching
current Isw of the device using current applied through the bias terminals. Specifically,
the Isw discussed here represents the total amount of current passing through the
constrictions just before the constrictions switched to the normal state.
The nanoSQUID Isw distribution measurements took place with the sample sub-
merged in a bath of liquid helium. The sample was placed in a copper-shielded sample
holder, but no magnetic shielding was used. The modulation current Imod was sup-
plied using an SRS variable battery source with two 20 kΩ resistors, one in series with
each terminal of the battery source. With Imod fixed, the distribution of Isw was then
measured by ramping I bias until a nonzero voltage appeared at the I bias terminal,
indicating that the constrictions switched to the normal state. The current ramp for
I bias was provided by an arbitrary waveform generator (AWG) in series with a 10 kΩ
resistor. The AWG output a 5 Vpp, 200 Hz triangle wave, corresponding to a current
ramp rate of 0.3 A/s. Extended details about this type of measurement can be found
in Section 7.1.1.
As we varied the injected current Imod , we observed the modulation of the device
Isw shown in Fig. 4-6. As the device is modulated by Imod , the switching current

59
Figure 4-6: Experimental results of the nanoSQUID being modulated by injected
current. Shown is the distribution of the nanoSQUID switching current (Isw ) varying
as a function of the injected modulation current (Imod ). Each vertical slice of the
graph corresponds to a a measurement of the Isw distribution for that value of Imod .
(inset) Two slices showing the distribution of Isw when maximally and minimally
modulated by Imod .

exhibits a triangle-wave pattern similar to that seen in Ref. [73], indicating a multi-
valued, approximately-linear current-phase relationship. Since the nanoSQUID forms
an unbroken superconducting loop, the shape of the Isw modulation can be under-
stood as follows. To maintain phase single-valuedness, current injected by Imod splits
between the two paths around the loop according to each path’s relative inductance.
One of the paths has a smaller inductance, and so carries a larger fraction of Imod .
The resulting imbalance of current flowing through the two constrictions reduces the
total Isw of the device. To make this analysis clearer, we can break up the contri-
butions of Imod into two constituent currents: Isym , the portion of the modulation
current which is divided equally between the two constrictions, and Iloop , a circulat-
ing current which has equal and opposite values through each constriction. These
components are shown in Fig. 4-1. Since our measurement of interest, Isw , is defined
as total amount of current passing through both constrictions when they switch, the
measurement of Isym is automatically absorbed into Isw , leaving only Iloop to affect the
value of Isw . Thus, we can view the effect of Imod as solely producing a loop current,

60
similar to how a magnetic field would induce a loop current in a conventional SQUID.
The periodicity of the Isw modulation arrives from the London quantization condi-
tion, which enforces an integer number of fluxoids in the loops. When Imod produces
enough circulating current, the device can counteract the induced current by allow-
ing a fluxoid in through one of the constrictions. Thus, the adjacent maxima of the
triangle wave shape correspond to Imod inducing a circulating current equivalent to
one fluxoid. One feature of note is that the distribution is not at an extrema when
Imod is zero. This distribution shift can be explained by a 4 % variation in Ic between
the two constrictions. We additionally verified that the shape and form of the current
modulation matched that of magnetic modulation, by applying a magnetic field to
the device.

4.4 Results analysis

By analyzing the results, we extracted several parameters from the device, including
the total device inductance, the kinetic inductance per square, the total inductance
of each current path, the loop current induced by each fluxoid, and the splitting ratio
of Imod between each constriction. The film’s total inductance can be calculated by
mod
Ltot = Φ0 /(2Isw ), where Isw
mod
is the depth of modulation of the device switching
current and corresponds to half the current induced by a fluxoid. From our experi-
mental results in Fig. 4-6 we found an Isw
mod
of 5.9 ± 0.1 µA, corresponding to a total
inductance of 175 ± 3 pH.
To extract the material’s kinetic inductance, we assumed that the kinetic induc-
tance per square was uniform over the entire patterned film. Kinetic inductance is
expected to increase with current density [78], but such increases are small except
within a few percent of the critical current. It is likely this assumption was violated in
the vicinity of the constrictions [79], but the constrictions represent a small fraction of
the total device inductance. Since the total inductance is just the summation of the
geometric and kinetic contributions, the film’s kinetic inductance per square was then
Lk = Ltot − Lg , where the geometric inductance Lg was numerically calculated, giving

61
a value of 16.7 pH. We then numerically calculated that there were 60.1 squares in
the loop, resulting in a kinetic inductance per square of 2.6 ± 0.1 pH/. This sheet
inductance was larger than the value predicted by Lk ≈ ~Rs /𝜋Δ0 = 1.5 pH/, where
Rs is the sheet resistance just above Tc and Δ0 is the superconducting gap energy
at zero temperature [29]. This difference likely due to degradation of the film during
the fabrication process, increasing Rs or decreasing the RRR. Lastly, by measuring
the periodicity of the triangle pattern, we found that Imod split between the two con-
strictions such that only 2.0 ± 0.1 % passed through the constriction on the left in
Fig. 4-1. This contrasted with our expectation that 8 % of Imod would pass through
that constriction.

4.5 Minimizing Lk

We also fabricated nanoSQUID devices with a number of other designs. One of these
in particular was designed an effort to reduce Lk as much as possible. The design we
chose was an hourglass-shape, shown in Fig. 4-7, which had very small number of total
squares in the loop, resulting in a minimum amount of kinetic inductance. Ideally, we
wanted to reduce Lk enough such that the amount of current induced by each fluxoid
(𝐼Φ0 = Φ0 /Lk ) was greater than the critical current of the constrictions. This goal
can be restated as we wanted 𝐼Φ0 /Ic > 1. Reducing Lk would have increased 𝐼Φ0 while
nominally leaving Ic the same. This would have allowed us to create a nanoSQUID
which could not trap any flux, and thus remove any chance for hysteresis in the device
since it would only have one state – the number of fluxoids 𝑛 in the device would
always be zero.
However, there was a fundamental problem with trying to reduce Lk to below
this level: in a constant-thickness Dayem-type bridge, the current path through a
single constriction – of even the most optimized geometry – always results in enough
inductance such 𝐼Φ0 /Ic ≈ 1. This is explained thoroughly in Ref. [77], and is demon-
strated schematically in Fig. 4-8. As a result of this Lk -Ic relation, a loop made from
a constant-current thickness film and two constrictions always had a 𝐼Φ0 /Ic ratio of

62
at least two, and thus could always store fluxoids. Changing the thickness of the
material does not help–for instance, if the thickness is doubled, Lk drops to half its
previous value and 𝐼Φ0 doubles, but simultaneously Ic is doubled, leaving the 𝐼Φ0 /Ic
ratio the same. The only way around this Lk minimum is to use multiple thickness of
superconducting material – this type of constriction is called as a variable-thickness
bridge in Ref. [77].

Figure 4-7: Hourglass nanoSQUID geometry designed to be as low-inductance as

possible.

Figure 4-8: Figure showing how a bridge which nominally comprises only a few squares
actually has more squares due to the path of the current flow. On the left is a
simulation of current flowing across a narrow constriction. It appears to be about 1
square in total, but the simulation reveals it is more than 3 squares total. This is due
a majority of the current taking an hourglass path, shown by the streamlines on the
left, and represented geometrically on the right.

63
4.6 Variations in the behavior of the nanoSQUID

Before performing a full characterization on the device shown in Fig. 4-1(a), we tested
a number of other devices and found that the output characteristics of the nanoSQUID
depended highly on the material from which they were fabricated, and also on the
uniformity of the constrictions.

4.6.1 Material dependency

We fabricated devices both from not only thin-film niobium (Nb) as described above,
but also thin-film niobium nitride (NbN). Generally speaking, these materials are
very similar: they are both type-II BCS superconductors, they have relatively high
bulk Tc values (∼9 K for Nb, ∼16 K for NbN), and their nominal critical current
densities are on the same order [80]. However, when it came to their operation as
nanoSQUIDs, their output was qualitatively different.
Shown in Fig. 4-6 and Fig. 4-9 are the results from measuring two different
nanoSQUIDs, one which was fabricated from a Nb film, and the other which was
fabricated from a NbN film. As can be seen in Fig. 4-6, the output of the Nb device
did not show any of the underlying fluxoid-state structure like can be seen in Fig. 4-9.
Only the uppermost edges of the valid regions can be seen in the Nb device. However,
in the NbN device, this underlying structure is visible–the edges of the triangle-wave
shape which constitutes the valid-region border extend beyond the border. With this
device, for a given I gate value there can be several distinct distributions of of Isw .
The presence of this underlying structure indicates that the device switched (con-
strictions became hotspots) when fluxoids entered or exited the loop. This process
is shown Fig. 4-10. When measuring the nanoSQUID, the gate bias I gate was fixed,
and then the I bias current was ramped. This trajectory through the I gate -I bias is
shown by the dashed blue line in Fig. 4-10. If, for example, the device started in
state 𝑛 = 2 (point A in Fig. 4-10), when it reached the edge of the 𝑛 = 2 state
(point B in Fig. 4-10), there would be enough current in the loop such that one of
the constrictions would be biased exactly at its critical current. At this bias current,

64
Figure 4-9: Measurements of the switching distributions for a NbN nanoSQUID. This
nanoSQUID was biased using an induced magnetic field from a solenoid instead of
current-biased.

the superconducting state is suppressed (due to 𝐽 ≈ Jc in the constriction), and it

becomes energetically favorable for fluxoids to leave the loop. As this fluxoid is leav-
ing, however, there are two scenarios that can occur, dependending on the material
from which the nanoSQUID is made. The first scenario is that the fluxoid exits out
(either coherently, through a phase slip, or a vortex), superconductivity is restored in
the constriction, and 𝑛 = 1 now. The bias current I bias can continue to be increased
without a voltage appearing on the device until it reaches the edge of the 𝑛 = 1 state
(point B in Fig. 4-10). Once at this border, there are no more valid states for the
nanoSQUID to relax into, and so the constrictions become normal voltage appears
on the device if I bias is increased.

The second scenario is that the fluxoid exits out (again either coherently, through
a phase slip, or a vortex), and the self-heating caused by the passage of the fluxoid
causes so much energy dissipation that a hotspot forms in the nanowire constric-
tion [63]. This is represented by the device switching at point B in Fig. 4-10 – even
though there is a valid state to switch into (𝑛 = 1), the dynamics of the fluxoid dissi-
pation create a hotspot. The presence of this hotspot has a long thermal lifetime and
so before the first hotspot can cool, all the current is diverted to the other constriction

65
where it causes a second hotspot. The result is that both constrictions switch to the
normal state and a voltage appears.

Figure 4-10: Diagram showing the nSQUID states as I bias is increased. First, I gate
is set and I bias is zero (point A) – the device begins in the state 𝑛 = 2. Next, I bias
increased until it reaches the boundary of the 𝑛 = 2 state (point B). At this time, if
the device is hysteretic, it may create a hotspot and latch. Otherwise, the device will
transition to the 𝑛 = 1 state by ejecting a fluxoid, and as I bias is further increased it
exits the valid region and forms a hotspot (point C).

The sheet resistivity of the nanoSQUID thin film determines which of these two
scenarios takes place [81] [82]. In lower-resistivity films like the Nb film, the thermal
energy dissipation that is incurred by the passage of the fluxoid is quickly drawn
out into the banks surrounding the constriction, and even if it causes a Joule-heated
region, this hotspot is quickly quenched because the Nb constriction is very low-
resistance even when normal and cannot easily sustain a hotspot. However, in the
NbN film this self-shunting process cannot occur due to the high sheet resistivity. If
enough energy is dissipated to cause a hotspot, this hotspot will be very effectively
self-Joule-heated by the current flowing through the high-resistance NbN constriction.

66
4.7 Use of the nanoSQUID geometry as an Lk metrol-
ogy tool
This device has proven to be a convenient metrological tool for extracting the ki-
netic inductance of superconducting thin films since it only requires low-frequency
DC currents. The design of superconducting devices which have kinetic inductances
often requires characterization of that inductance to achieve optimal device perfor-
mance, for example tuning the L/R times of superconducting nanowire single photon
detectors or nTrons. Typically, these Lk values are measured by microwave reflection
measurements using a network analyzer [79], or by measuring the magnetic penetra-
tion of the film using two-coil mutual inductance measurements [83]. By patterning
a current-modulated nanoSQUID on the same film as these devices, it instead be-
comes possible to directly extract the thin-film inductance per unit square using only
low-frequency currents–no microwave characterization or tunable magnetic fields are
required.
In summary, we have demonstrated the first modulation of a nanoSQUID using
injected current rather than an applied magnetic field. By adding current asymmet-
rically to the two constrictions of the nanoSQUID, we were able to modulate the
switching current of the device. This modulation method is a direct analog to in-
ducing a loop current by applying a magnetic field. This technique has immediate
applications as a closed-loop feedback mechanism for scanning-SQUID microscopy.
Although the device described here has a large total inductance, and thus low sensi-
tivity when operated as a magnetometer, this method of modulation should generalize
to nanoSQUIDs of any design.

67
68
Chapter 5

The nanocryotron (nTron)

Superconducting devices are vitally important in several fields such as magnetic-field

sensing, quantum and classical computing, photon sensing in communications, and
astronomy. These areas require sophisticated electronics for control, amplification,
and related information processing. To address the challenges posed by these ap-
plication areas, superconducting device families such as Rapid Single-Flux Quantum
(RSFQ) electronics have been developed for the past several decades as possible scal-
able electronics technologies. But all these efforts have been based fundamentally on
the Superconducting Quantum Interference Device (SQUID). The SQUID provides
the non-linearity and gain that is fundamentally required in order to realize sensing
and amplification in electronic systems.
Unfortunately, systems based on SQUIDs (including RSFQ electronics) suffer from
major disadvantages which render them impractical for a variety of applications and
environments. These disadvantages include low gain, high sensitivity to magnetic
fields, difficulty in driving large-impedance loads, and challenges in fabrication. De-
vices based on the SQUID must be biased with substantial amounts of current, but
are limited in how much output current they can source. The result is a device with
low gain. In addition, the requirement that these devices include Josephson junctions
– ultrathin tunneling barriers – renders them notoriously sensitive to fabrication im-
perfections. A variation of an atomic layer in barrier thickness can radically change
the operating point of a device. Finally, SQUIDs are intrinsically the most sensitive

69
magnetic field sensors available. This feature is a blessing and a curse, as SQUID-
based computing devices must be heavily shielded in order to operate. On top of these
issues, the challenge in reproducibly fabricating high-Tc Josephson junctions has pre-
vented SQUID-based active electronic devices with critical temperatures above ∼15 K
from being developed. In large part, the practical concerns outlined here have been a
barrier to the broader application of superconducting systems. By circumventing the
usage of SQUIDs and Josephson junctions entirely, the devices described in this chap-
ter overcome these impediments while at the same time operating with characteristics
that make them suitable for integration with existing SQUID-based systems.
The text that follows is a reprint of work that was originally published in Nano
Letters [84]. Reprinted with permission from McCaughan, A. N. and Berggren, K.
K. A Superconducting-Nanowire Three-Terminal Electrothermal Device. Nano Lett.
2014, 14 (10), 5748-5753, copyright 2015 American Chemical Society.

5.1 Challenges of superconducting circuitry

Superconducting circuits have a long history, beginning 50 years ago with Dudley
Buck’s invention of the cryotron [32] [77]. The cryotron was a device composed of
intertwined superconducting wires, in which a gate wire induced a magnetic field
that would switch the channel wire between a superconducting and resistive state,
thus enabling active control of the channel resistance. Since then, a number of two,
three, and four terminal superconducting logic devices have also been introduced
[36] [33] [34] [38] [35] [85], but were not developed beyond basic characterization.
One exception has been the Josephson junction, which has found widespread success
in a variety of fields, thanks in large part to their >100 GHz operating speeds and
sub-aJ/bit power consumption (9, 10). However, the fundamental property that
makes Josephson junctions attractive–their manipulation of single flux quanta–also
limits the scope of their application, especially in areas that require driving large
impedances, fanning out digital signals, or operating in noisy magnetic environments.
Additionally, building small-scale circuits with Josephson junctions requires either

70
access to a foundry, or very well developed processes as a junction’s key parameters
depend sensitively on sub-Angstrom-scale thickness variation of the tunneling barrier.
To address some of the challenges faced by superconducting circuitry, we have de-
veloped the nanocryotron (nTron). Similar to the cryotron, the nTron uses an input
gate current to induce changes in the resistivity of the channel. However, unlike the
cryotron, the nTron uses a localized, Joule-heated hotspot [10] [11] [12] formed in
the gate to modulate current flow in a perpendicular superconducting channel. The
rapid transition to a highly resistive state enables the nTron to drive large impedances
(>100 kΩ) and fanout to multiple devices (>4). Additionally, although the input cur-
rent sensitivity is sufficient to detect single flux quanta (1-𝜎 grey zone of 66 nA), the
nTron does not require superconducting loops and so can circumvent the issue of flux
trapping in noisy environments. Furthermore, because the hotspot effect occurs in
all known superconductors [86] [87], we expect the design to be extensible to other
materials, providing a viable path to digital logic, switching, and amplification in
high-temperature superconductors. We have characterized the nTron, matched it to
a theoretical framework, and demonstrated a half-adder circuit made from nTron logic
gates. In addition, we have used it as a low-jitter (<25 ps) digital amplifier for su-
perconducting nanowire single-photon detector (SNSPD) pulses. Initially, the nTron
will most likely find use in driving digital storage address lines in Josephson-junction-
based computers, in amplifying SNSPD pulses (e.g. for feed-forward applications in
photonic quantum computing), and possibly in other cryogenic applications requiring
small-signal discrimination with sub-25-ps timing accuracy and moderate (e.g. below
1 GHz) repetition rates.

5.2 nTron device description

The nTron is a thin-film superconducting device with a gate, a drain, and a source
terminal, all connected contiguously with no junctions (Fig. 5-1A). As shown in Fig. 5-
1B, the gate terminal perpendicularly intersects the side of the channel via a narrow
bottleneck called the choke. Current entering the gate terminal, I gate , switches the

71
phase of the choke from the superconducting (S) to the resistive (R) state. The S→R
phase transition is induced in the ∼15-nm-wide choke by locally exceeding the critical
current density Jc of the niobium nitride film. In turn, the resistive phase of the
choke induces a nonlinear suppression of the critical current of the channel, Icchannel .
The resulting dependence of the channel critical current on the gate input current,
Icchannel (I gate ), enables the nTron to produce robust switching and gain. Fig. 5-2
depicts the characterization of a non-inverting nTron amplifier circuit. The observed
form of Icchannel (I gate ) is ideal for a digital logic family: nearly zero modulation of the
channel critical current was seen until a threshold of gate current, Icgate = 2.9 µA,
was reached. Exceeding that threshold produced a 30.5 ± 0.5 % reduction in Icchannel
from its base value Icchannel (0) = Ic0 . This reduction occurred coincidentally with
a nonzero resistance measured at the gate terminal, indicating that the formation
of the resistive hotspot in the choke was responsible for the suppression of Icchannel .
Bias currents greater than 0.9 Ic0 resulted in undesired behavior such as photon- and
noise-induced hotspot generation, while operating the devices at lower bias currents
improved their robustness to source noise and typical ambient magnetic noise.

5.2.1 Fabrication of the nTron

The nTrons described here were fabricated from a contiguous 10 nm film of niobium
nitride (NbN) deposited on a single 2 inch R-plane sapphire wafer. For the simula-
tion, parameters extracted from the device and the film from which it was fabricated
included (1) the device critical current density Jc = 6.6 MA/cm2 , (2) the critical tem-
perature Tc = 12.6 K and its transition width ΔTc = 1.1 K of the film, and (3) the
device sheet resistance Rs = 285 Ω/. For patterning, we spun on 50 nm hydro-
gen silsesquioxane (HSQ) resist and exposed the device patterns in a 125 kV Elionix
electron-beam lithography tool. We then etched the NbN around the patterned HSQ
to complete the nTron fabrication. Contact pads were added by photolithography of
1-micrometer-thick Shipley S1813 photoresist, followed by evaporation of 10 nm Ti
and 25 nm Au and then liftoff. Electrical connections between the sample mount and
device contact pads were made using aluminum wirebonds. For all experiments de-

72
Figure 5-1: (A) Three-terminal circuit symbol. The position of the gate arrow denotes
the location of the choke relative to the narrowing of the channel. (B) SEM of a
fabricated nTron, the inset depicts a close-up of the choke, the area in which the
resistive hotspot is first formed.

scribed here, samples were submerged in a bath of liquid helium. Experiments took
place in ambient magnetic conditions.

5.3 nTron operation

During operation, the nTron moves between three distinct states (Fig. 5-3). Under
the bias condition Icchannel (Icgate ) < I bias < Ic0 , when a logical LOW (I gate < Icgate ) is fed
into the gate input, the channel remains superconducting, and when a logical HIGH
(I gate > Icgate ) is input, the channel becomes resistive. In this context, we used the
nTron as a discrete digital element for performing logical functions. Digital operation
was reproduced for load impedances of 50 Ω, 100 Ω, 1 kΩ, 10 kΩ, 100 kΩ, and the open-
circuit case. In the open-circuit case, the bias condition of Icchannel = 0.85Ic0 = 90 µA
yielded an output voltage of 8.1 V and an input-output isolation of 42.7 kΩ. For the

73
Figure 5-2: Circuit schematic and output characteristics for an nTron in a non-
inverting amplifier configuration. I gate was fixed and I bias was swept from 0 to 120 µA.

10 kΩ case, the leakage current into the load during the digital LOW state was less
than 0.4 nA, within the measurement error of our testing apparatus.

Fundamental to the operation of the nTron is the phenomenon of localized critical-

current suppression, wherein a hotspot sustained by Joule heating suppresses the
superconducting characteristics of the nearby material. Phonons and quasiparticles
diffuse from the hotspot to the surrounding superconductor where they interact with
the superconducting bath, depleting the local Cooper pair population as they relax
back to equilibrium [88]. In the case of thin-film NbN, out-diffusion of hot electrons
is the primary means of thermal energy transfer from the hotspot to the surrounding
material [89] [90]; the characteristic diffusion coefficient for non-equilibrium electrons
has been measured [24] to be 45 nm2 /ps.

74
Figure 5-3: Numerical simulation of the nTron depicting the three states of operation.
OFF state: The device is fully superconducting, bias current is drained through the
channel to ground. Transition state: Current is added to the gate input, forming a re-
sistive hotspot which locally suppresses superconductivity. (inset, upper) Closeup of
the resistive hotspot forming in the choke. (inset, lower) Contour map of Jc suppres-
sion extending from the hotspot. From inner to outer, the bands represent reductions
in Jc by 0 % (blue), 25 % (light blue), 50 % (green), 75 % (orange), and ≥99 % (ma-
genta). ON state: The critical current of the channel is reduced sufficiently that the
bias current triggers the formation of a resistive hotspot in the channel.

5.4 Simulation and design parameters

To corroborate our measurement results and facilitate future designs, we simulated

the nTron device geometry using an established theoretical framework, the two-
temperature model [55], which uses an effective electron temperature Te to represent
the temperature of populations of quasiparticles and Cooper pairs. The simulation
used no free parameters–we instead employed measurements from the device as well
as empirical parameters for thin-film NbN found in the literature [90] [24]. The simu-
lation results showed that when a hotspot was formed in the choke of the nTron, the
effective electron temperature Te of the surrounding ∼100 nm was increased. This
temperature increase corresponds to a decrease in Jc over the same radius, effectively
reducing the total channel critical current. As a result of the hotspot formation, the
calculated Ic0 was reduced by 28 % of its original value, closely matching the mea-

75
sured value of 30.5 %. Additionally, just above Icgate the resistance of the simulated
gate hotspot was 823 Ω, which was comparable to the measured resistance of 832 Ω.
The nTron consists of a single layer of thin-film superconducting material, so its 2D
geometry defines its operation. The layout of the device has several essential design
elements, resulting in a large design parameter space in which future implementations
may be optimized. The most critical design element is the size and location of the
choke region, which is the point of highest current density for the gate input current
and is where the hotspot first forms I gate > Icgate . The choke hotspot is the key to
the nTron’s functionality; it proximitizes the nearby superconducting channel and
induces a suppression of the critical current in that area. The width of the choke
defines the input current level required for the gate hotspot to form and produce
a logical HIGH, which can be approximated by Icgate = Jc 𝑑w choke , where Jc is the
critical current density, 𝑑 is the substrate thickness, and w choke is the width of the
choke. As we observed in our simulation, the formation of the hotspot only suppressed
Jc within approximately one diffusion length 𝐿𝐷 (∼100 nm for thin-film NbN) of its
perimeter. Accordingly, the width of the channel must be on the same length scale for
the hotspot to generate a sharp dropoff in Icchannel . The shape of the channel presents
another design parameter, as it affects the growth of the hotspot after switching. The
shallow swept-curve shown in Fig. 5-1 controls the direction of the hotspot growth–it
will preferentially grow towards narrower regions due to the higher current density–
but does not limit the maximum growth of the hotspot. However, a larger variation
of the width along the channel can produce an nTron with a constrained hotspot size,
which may be desirable in some applications.

5.5 Digital applications and characterization

To demonstrate the suitability of the nTron for digital applications, we used it to build
AND/OR/NOT/COPY gates (Fig. 5-4), and from those constructed a half-adder
(Fig. 5-5). We also performed a number of additional characterization experiments
to demonstrate RF operation, sensitivity, and robustness of the nTron.

76
Figure 5-4: Digital gates based on the nanocryotron. (A) Schematic of a set of
universal logical gates from the basic three-terminal nTron. The AND gate and OR
gate are topologically identical, and are only differentiated by their bias conditions.
AND/OR/COPY were constructed purely from nTrons, while the NOT gate required
a shunt impedance for the bias (in this case a resistor). (B) AND-gate timing diagram
for pipelined logic propagation. Once gates A and B have valid inputs, the bias current
is enabled and the resulting output can be used as an input for the next stage. 𝜏𝑏𝑖𝑎𝑠
denotes the propagation delay due to the low-rate bias electronics.

5.5.1 nTron logic gates

Fig. 5-4A shows the COPY/NOT/AND/OR gates. All of the gates required only
one or two nTrons, except for the NOT gate which used an additional 330 Ω shunt
resistor and another nanowire constriction in addition to a single nTron. The COPY
gates operate similar to a non-inverting version of the FET: when a logical HIGH
(I gate > Icgate ) is input to the gate, the channel becomes resistive and the channel
bias current is diverted into the output, generating a HIGH output. When a logical
LOW (I gate < Icgate ) is input to the gate, the channel bias current drains directly to
ground without generating a resistive region in the channel. As a result, the output is
effectively shorted to ground and is thus a logical LOW. The AND and OR gates are
the result of putting two nTrons in parallel. In the case of the OR configuration, when
either gate input received a logical HIGH, the combined Icchannel of the two nTrons
would be reduced enough such that the channel bias current would switch the device.
The AND configuration was achieved by biasing the channel below this point, to
the point where the combined channel critical currents only dropped below the bias
current when both gate inputs were HIGH. This topological equivalence between the
AND and OR gate opens the possibility to dynamically reprogram an otherwise fixed
logic circuit.

77
 For the NOT gate, when a logical LOW was input, Icchannel was greater than the
bias current, allowing all the current to pass through it without switching. However,
the pulldown constriction critical current was designed to be smaller than the full
channel bias current, which caused the pulldown constriction to switch and diverted
the excess current into the output, producing a logical HIGH out. In the case of a
HIGH input, the suppressed Icchannel (I gate ) was less than the bias current. As a result,
the nTron transitioned to the ON state, and current was diverted to ground through
the shunt resistor. The pulldown constriction served to tie the output to ground,
even in the presence of the unavoidable small current leaking through the resistive
channel.

5.5.2 The nTron half-adder

The half-adder experiment consisted of fabricating several individual gates on a sin-

gle chip, connecting them via wirebond, and biasing them in a pipelined manner to
perform the summing computation. We built a custom 16-channel combined ADC
and DAC system to handle the multiple gate and bias inputs, and to read out their
bias voltages. For the gate inputs and channel biases, current sources were approx-
imated using the DAC voltage channels in series with 100 kΩ resistors. The ADC
channels were used to record the status of each gate’s output. Due to the pipeline-
propagation of the half-adder, in combination with the low-rate DAC/ADC system,
the computation of the half-adder outputs required approximately 0.8 s to complete.
No external amplification was necessary for readout, and the lower bit and carry
bit and generated 38.8 mV and 17.0 mV respectively across their 330 Ω load resistors.
The sampling rate of the room-temperature electronics setup limited the computation
rate to 1.2 Hz. In this mode, longer time scales present more opportunity for noise to
erroneously switch the devices. Despite this disadvantage, exercising the circuit over
4,000 cycles spanning 55 minutes the circuit produced only seven errors.
We operated the half-adder in a latching, pipelined fashion, where valid inputs
were translated into valid outputs for the next stage only upon the enabling of the
gate bias current (Fig. 5-4B). After the computation was completed and the final

78
outputs were recorded, all the input and bias currents were shut off to unlatch the
gates and reset the computation. For the purposes of this demonstration, operating
in the latching regime enabled us to tolerate potentially large fabrication defects in
these first devices. In the latching regime, variations between the OFF-state input
impedances of the gates did not matter because each stage was able to drive arbitrarily
large input impedances in the next stages–to operate the nanowires in a non-latching
regime, output impedances must be more carefully controlled [52]. No electrical or
magnetic shielding was necessary; the nTron does not require fully-superconducting
loops to function, and so resistive wirebonds and contact pad connections were used
to prevent flux trapping during operation.

5.5.3 Measuring the threshold sensitivity of the nTron

We tested the nTron as a comparator to characterize the current sensitivity of the

gate input (Fig. 5-6). The experimental setup was the same as that of the 10 MHz eye
diagram characterization, only differing in the current bias and readout scheme. First,
we applied a 52.8 µA bias current to the channel in order to prime it for transition
from the ON to OFF state when I gate exceeded Icgate . We then ramped the gate
current from 0 to 15 µA at a rate of 16.7 nA/µs, recording the output current through
the load resistor RL on a 1 GHz oscilloscope. We repeated this experiment 11,000
times, plotting the results as a histogram (Fig. 5-6B). From the histogram, we found
a mode switching value of 2.91 µA with a 1-𝜎 grey zone of 66 nA.

5.5.4 Measurement of a pseudo-eye-diagram at 10 MHz

To characterize the device at higher frequencies than the operating frequency of the
half adder experiment described above, we input a 10 MHz square wave into the
device to generate a modified eye diagram. With a 1.46 kΩ output load, the nTron
was able to convert a 3.10 ± 0.02 µA input square wave into a 62.7 ± 1.2 µA output
square wave, corresponding to a signal gain of 20.2 (even though this circuit is a
digital comparator, in this chapter we use the term gain to describe the ratio of

79
the output current to the gate input current). At the sampling point, the signal-
to-noise ratio was 168. Applying a magnetic field perpendicular to the device plane,
swept between ±7.4 mT, had no observable impact to the eye diagram characteristics.
Additionally, we measured the input-level sensitivity by operating the device as a
current comparator, measuring an Icgate of 2.91 µA with a 1-𝜎 grey zone of 66 nA.
Using the experimental setup shown in Fig. 5-7A, we generated a 10 MHz eye
diagram for the nTron digital amplifier. We accomplished this by sending one current
square wave into the device channel, and one current square wave into the gate
port (delayed by approximately 10 ns relative to the channel square wave). The
channel square wave primed the nTron to switch from the ON to the OFF state,
such that when the rising edge of the gate square wave arrived, the nTron switched,
generating an output current which was read out by the scope. Note that the eye
diagram shown in Fig. 5-7B is modified from a standard diagram: it only shows the
transitions LOW→HIGH and HIGH→LOW, and omits the LOW→LOW transition
as well as the HIGH→HIGH transition. The testing was performed in the latching
regime, meaning the bias was shut off each cycle, eliminating the possibility of a
standard HIGH→HIGH transition. The LOW→LOW transition was also omitted,
as the square-wave generator used for this test did not have a pseudo-random bit
stream (PRBS) function. The two shown transitions were captured by allowing the
oscilloscope to trigger on both the rising edge and falling edge. This demonstration
was primarily qualitative, meant to show the large signal-to-noise ratio at 10 MHz-
future work will include the high-frequency analysis of the nTron, as well as the
generation of a more typical (and lower SNR) eye diagram, covering both the latching
and non-latching regimes with all four transitions.

5.5.5 Integration with a superconducting nanowire single-photon

detector

We tested the nTron’s ability to amplify pulses as well as established an upper bound
to its jitter by using it to read out a superconducting nanowire single-photon de-

80
tector [50] (SNSPD). When detecting a photon, SNSPDs produce millivolt-scale mi-
crowave pulses with sub-100 ps rising edges. The integrated device used these pulses
as input into the gate of a nTron amplifier. We monolithically fabricated the SNSPD
and nTron on the same film within a 100 µm2 area, connecting the SNSPD output to
the gate of the nTron. Our circuit design (Fig. 5-8) enabled us to bias each device
separately, as well as simultaneously read out the unamplified SNSPD pulses and
nTron-amplified pulses for comparison. When the SNSPD was illuminated with a
1550 nm sub-picosecond laser, output pulses were produced from both devices con-
currently. We compared the SNSPD output to the nTron-amplified output, and
observed a factor of 2.9 increase in signal pulse amplitude. This increase in ampli-
tude proportionally increased the slew rate of the rising signal edge, resulting in a
reduced jitter when measured by our scope. This test also served to demonstrate the
non-latching operation of the nTron: after the pulses were produced, both devices
reset on their own within ∼10 ns, without any modification of the bias conditions.
Jitter was measured between the sync edge of the sub-ps laser and the rising edge of
each device’s output pulses. As can be seen in Fig. 5-8, the unamplified pulses had
a full-width half-max jitter of 41.3 ± 0.3 ps, while the corresponding nTron-amplified
pulses showed a reduced full-width half-max jitter of 23.8 ± 0.2 ps. This value of
23.8 ps is the total system jitter, and thus also sets an upper bound to the nTron
input-to-output device jitter, which is presumably smaller, as state-of-the-art NbN
SNSPDs [91] alone produce jitters on the order of 20-40 ps.

As shown in Fig. 5-9, the SNSPD and nTron pulse amplifier were integrated as sin-
gle circuit. Current biasing was accomplished through the use of inductive splitting,
where the inductance was provided by the kinetic inductance of the nanowires. The
SNSPD nanowires had a width of 60 nm, and the inductor nanowires had widths of
200 nm. The entire circuit occupied an area of approximately 100 µm2 . The 50 Ω lines
were high-frequency coaxial cable running between the sample and room-temperature
electronics. During operation, we biased the SNSPD at 35 µA and the nTron at 95 µA.
Expected operation was ensured by measuring: photon sensitivity for the SNSPD and
amplifier when biased separately; critical current suppression in the nTron channel

81
when the SNSPD was overbiased (creating a hotspot in the gate); count rate from
both outputs when biased together; count rate vs. I bias ; and count rate from the
SNSPD vs ISNSPD . The results of these measurements allowed us to conclude that (1)
the nTron amplifier, at a bias of 95 µA, was not photosensitive; (2) there was a 1:1
correspondence in counts between the two outputs (one amplifier pulse per SNSPD
pulse); (3) no counts were generated in the amplifier when only the SNSPD was
biased, and vice-versa.

5.5.6 Power dissipation and clock rate

In addition to jitter characterization, to assess the utility of the nTron in future

applications, we discuss here its potential clock rate limitations. The maximum clock
rate of an nTron is determined by both the superconducting and thermal properties
of its constituent materials. Although we presently have no direct measurement of the
maximum operating speed of an nTron circuit, their thermal and electrical properties
are expected to be identical to SNSPDs, whose reset times are well understood [90]
[51]. In SNSPDs, the limiting timescale for reset is either the inductive electrical
time constant or the thermal recovery time from the normal to superconducting state,
whichever is longer. After the thin film returns to the superconducting state, high-
energy phonons must escape into the substrate for the device to return to its initial
non-resistive state. This geometry-independent recovery period has been measured
to be below 100 ps [92] [93], suggesting an maximum clock rate on the order of 1 GHz
for thin-film NbN.
In many cases, switching power dissipation is a factor in determining applicability
of a new electronic device. To estimate the optimal power usage per switching event
in the non-latching (pulsed) regime, we used conservative values for the nTron device
inductance (∼1 nH) and the channel bias current (∼100 µA). These estimates yield
a minimum switching-energy of 𝐸 ≈ 𝐿𝐼 2 = 1 × 10−17 J/bit for our initial devices,
within two orders of magnitude of current SFQ technologies [94]. These speed and
power estimates may be improved upon by using different materials or geometries or
by scaling to smaller dimensions, which may be desirable in digital applications. Even

82
for applications in high-performance computing, where Josephson junction electronics
may be preferable, the nTron can provide ancillary utility in the form of interfaces
with sensors and high-impedance loads such as readouts and data-storage address
lines.

5.6 High-Tc YBCO nTrons

In a collaboration with Riccardo Arpaia and Floriana Lombardi of Chalmers Uni-
versity, we were able to test an nTron fabricated from the high-temperature super-
conductor YBa2 Cu3 O7 (YBCO). YBCO is an exotic cuprate compound which has a
Tc of 90 K, well above the boiling point of liquid nitrogen. The Chalmers group had
a well-developed thin-film YBCO patterning process, and so by sending them the
geometry of the nTron, they were able to recreate the geometry in YBCO, just as we
had done with NbN. An image of a chip they fabricated, which was 50 nm of YBCO
capped with 50 nm of gold, is shown in Fig. 5-10.

5.6.1 Cryostat experimental setup

Testing the YBCO nTron required some experimental alterations from the NbN mea-
surement setup. YBCO does not tolerate moisture well and thus cannot be cooled
down except under vacuum, for fear of degrading the material with condensation.
Instead of dunking the sample directly in liquid helium, we built a custom PCB sam-
ple mount for the sample. We were able to aluminum wirebonds to connect to the
YBCO device because the YBCO was capped in gold – with bare YBCO, it likely
would have been difficult to make a solid connection, as YBCO is ceramic-like in most
of its physical properties and thus a bond is not easily formed by the wirebonder’s
ultrasonic bonding process.
The PCB we built with the mounted chip is shown in Fig. 5-11. For each nTron
bonded, there was a direct input to the gate, as well as a four-point connection setup
which measured the voltage across the channel (two points on the drain terminal,
two on the source terminal) and passed that voltage to a SRS560 low-noise voltage

83
preamplifier. The reason the voltage preamplifier was necessary in this setup and
not in the NbN setup was because of the thick gold capping layer on the YBCO.
This layer reduced the normal state resistivity of the nTron channel to less than 1 Ω,
making it difficult to distinguish the superconducting state from the normal state
without amplification.

5.6.2 YBCO nTron results

After cooling down the device to 4 K under vacuum in the cryostat, we measured the
IV curve of the nTron channel. The result of the IV curve for zero gate bias is shown
in Fig. 5-12. The nTron channel, which was 400 nm wide, had a critical current of
6.1 mA and was non-hysteretic due to the large conductance of the capping gold.
We then measured the IV curve as a function of gate bias, which is shown in
Fig. 5-13. These measurements showed that for every 200 µA of gate current added,
the IV curve was shifted by exactly 200 µA. This result indicates that there was no
hotspot-effect action being produced at the nTron choke. The channel critical cur-
rent varies linearly with the gate input, even when the gate choke is biased above its
critical current. In normal nTron operation, the change of the choke from supercon-
ducting normal should trigger a flood of quasiparticles into the channel, suppressing
the superconductivity and shifting the IV curve significantly more than the 1:1 ratio
seen in Fig. 5-13. The most likely reason for the lack of the hotspot effect is that
the gold provides an excellent thermal and quasiparticle shunt, carrying away hot
electrons before they are able to deposit energy on the superconducting channel. The
next step for testing the nTron in high-temperature superconducting materials will
be to use YBCO devices that are not capped in gold.

5.7 Conclusion
The central result of this section was the design, characterization, and implementation
of the nTron, a superconducting three-terminal electrothermal device that does not
rely on Josephson junctions. The nTron was used to demonstrate digital operations,

84
as well as pulse amplification with picosecond timing resolution. The nTron has im-
mediate application to the readout of superconducting sensors such as those used in
quantum computing (e.g. flux qubit readout, transmission line multiplexing), commu-
nications (e.g. single-photon detectors, driving optical modulators), and astronomy
(e.g. as an alternative amplifier to SQUIDs for reading out large detector arrays). Its
operating characteristics are also complementary to Josephson-junction-based tech-
nologies, and integration could enable those technologies to perform otherwise-difficult
tasks such as memory line-driving and fanout in exascale computing. Additionally,
the ease of fabrication, transistor-like logic-gate design, and extensibility to other ma-
terials mean that small-scale superconducting circuits are now accessible with minimal
fabrication investment and straightforward circuit designs.

85
Figure 5-5: Experimental demonstration of an nTron half-adder. (A) Half-adder
circuit schematic constructed from logical gates. Single inputs were provided into
the initial (yellow) COPY gates, which acted as buffers for the signals Input A and
Input B, each with a fanout of three. Connections to ground and between gates were
made with low resistance, non-superconducting links. (B) Per-channel output for the
half-adder for computation of 0+0, 0+1, 1+0, and 1+1, repeated twice. HIGH (1)
and LOW (0) current values were input to Input A and Input B, and after a bias
electronics delay 𝜏𝑏𝑖𝑎𝑠 , the lower bit and carry (upper) bit outputs represented the
resulting sum of the inputs. The red text overlay of ones and zeros corresponds to
HIGH and LOW values.

86
Figure 5-6: The current comparator experiment used to test the input sensitivity of
the nTron. (A) Circuit diagram for the nTron current comparator. The channel was
biased at a fixed value, and the gate was ramped until output appeared at the scope.
(B) Histogram of I gate values for the gate current at which the comparator switched
and produced an output voltage at the scope.

87
Figure 5-7: Circuit schematic for 10 MHz eye diagram experiment. (A) Circuit dia-
gram for the nTron 10 MHz eye-diagram experiment. The area in blue represents the
portion of on the sample holder and submerged in liquid helium at 4.2 K. Placing
the resistors close to the device allowed us to convert the incoming voltage square
waves to a low-amplitude current square waves. The resistors 𝑅𝐿 , 𝑅𝑏𝑖𝑎𝑠 , and 𝑅𝑔𝑎𝑡𝑒
were 1.46 kΩ, 20.8 kΩ, and 42.0 kΩ, respectively (as measured at 4.2 K). (B) 10 MHz
modified eye diagram output taken directly from oscilloscope

88
Figure 5-8: Jitter measurements for an nTron integrated as an amplifier for a su-
perconducting nanowire single-photon detector (SNSPD) pulses. Detection of laser
photons from a sub-ps laser by the detector (inset, purple ‘S’ box) generated an elec-
trical pulse on Port 1 (inset, red) and also triggered a concurrent, amplified pulse from
the nTron on Port 2 (inset, blue). Plotted is a histogram of the relative delay between
the laser sync edge and the resulting electrical pulse edges of the unamplified SNSPD
(red dots) and nTron-amplified output (blue dots). Gaussian fits to each data set are
shown as solid lines. The reduced jitter in the amplified signal is due to increased
signal amplitude. (upper right) Device schematic of the integrated SNSPD-nTron
pulse amplifier.

89
Figure 5-9: Circuit schematic for the SNSPD and nTron pulse amplifier experiment.
(A), Device circuit schematic. The inductors were made by patterning long nanowires,
which intrinsically produce kinetic inductance. The length of the inductor nanowires
(and thus their total inductance) were scaled against the SNSPD, which had an
approximate kinetic inductance of Lk ≈ 25 nH. (B) Room-temperature readout and
bias electronics. Pulses generated from the device and output to the coax in (a)
arrived at the other end of the coax, shown in (b), where they were amplified with
three 20-3000 MHz amplifiers in series before being input to the scope.

Figure 5-10: Microscope image of the YBCO chip created for us by Lombardi group
at Chalmers University. The material is 50 nm of YBCO capped with 50 nm of gold
on an MgO substrate.

90
Figure 5-11: YBCO sample wirebonded to the custom PCB and mounted in the
vacuum cryostat.

Figure 5-12: I-V curve shown for the YBCO nTron, with I gate = 0.

91
Figure 5-13: I-V curves of the YBCO nTron versus as a function of varying I gate .
Shown are values of I gate in increments of 200 µA.

92
Chapter 6

The current-crowding cryotron

(yTron)

The yTron is a three-terminal device which is capable of sensing superconducting loop

currents inline without perturbing the superconducting state. This functionality is
unique to the yTron, because typically this type of measurement must be done either
by magnetically coupling to an external loop or by perturbing the superconducting
state in enough to allow flux to pass out of the loop. In this chapter I describe the basic
description, characterization, and application of this device. Like the nanoSQUID
chapter, the text that follows constitutes a preliminary writeup of work which will be
submitted to a journal for publication.

6.1 Device description

The yTron is made from a single layer of superconducting niobium nitride (NbN)
thin film which has been patterned into a “Y” shape. As shown in Fig. 6-1, the drain
and the gate terminals form the upper arms of the Y, joining together at a sharp
intersection where they connect to form the source terminal. The width of the source
terminal is equal to the summed widths of the drain and the gate terminals.
The material used to make the device shown in Fig. 6-1 was thin-film NbN which
was deposited on SiO2 in a DC magnetron sputtering system at 800 ∘C. This NbN

93
Figure 6-1: Scanning electron micrograph of a yTron with a 200 nm gate and 100 nm
channel. The low contrast of the edges that form the intersection are due to the
tapering of the e-beam resist in that region.

film was measured to have a sheet resistivity of 437 Ω/, and a Tc of 10.65 K. After
deposition, contact pads shapes were patterned onto the film via photolithography.
After patterning, 10 nm Ti and 50 nm Au were deposited to form the contact pads.
After performing liftoff to remove the excess Ti and Au, the sample was cleaned by
sonicating it in acetone for 5 min. Once cleaned, 4% HSQ was spun onto the sample at
3 krpm for 60 s, creating an HSQ film approximately 50 nm thick. The yTron patterns
were then exposed into the HSQ resist by electron-beam lithography on an Elionix
125 kV tool which had a beam current of 1 nA with an areal dose of 3840 µC/cm2 .
The patterns were developed by developing the sample in 25% TMAH for 2 min and
then rinsing with deionized water for 30 sec. Lastly, the sample was etched in 1.33 Pa
(10 mTorr) CF4 in a PlasmaTherm RIE at 50 W of RF power for 3 min. The 50 W
of RF power was spread over 100 cm wafer in the etcher, giving a an areal etching
power of 6.4 mW/cm2 .

94
Figure 6-2: Fabrication steps for patterning the yTron out of a thin NbN film. (a)
NbN is deposited on an SiO2 substrate. (b) Titanium-gold contact pads are added by
a photolithographic liftoff process. (c) The e-beam resist HSQ is spun on the sample.
(d) The HSQ is patterned by an e-beam tool and developed. (e) The sample is etched,
leaving NbN only in the areas protected by the HSQ and contact pads.

6.2 Device operation

The functionality of the yTron as an inline current sensor comes from the dependence
of the channel switching current on the amount of current flowing into the gate.
Although counter-intuitive, adding current to the gate actually increases the switching
current of the channel. This increase is the result of to current crowding [57] [95] due
to current density mismatches at the intersection of the two upper arms. Current
crowding describes the increase in current density that takes place for currents that
are bent around sharp features, as shown in Fig. 6-3. The yTron uses this phenomenon
as a means controllably break down superconductivity at the intersection between the
gate and the channel nanowires.

6.2.1 Current crowding and the channel critical current

We can explain the impact of current crowding using the three examples shown
in Fig. 6-4. In Fig. 6-4(a), the yTron is biased such that the current densities through
the gate and the channel are equal. As a result, the current densities on either side of

95
Figure 6-3: Simulation of current flowing around a sharp corner. Current-streamlines
are shown, and the coloration indicates the current density, which is at a maximum
around sharp corner feature.

intersection are equal and the current streamlines join together smoothly as they flow
to the source. In Fig. 6-4(b), the current flowing into the gate has been reduced, and
as a result the current streamlines from the channel bend slightly around intersec-
tion point so that they can redistribute evenly by the time they arrive at the source.
In Fig. 6-4(c), the gate current has been shut off, and the current streamlines from the
channel bend sharply around the intersection point as they redistribute. The current
flowing through the channel is subject to a large amount of current crowding as it
bends around the intersection corner. In other words, the channel current density–
which is uniform at the channel terminal entrance–increases in magnitude near the
intersection. The increased current density acts as a weak point for the breakdown
of superconductivity, either through the entry of vortices [63] or the formation of a
hotspot [60]. As a result of this process, reducing the gate current decreases Isw
ch
, the
switching current of the channel.

Modulation of the yTron channel switching current can be best explained by ex-
amining the horizontal current components near the intersection point as shown in
Fig. 6-5. Components of the gate current which are tangential to the intersection
boundary (and thus produce current crowding) can oppose and cancel out the hor-

96
Figure 6-4: Current flow streamlines in the current-crowding cryotron for various gate
biases. (a) The gate is biased at the same current density as the channel, and there
is minimal current crowding at the intersection. (b) The gate is at half the channel
current density. (c) The gate carries no current, and as a result the streamlines
from the channel curve sharply around the intersection, causing significant current
crowding.

izontal components from the channel. This cancellation results in a total reduction
of the current crowding at the intersection. In this way, current added to the gate
diminishes current crowding in the channel, and as a result increases the channel
switching current.

6.2.2 Output characterization

Fig. 6-6 shows the characterization of a device with the same dimensions as the yTron
shown in Fig. 6-1. The dependence of the channel switching current Isw
ch
on the gate
current input I gate is approximately linear over a large range of I gate values. This
linearity occurs because the magnitude of the current crowding is dependent on the
horizontal components of the current density, with respect to the intersection edge.
Equal current densities on either side of the intersection produce a minimum in the
current crowding, because the horizontal components at the intersection point are

97
Figure 6-5: Simulation of two yTron bias points showing the summation of horizontal
currents. (a) Current flowing in from the upper left arm and current flowing from the
upper right arm produce horizontal current components which mostly cancel each
other out, reducing current crowding at the intersection point. (b) Current only
flowing in from the upper right arm. In this scenario there is no cancellation of
horizontal current components, and so there is a large amount of current crowding at
the intersection point.

equal and opposite, and thus cancel each other out. This cancellation is conceptually
similar to a Wheatstone bridge, where current flow across the midsection of the bridge
is minimized when the contributing currents from either side are equal and opposite.

Figure 6-6: Channel switching current modulation versus gate current.

In developing the yTron, we tested several devices with varying widths of the
channel and gate nanowires, and found that the ratio of the widths of the channel
and gate determines the dependence of Isw
ch
on the gate current. Specifically, this

98
ratio determines the slope of the linear regime of Isw (I ) in Fig. 6-6. For example,
ch gate

the device pictured in Fig. 6-1 has a channel width of 200 nm and a gate width
of 100 nm, giving a channel-to-gate width ratio of 0.5. This 0.5 ratio corresponds
approximately to the slope of the linear portion of Fig. 6-6, which is 0.61. We tested
a variety of devices with gates and channels ranging from 100-800 nm, and which
had channel-to-gate width ratios between 1:8 and 8:1. We found that the slopes of
their Isw (I ) graphs ranged from 0.33 to 2.93, and although there was quite a bit
ch gate

of scatter with respect to the absolute slope values (even between nominally-identical
devices) generally the smaller the channel-to-gate ratio, the lower the slope value.

Looking closely at Fig. 6-6, we observed that the dependence of Isw

ch
on I gate had
well-segmented regions of behavior. Around I gate = 0 there was a linear region of
(I ) , where the current densities on either side of the intersection summed lin-
ch gate
Isw
early to produce current crowding as described earlier. This region was where yTron
would be most useful as a current sensor, as the straightforward linear dependence
can be used to infer I gate from Isw
ch
measurements. However, as shown in Fig. 6-6
there are two additional regions on either side of that range. For the device mea-
sured in Fig. 6-6, these regions corresponded to when I gate was above 38 µA or below
−14 µA. We believe the change of Isw (I ) away from a simple linear slope occurs
ch gate

when design or fabrication imperfections around the intersection begin to cause more
current crowding than the intersection point itself. When we tested device on the
same chip with different dimensions, the size of this linear region was variable – in
one case, it ranged from −5 µA to 20 µA, and in another cases it ranged from −25 µA
to 50 µA. It will not be possible to rule out other causes without further characteriza-
tion, but if small defects existed a few tens of nanometers away from the intersection
point, they could act as focal points for current crowding which only exceeded the
intersection current crowding once the gate current became high or low enough.

99
6.3 Device design considerations

6.3.1 Material considerations

There are several aspects of the yTron design which impact its operation. The first
characteristics which must be considered are those of the superconducting material
from which the device is fabricated. The superconducting film thickness must be less
than the material’s penetration depth 𝜆 in order for the current crowding to work
as described. In a thicker superconductor with a non-uniform kinetic inductance,
current will not be distributed evenly across the cross-section of each arm of the
yTron, altering the effect of current crowding. By making the device from a film
thinner than 𝜆, the device has an (approximately) uniform sheet kinetic inductance
that enforces the current redistribution shown in Fig. 6-4.
The other material parameter which must be considered when designing a yTron
is the superconducting coherence length. To minimize tunneling effects, arms of the
yTron should be wider than the coherence length (in thin-film NbN 𝜉 ≈ 4 nm [96]).
This minimum width is necessary because the yTron relies on an non-uniform current
distribution across the width of the wire to produce current crowding. This distri-
bution would not be possible if, for example, all dimensions of the yTron channel
were narrower than the coherence length and thus the channel was effectively one
dimensional.

6.3.2 Geometric considerations

The geometry of the yTron determines its operation. The dependence of the channel
switching current Isw
ch
on the gate current I gate is based on three factors: (1) the sharp-
ness of the intersection point, (2) the widths of the channel and gate nanowires, and
(3) the angle at which the gate and channel intersect. In testing various geometries,
we did not vary the third factor, and so will not detail its impact here. Nominally,
an infinitely sharp point at the intersection tip would result in an infinite amount
of current crowding for any streamline bending around the intersection. As a result,

100
the current density at the corner should diverge is cases like Fig. 6-4b and Fig. 6-4c.
However, the sharpness of the intersection point will always be ameliorated by two
factors: rounding caused by the practical fabrication limits of e-beam lithography
(∼15 nm for the device in Fig. 6-1), and a radius-of-curvature effect (as described in
Ref. [57]) in the superconductor produces a rounding of the intersection point on the
order of the material superconducting coherence length, even for a perfectly sharp
intersection.
In the scenario where the lithographic rounding is significantly smaller than 𝜉, we
can bound the effects of current crowding by approximating the yTron intersection as
a 180∘ hairpin turn. With this approximation, the channel switching current can be
reduced by at most a factor of 𝑅 = 𝜋𝜉/𝑤, where 𝑤 is the width of the nanowire. For
√︀

the device described here, 𝜉 = 10 nm and 𝑤 = 200 nm, and so the range of values Isw
ch

can take will be between 39.6 % and 100 % of the switching current of a 200-nm-wide
wire. This value sets a limit to the dynamic range of the yTron.

6.4 Operating modes of the yTron

The yTron can either be used as a latching digital discriminator, or a three-terminal

controllable weak link. These two modes are schematically drawn in Fig. 6-7. The
mode in which the device operates depends on whether the channel IV curve is hys-
teretic, which in turn depends on the shunt resistance in parallel with the channel
nanowire [1] [82] [81]. If the channel IV curve is hysteretic (large shunt resistance), the
device operates as a latching digital discriminator: when the current in the channel
exceeds Isw (I ), a hotspot will form and stabilize, producing a DC output voltage
ch gate

between the drain and source terminals. Conversely, if a low enough shunt resistance
is present across the channel and it becomes non-hysteretic, the channel cannot sup-
port a stable hotspot. Instead, the channel will behave like a Dayem-bridge-type
weak link, and when biased above Isw (I ) flux will be allowed to pass across the
ch gate

channel by way of vortex crossings. Fig. 6-8 shows IV curves of the yTron described
here for various values of I gate . Each curve looks approximately like a standard IV

101
curve for a superconducting NbN nanowire, with a switching current that depends
on I gate . Since the material used was thin-film NbN, the curves are highly hysteretic
as expected.

Figure 6-7: The two operating modes of the yTron, which match the operating modes
of a typical nanowire [1] based on whether or not the nanowire is hysteretic. (a) The
channel is shunted by a small resistance in parallel. Flux flows across the channel, but
the small resistance shunts the bias current and prevents a stable Joule-heated hotspot
from forming. (b) The channel has a large shunt resistance. Significantly more power
is dissipated in the channel, allowing a self-sustaining Joule-heated hotspot (normal
region) to form.

Effectively, the yTron acts as a three-terminal controllable weak-link: the amount

of current flowing into the gate terminal controls the passage of flux across the channel
terminal. At no point during this process is the current in the gate modified; the
vortex flow (or hotspot) is isolated in the channel nanowire, leaving the current flowing
through the gate undisturbed. This allows nondestructive readout of superconducting
loop currents, which previously was only possible through magnetic coupling.

6.4.1 Isolation of the gate from the channel

Isolation between the gate and channel is a key feature of the yTron. Since the yTron is
fabricated from a continuous superconducting film, “isolation” in this context does not
mean electrically disconnected like the gate of a MOSFET, but instead the isolation
of the gate from changes in the channel. The key feature of the yTron is that even

102
Figure 6-8: IV curves of the yTron channel for different values of I gate . Each IV curve
looks approximately like a nanowire with a different Ic value.

when the state the channel changes–e.g. between the superconducting state, vortex
flow state, or hotspot state–the superconducting state of the gate nanowire is not
perturbed. Due to the indirect nature of the current-crowding-based modulation of
ch
Isw , a voltage state in the channel (source to drain) does not produce any voltage
on the gate (gate to drain). As an example, let us assume we have a non-hysteretic
yTron whose channel is biased at Ich , just below Isw
ch
. When I gate is reduced, Ich > Isw
ch
,
and vortices will begin to flow across the channel, producing a voltage between the
source and drain terminals. Despite the fact that flux is passing across the channel
nanowire, the gate nanowire is still entirely superconducting and no flux is able to
cross. One concern we had when testing the yTron was that excited quasiparticles
generated by the hotspot or vortex crossings could diffuse into the gate and perturb
the current flowing there. In thin-film NbN, this diffusion length is ∼100 nm [84], on
a similar scale to the nanowire widths. However, we found that as long as the total
power dissipation in the channel was below 350 nW, it did not impact the switching
current of the gate.

103
6.5 Measurement details

The results of Fig. 6-6 were measured using a current-sweeping setup. Each vertical
slice of the graph corresponds to a set of 100 measurements of Isw
ch
for a given value
of I gate . Measurement of each distribution began by setting the I gate input, using
an SRS variable battery source with a 10 kΩ series resistor in front of the positive
terminal. A current ramp was then applied to the channel using an arbitrary waveform
generator in series with a 10 kΩ resistor. Isw
ch
was then recorded as the value of the
current ramp when a voltage appeared at the channel input (triggered at 1 mV),
indicating the switching of the channel from the superconducting to normal state.
This measurement was then repeated an additional 100 times to yield the distribution
for Isw (I ). Extended details about measuring the critical current distribution can
ch gate

be found in Section 7.1.1. All measurements were made at 4.2 K by submerging

the sample directly into liquid helium. The sample was shielded from spurious RF
radiation by a thin copper box, but no magnetic shielding was used.

6.6 Inline, nondestructive measurements of a quan-

tized superconducting loop current

Due to its ability to read inline currents, a natural application for the yTron is the
readout of quantized currents in a superconducting loop. By placing the gate of
the yTron inline with a superconducting loop, we were able to use the yTron to
nondestructively read out the number of discrete fluxons (𝑛) trapped in a the loop.
We successfully resolved the adjacent fluxon states (𝑛, 𝑛+1, etc) of the loop, and were
able to read out those states several thousand times consecutively without changing
the value of 𝑛. This application was possible because the gate (loop) current can
be inferred from Isw
ch
, which can be measured without allowing flux into or out of
the loop. Each trial of the experiment consisted of two alternating steps: First we
measured Isw
ch
several thousand times consecutively (Fig. 6-9). We then heated part
of the gate-source loop above Tc to break the superconducting loop temporarily and

104
allow the number of stored fluxons 𝑛 to change randomly (Fig. 6-10).

Figure 6-9: Readout procedure for the inline nondestructive measurement of the
superconducting loop. The gate-source loop started out at rest (left), and then the Ic
of the channel was measured by ramping 𝐼𝑟𝑒𝑎𝑑 until a hotspot formed in the channel
(right). 𝐼𝑟𝑒𝑎𝑑 was then turned off and the system returned to rest (left). This process
was able to be repeated several thousand times without changing 𝑛, the number of
fluxons trapped in the gate-source loop.

By measuring the channel current after heating it repeatedly, we observed step-

like behavior in Isw
ch
, which is shown in Fig. 6-11 which indicates the nondestructive
readout of quantized loop currents. During each trial, the distribution of Isw
ch
was
fixed. Only when the loop was heated and re-cooled did the distribution of Isw
ch

change. Moreover, when the loop was heated and re-cooled, the distribution of Isw
ch

shifted by integer multiples of 2.48 µA, corresponding to the addition or removal of

quantized currents in the gate-source loop. The fixed distribution of Isw
ch
during the
first step indicates that the readout process Isw
ch
did not alter 𝑛. Correspondingly, the
evenly-spaced and non-overlapping Isw
ch
measurements between trials indicate we were
able to read out individual, adjacent fluxon states (e.g. 𝑛, 𝑛 + 1, 𝑛 + 2). Since the
film had an inductance of ∼10 pH/, we calculated that our loop had an inductance
of approximately 200 pH/, meaning each fluxon contributed 10 µA. This fluxon
current value corroborates well with our measurements of other standalone yTrons of
similar dimensions, which have shown that for every 2-3 µA of gate current added,

105
Figure 6-10: Procedure to change the number of fluxons 𝑛 in the superconducting
gate-source loop. (a) The entire device starts out unbiased, completely superconduct-
ing. 𝑛 fluxons are stored in the gate-source loop. (b) An applied electrical current
from an external wire creates a hotspot in part of the gate-source loop, breaking the
superconductivity and allowing flux to enter or leave the loop randomly. (c) The
number of fluxons in the loop has changed from 𝑛 to 𝑚.

ch
Isw changes by 1 µA.

6.7 Outlook

In conclusion we have developed, characterized, and applied the yTron, a new su-
perconducting device which is able to sense superconducting currents inline without
perturbing them. We expect the yTron will find application as an inline current sen-
sor for devices such as transition edge sensors and superconducting nanowire single
photon detectors (SNSPD). In these devices, the yTron may be used to produce large
(∼10 mV) ancillary outputs for multiplexing or time resolution purposes, without
otherwise disturbing the primary readout method. Additionally, as a three-terminal
controllable weak link, the yTron can serve as a logic or memory element with which
to process and record detection events. Monolithic integration of the yTron with an
SNSPD is feasible based on our previous work with the nanocryotron (nTron) which
has similar fabrication to the yTron.
Since the yTron functionality comes from current crowding which occurs in every
superconductor, it should be possible to fabricate it from any superconducting mate-

106
Figure 6-11: Sequential trials of measurements of a quantized superconducting loop
using the yTron as an inline readout. Each dot corresponds to the median value of
100 measurements of the Ic of the yTron channel. The bars around each dot indicate
the standard deviation of the Ic measurements for that trial. Between each trial,
the loop was heated and cooled to allow fluxons to enter and leave. The step-like,
evenly-spaced division of Ic values indicate that the yTron was able to read out the
quantized current stored in the superconducting loop.

rial, although considerations must be made since the effective radius of curvature of
the intersection point may be mitigated by the coherence length of the material. Ad-
ditionally, the form of the yTron lends itself well to a CMOS-type fabrication scheme:
it could be fabricated vertically by using a oxide layer between two superconducting
thin films as the barrier between the gate and channel wires.

107
108
Chapter 7

Experimental techniques

In the process of testing the nTron, nanoSQUID, and yTron, a number of experimental
techniques were developed that were common to all three. These techniques ranged
from equipment automation using Python to building sample holders using a laser
cutter.

7.1 Measurement and automation

7.1.1 Critical current measurements

Since all three devices were made from superconducting nanowires, it was crucial to
have a consistent and well-characterized method of measuring the critical current Ic
of the devices. Due to the dynamics that take place during the superconducting to
resistive transition in a thin film, measurement of the switching current of a nanowire
depends greatly on the readout and bias circuitry.

Factors that affect Ic measurement

At its most fundamental level, the transition from a superconducting to resistive

state for a superconducting nanowire under a current bias is a stochastic process [97]
[98]. Even neglecting quantum fluctuations which may vary Ic with time, there will
always be Johnson-Nyquist noise [99] present on the nanowire. The presence of this

109
noise means that even with a fixed applied current, the actual current magnitude
in the nanowire is constantly shifting. Outside of these internal processes, there
are additional sources of noise such as thermally-radiated photons from the testing
chamber [100], as well as ambient electromagnetic noise which can induce currents
in the nanowire. These properties of the nanowire demand that Ic be measured as a
distribution, not as a single measurement.

Even allowing that Ic is a distribution, there is no canonical Ic measurement: in

fact, there are several measurement methods upon which the Ic value can be based,
and all will produce different distributions (e.g. shift the median value of the distribu-
tion). Take for example a long nanowire like an SNSPD which has a large amount of
kinetic inductance and a low shunt. When an SNSPD is current-biased, the nanowire
will stochastically output pulses once the current bias is large enough. Clearly in
the absence of light sources, the output pulses are the result of the wire being over-
biased, but it is not clear whether the lowest bias value that produces a single pulse
should be called Ic . Especially since this output pulse behavior is not universal to
all nanowires–in short nanowires with low inductance, these pulses are often so small
they cannot be read out by a standard amplifier setup.

Due to the properties of the NbN nanowire-based devices we often tested, we chose
to interpret Ic as the latching current measurement for a wire. For a highly resistive
material such as thin film niobium nitride, the latching current is very distinctive
and straightforward to measure, regardless of device length or inductance [52] [54].
The latching current is characterized by the lowest current value at which the voltage
across the device jumps from zero to a finite steady state value and does not return
to zero. This current value corresponds to the creation of a resistive hotspot in the
nanowire that is stable as long as the applied current bias is not reduced.

The one requirement for the latching measurement to be straightforward and

easy to measure is that the nanowire readout must be hysteretic. In this instance,
hysteresis means the nanowire will have a different I-V curve depending on whether
current is swept in the positive direct or negative direction. This discrepancy occurs
because a hotspot forms at high currents, and the presence of that hotspot can change

110
the point at which the nanowire transitions from resistive to superconducting.

If the nanowire is not hysteretic, instead of the readout voltage jumping sharply
from zero to a finite voltage, there will instead be a gradual increase in voltage.
The reasoning for this behavior is described in detail in Ref. [59], but for a brief
explanation we will give the following example. Consider a 100-nm-wide nanowire
with sheet resistivity Rs of 500 Ω/ that is being measured by an oscilloscope with
an input impedance of 1 Ω. A current source inputs a current ramp which starts at
zero and increases by 1 µA per second. At 𝑡 = 8 s, the current in the nanowire is
at 8 µA and the wire is still fully superconducting, meaning the voltage is still zero.
When the current reaches 9 µA though, assume the nanowire can no longer carry all
the current and enters the resistive state. However, this resistive state is actually
a hotspot which when created always has a minimum size of on the order of the
thermal diffusion length, which for NbN is ∼100 nm. At this point the nanowire has
a resistivity of approximately one square, or 500 Ω. Instead of flowing only through
the nanowire, the current will then split between the nanowire and the oscilloscope,
which as we specified has a resistance of 1 Ω. As a result of this splitting, the majority
of the current that was in the nanowire will be diverted to the oscilloscope input
resistor, leaving the nanowire with only a few nanoamperes still passing through the
resistive hotspot. As a result, the hotspot is able to cool, the superconducting state
is restored, and once again the current begins to divert from the oscilloscope input
impedance back into the nanowire, and the process repeats.

This oscillation is a well-known process called as relaxation-oscillation [3]. If we

observe the voltage on the scope in this regime, we will either see a series of pulses,
or if the pulse frequency is higher than the scope input bandwidth, the scope will
just read the averaged voltage of the pulses which will be slightly above zero volts.
If we increase the current bias to 10 µA at this point, this dynamic process will shift
slightly and spend more time in the hotspot state, and as a result the average voltage
the scope reports will be increased slightly. As the current bias increases, the process
of going into and out of the hotspot state will generate so much heat the wire will be
transition to being completely resistive. However, this will not be a sharp transition

111
in the voltage readout, making it difficult to determine at which current value this
occurs.
By the previous example, we can see how the nanowire switching process can
actually be an unstable, dynamic process that depends on the external readout circuit.
The load impedance the nanowire sees can affect the measurement process–a low
load impedance will cause the nanowire to be non-hysteretic, robbing us of the sharp
superconducting to resistive transition seen in hysteretic nanowires. Taking this a
step further, even when the nanowire is hysteretic changing the load impedance can
shunt the quantum and thermal noise in the nanowire, changing the Ic distribution.
Another factor that affects the Ic distribution is the rate at which the applied current
bias is ramped. Due to the stochastic nature of the switching, the higher the current
ramp rate (larger A/s) the larger the median of the Ic distribution will be. Thus, it
behooved us to find a consistent method that allowed for a straightforward, repeatable
implementation of Ic measurements.

The standardized Ic measurement

The testing setup used to measure the nTron, nanoSQUID, and yTron Ic can be con-
structed with common lab equipment. Specifically, what is needed are an oscilloscope,
an arbitrary waveform generator (or sine wave voltage source), lowpass filters, and
resistors in RF-shielded boxes. A typical setup is shown as a circuit schematic in
Fig. 7-1.
For our setup, we used a LeCroy 620Zi oscilloscope, which allowed for remote
control using a Python interface. This remote interface is detailed in Section 7.1.2.
This oscilloscope had an input bandwidth of 2 GHz, but since we were only interested
in voltage measurement timescales on the order of microseconds, we internally lim-
ited the bandwidth to 20 MHz in order to reduce high-frequency noise. To generate
the current ramp, we used a Agilent 33250a arbitrary waveform generator (AWG)
to generate a voltage ramp. As shown in Fig. 7-1, the AWG voltage output was
connected to the scope channel 1 to serve as a reference, and was also split off and
connected to a 10 kΩ resistor. The AWG additionally had a 1.9 MHz lowpass filter

112
Figure 7-1: Circuit diagram of the Ic sweeping setup. DUT stands for “Device Under
Test” and refers to the nanowire being measured. The arbitrary waveform generator
(AWG) was a Agilent 33250A, the 2 MHz lowpass filter was a Mini-Circuits BLP-
1.9+, and the 80 MHz lowpass filter was a high-rejection Mini-Circuits VLFX-80+.

in front of it, to block digital noise caused by the internal digital-to-analog converter
(DAC). The resistor was in series with the nanowire and converted the voltage signal
to a current signal, such that one volt at the AWG produced 100 µA through the
nanowire. Channel 2 of the oscilloscope was then connected between the resistor and
nanowire to serve as the device output voltage in a two-wire measurement setup.
Although a four-wire measurement would have been more exact, these measurements
typically produced tens of millivolts of output voltages. A two-wire measurement was
more than adequate due to the comparatively small voltage drops along the connec-
tions. Both oscilloscope channels had their input impedances set to DC 1 MΩ input
impedance, to guarantee the nanowire saw a high load impedance and would remain
hysteretic.

All of the connections were made using BNC or SMA coaxial cables, in order to
shield the device from high-frequency RF noise from the environment. Exposing a
small unshielded area (such as using a resistor not in an RF-tight box) made it possible
to see spurious radiation from cell phones and FM radio stations on the oscilloscope
readout. To further ensure the isolation of the device from high frequency noise,

113
lowpass filters were placed at all connections facing the device, as can be seen in
Fig. 7-1. Two types of lowpass filters were used in series: high-rejection lowpass
filters (Mini-Circuits VLFX-80+), which absorbed frequencies from 80 MHz up to
20 GHz, and lower-frequency lowpass filters (Mini-Circuits BLP-1.9+) which absorbed
frequencies down to 1.9 MHz but had poor absorption above 200 MHz.

With this setup in place, the Ic sweep measurement was ready to begin. The
AWG was set to produce a 200 Hz, 20 Vpp sine waveform, so that the device current
was swept between ±1 mA every few milliseconds. This timescale was chosen because
it often took several tens of microseconds for the voltage to reach steady-state once
the device had latched. This settling time occurred because at low frequencies the
coax cabling and lowpass filters looked like they had capacitances to ground of several
nanohenries. Because of the nanohenry-scale capacitance, the RC time constant for
transient effects was the series resistance time the capacitance, 10 kΩ × 1 nH = 10 µs.
Since we wanted the voltage readout to reflect the steady-state value as we swept the
current, we needed the transient timescale to be much shorter than the current ramp
timescale. It was also important to make the sine wave Vpp large enough such that the
measured Ic occurred during the approximately-linear portion of the sine wave voltage,
so that the ramp rate (which affects the Ic distribution) was approximately constant
during the superconducting to resistive transition. Typically, this meant adjusting
the AWG Vpp such that the current ramp swept to at least twice the measured Ic .

While the equipment for the Ic sweep was initially set up, it produced an output
on the oscilloscope that looked like Fig. 7-2. As shown in that figure, the oscilloscope
simultaneously reads out the AWG reference voltage as well as the nanowire volt-
age. Then by triggering on the rising edge of the device voltage (the jump from the
superconducting to resistive state), the reference voltage at that point was recorded
as a single Ic measurement. This process was typically repeated 100 to 1,000 times
to produce the Ic distribution. Since we were using a digital oscilloscope, however,
some factors had to be taken into account in order to get an accurate reading of the
reference voltage. These factors are detailed in the following section.

114
Figure 7-2: Oscilloscope voltage traces for an Ic sweeping measurement. Shown in
yellow is the sine-wave reference voltage produced by the AWG. In pink is the voltage
of the device under test (in this case an SNSPD).

Avoiding digital noise in the Ic measurement

Initially, the technique for extracting the Ic current from the measurement setup was
as follows: (1) trigger on the superconducting to resistive voltage jump; (2) at the
trigger/jump point, take the corresponding voltage datapoint from the AWG reference
voltage; (3) Divide this voltage datapoint by the series resistance to get Ic .
However, while taking thousands of Ic measurements, we found that there were
strange striations in the resulting histogram which did not seem to correspond to any
physical phenomenon. Normally, the Ic sweep distribution should look like a smooth
Gumbel distribution [101] of the form 𝑃 (𝑥) = exp (−(𝑥 + 𝑒−𝑥 )) . However, as shown
in Fig. 7-3 the general shape of the distribution was correct, but it was not smooth.
The distribution looked like it suffered from digital noise, but at the same time it was
not purely digital noise because the striations were not the result of truncated bits,
which would make most bins of the histogram zero.
It appeared the full distribution was actually comprised of periodic Gaussians of

115
Figure 7-3: Ic sweep distribution measurement, shown in terms of the raw voltage
(to convert to the Ic current divide by 10 kΩ). The Ic was swept 300,000 times
to build this histogram. (inset) Zoomed portion of the distribution showing the
unexpected striations (periodic Gaussian-like shapes) of the measured distribution.
These striations were due to the digital nature of the digital oscilloscope.

different heights. At first we believed this was the result of the AWG producing
a very digitized sine wave with discrete steps. This would have made sense beaus
the AWG is fundamentally just a high-speed digital-to-analog converter (DAC), so
possibly what we were seeing was that each sub-Gaussian corresponded the noisy
measurement of single-bit DAC step. But no matter what we did to filter the AWG
output – we tried many combinations of bandpass and lowpass filters – the striations
would not disappear from the measurement.

We then decided it could be a problem with the actual scope measurement. In our
initial implementation, the scope waited for the nanowire voltage jump, then it read
the datapoint of the AWG voltage at that time to determine Ic . But the problem
with this approach was that the voltage datapoints were digitally recorded. In the
example shown in Fig. 7-3, each sub-Gaussian is about 0.1 mV, which for the scope
measurement settings (20 mV/div) corresponds to about 11 bits of precision. That
is close to the edge of how well a very fast analog-to-digital converter (ADC) can
function, so any precision beyond those 11 bits is essentially just the scope guessing.

116
 With this discovery, it became clear that it was not a good idea to just pick one
datapoint from the AWG reference waveform and use only that datapoint for the
switching current. Instead, we slightly changed the way the scope read out Ic : when
the scope triggered on the nanowire, we then took the mean of several thousand
datapoints of the AWG reference voltage around the trigger time. Since the AWG
reference voltage was slowly-varying, these several thousand datapoints were essen-
tially the same voltage level, comprising only a few hundred nanoseconds worth of
data. We accomplished this mean function by zooming the scope to 200 ns of width,
and using the built-in scope mean function to generate a single value from the all the
datapoints in the (approximately flat) AWG voltage waveform.
By taking the average of several thousand datapoints, we were able to greatly
increase the accuracy of each individual Ic reading. After running another sample
test, we were able to get the smooth Gumbel distribution we initially expected. This
distribution is shown in Fig. 7-4 and represents the measurement technique used for
all Ic distribution measurements listed in this thesis.

Figure 7-4: Ic sweep distribution measurement with the improved measurement tech-
nique, eliminating the striations in the distribution (to convert to the Ic current divide
by 10 kΩ). The Ic was swept 58,000 times to build this histogram. (inset) Zoomed
portion of the distribution showing the corrected distribution which does not have
periodic behavior.

117
7.1.2 Measurement automation with Python

The ability to control experimental equipment remotely was crucial to configuring

repeatable measurements with as little setup time as possible. By writing functions
and scripts in Python, we were able to build standardized test configurations that
could be implemented in a few seconds, rather than tens of minutes of manual setup.
It also allowed us to ensure that the measurement setup was the same each time; as
long as the script configuring the equipment interface remained the same, the resulting
equipment configuration was the same. The Python code which was used to control
the equipment described in this thesis can additionally be found in Appendix A.

Remote control of equipment using GPIB

Many pieces of scientific equipment come with the ability to be remotely controlled.
Typically, communications with the instrument is accomplished using a General Pur-
pose Interface Bus (GPIB). The GPIB interface specification is also known as IEEE-
488 and dates back to the 1960s, but is often still implemented on modern equipment
due to its robust operation and long history as a standard equipment communica-
tions protocol. The physical GPIB setup typically consists of a single GPIB controller
(which plugs into a computer using USB) which is connected by GPIB-specific cables
to several pieces of equipment. These two components are shown in Fig. 7-5. Each
piece of equipment has an address associated with it (usually between 1 and 24), and
the computer can send messages or queries to a specific instrument by specifying its
address to the controller. Thus, the controller handles the signal-level communica-
tions protocol, and the computer just needs to provide (1) the GPIB address of the
instrument and (2) the message to send to the instrument. The messages sent by
the computer to the instruments are typically text strings. These strings can either
be queries, which generate a response, or commands which direct the instrument to
perform some function and do not generate a response. Sending a query is usually
a two step process: first, the computer issues a query command, and second, the
computer reads the text string response from the instrument.

118
 Figure 7-5: Photograph of a GPIB cable connector and a GPIB-USB interface.

Automating measurements using Python and GPIB

The majority of the measurements done in this thesis were performed using Python
scripts which controlled instruments through a Python-GPIB interface. The Python
software package used to talk to the GPIB was called PyVISA. PyVISA abstracts
simple command/response communications for several types of protocols (TCP/IP,
LXI, GPIB) away from the details of the protocol and lets the user communicate with
instruments using simple Python commands.
Instruments are created in Python by specifying a GPIB address to PyVISA.
PyVISA will then return an object which represents the instrument. The user can
then interact with the instrument by using the instrument-object functions read and
write.
A typical Python interaction with an instrument works like the following exam-
ple which describes reading the voltage on a Keithley 2001 multimeter. The user
first finds the GPIB address of the instrument (typically found in the instrument’s
configuration) and uses that address with PyVISA to create a Python object repre-
senting that instrument, for example my_multimeter. The user then looks up the
command for the desired action, in this case reading out a voltage. Commands
for a given piece of equipment are usually listed and described in a section of the
equipment manual called “Remote Programming”, “SCPI Command Reference”, or
similar. In this case, the Keithley 2001 multimeter command to read a voltage can

119
be found in the “IEEE-488 Reference” section of the manual, and is listed as the
command READ?. The user can send this command to the instrument using the
write command of the my_multimeter object, so that the full Python command
is my_multimeter.write(‘READ?’). The string ‘READ?’ will then be received and
processed by the multimeter, and since the string is a command that asks for a re-
sponse (usually indicated by a question-mark at the end of the command string), the
instrument will return a response string. If the Keithley multimeter had a voltage of
3.81 V attached to it, it would send the response string ‘3.81’. This response string
sits in an input buffer when it is received by the computer, so it is not assigned to
any Python variable until the read function is issued. To put the response string
into the Python variable the_voltage, the user issues the command the_voltage =
my_multimeter.read(). At this point, the string ‘3.81’ is contained in the variable
the_voltage and can be saved, converted to a floating point number, etc.

7.2 Sample holder design

The sample holders that were used to test the devices described in this thesis were
built out of low-cost materials and designed to allow for fast testing turnaround.

7.2.1 Sample holder construction

Although most superconducting experiments take place in a cryogenic vacuum envi-

ronment with careful attention paid to the rate of cooldown and vacuum properties,
the thin films we tested were found to be very resistant to moisture and extreme
thermal shock. We routinely cooled our superconducting nanostructures from room
temperature to 4.2 K in under a minute by submerging the samples directly into a
dewar of liquid helium, and observed no degradation in metrics of import such as Ic ,
Rs , or Tc . These samples were often removed from the liquid helium without special
precautions as well, and often there would be significant condensation left on the
devices as they warmed up–but again no pattern of degradation was observed.
This quick cooldown time allowed us to design, fabricate, and test superconducting

120
devices with a total turnaround time as low as 24 hours, but also required us to make
some special considerations when building the sample holder. For instance, due to the
submersion into liquid helium, it was important that the sample holders were built
from low-heat-capacity materials that would not boil off large amounts of expensive
liquid helium [102]. The sample holder also had to withstand the thermal shock from
the cooling process, and not break or detach due to thermal expansion. Additionally,
the holder needed to be sized such that it could fit down the neck of the dewar, and
be detachable from the submersion stick so that it can be brought to the wirebonder
for making electrical connections.

We found that an effective sample holder design consisted of an acrylic base, an

acrylic cover, a clip to mechanically hold down the (typically 1 × 1 cm) sample, a
PCB with which to make electrical connections, and a rod for dropping the sample
into the liquid helium. We typically cut the acrylic base and cover out of a large piece
of acrylic using a laser cutter. This allowed for rapid prototyping of sample holder
designs. Some finished examples of the sample holder are shown in Fig. 7-6 and .

The first step we took when constructing the sample holder was to design a to-
scale 2D representation of the holder in the open-source vector graphics software
package Inkscape. Our dewar mouth had a diameter of approximately 5 cm, and so
the holder base was typically designed to be long and narrow, about 3 to 4 cm wide
by 10 cm long. One example design is shown in Fig. 7-8. Since there were no large
mechanical forces being applied to our test setup, creating holes in the acrylic for
4-40 tapping was adequate for most mechanical connections. In the design, we chose
to use exclusively 4-40 screws in order to reduce the complexity of assembling and
disassembling the holder. Once the vector design for the base was completed, the
design was cut out from a sheet of acrylic 1/4 inch thick.

Once the sample holder base was complete, a custom PCB was designed and
fabricated (either by a PCB foundry or the process detailed in Section 7.2.2) to carry
electrical signals to the sample. This PCB typically had traces leading up to the
1 × 1 cm sample area. The aluminum wirebonds we used unfortunately were not
able to bond to the solder coating on these traces, likely because the solder was too

121
Figure 7-6: Photograph of three different custom-built sample holders. The two
sample holders on the left were constructed from PCBs that were fabricated in-house
using the etching process described in this thesis. The other sample holder (green
PCB) had a PCB which was designed in-house but purchased from a commercial
PCB company.

soft for the ultrasonic bonding process. Initially, we ground the solder off with a
dremel to expose the underlying copper trace, but often that copper would oxidize
after a few days and require re-polishing before wirebonds would successfully stick
again. We found that the best solution to make a robust wirebonding target was to
cut the ends off of a gold-coated pin header or other gold-coated part and solder it
down to the PCB trace. These parts typically had an electroplated gold finish several
tens of nanometers thick, which provided an ideal wirebonding target surface. The
result of soldering a gold-coated part (Digikey part number 1003-1626-ND) is shown
in Fig. 7-9. On the other end of the PCB trace was a microwave connector such
as SMA, SMP, or U.FL which attached to a coax and led to the room-temperature
electronics. Because we were submersing the whole holder into liquid helium, we had
a significant amount of cooling power and were able to use standard flexible copper
coaxial cables without worries of inadvertently heating the sample.

Samples were typically held in place by a beryllium copper (BeCu) clip. The

122
 Figure 7-7: Finished sample holder with cover on top.

Figure 7-8: Vector drawing for an acrylic sample holder drawn in Inkscape. All of
the solid lines represent cuts performed by the laser cutter. The smallest holes are
holes meant for 4-40 tapping

clip material was chosen because BeCu is one of the few materials which retains
its spring-like characteristics even at cryogenic temperatures–many other materials
become hard and brittle [102]. Custom-shaped BeCu was expensive, and sheets of
BeCu were difficult to form, but we found a solution by ordering BeCu “finger stock”.
This type of item is typically sold in long strips and is used for shielding purposes,
but we repurposed it by cutting short sections and using their spring-like qualities to
pin down the sample. Two examples of these clip can be seen in Fig. 7-6.
Once the sample holder components were complete, assembly for testing was ready.
The sample would be loaded into the BeCu clip, electrical connections would be made
by wirebond from the PCB traces to the devices on the sample, and the cover was

123
Figure 7-9: Photograph of gold-coated components which have been cut into pieces
and soldered to the PCB to act as a wirebonding targets.

screwed down on top of the assembly to prevent inadvertent contact with the sample
or wirebonds. Altogether, the assembly was very robust and only infrequently were
there mechanical problems such as detaching wirebonds. The PCB would then have
coaxial cables attached to it, and the whole assembly would be mounted on the end
of a thin stainless steel tube for submersion.

7.2.2 In-house PCB fabrication

For many of the experiments detailed in this these, PCBs were commonly used as
sample holders since they allowed for easy integration of lumped element circuit com-
ponents, and were bondable with the wirebonding processes needed to connect to
the gold pads of the samples. Unfortunately, many PCB foundries can take days or
sometimes weeks to fabricate and ship a simple PCB. This is confounded further by
the requirement to learn EAGLE or an equivalent circuit-layout program to produce
the Gerber files necessary for the foundries. In many cases the sample holder PCBs
were simply a few traces on one side of the PCB, and did not require the advanced
routing techniques available in the circuit-layout program. Instead, these traces can
be drawn in a basic graphics design program such as Inkscape.

124
 We discovered that the Epilog Mini laser cutter we used to cut acrylic could be
used as a raster patterning tool for patterning simple PCBs. Were were successfully
able to fabricate PCBs with features sizes of less than 0.5 mm using this technique.
This resolution could likely be improved with further optimization. Here we describe
the PCB fabrication technique from start to finish. First, the supplies that are needed
are:

∙ Laser cutter

∙ Copper-clad PCB board

∙ Spraypaint (Used Krylon “Eco-Guard” Latex Enamel spraypaint, black, worked

well because it was fast-drying and low odor.)

∙ Isopropyl alcohol

∙ Ferric chloride copper etchant

As shown in Fig. 7-10 first coat the copper side of the PCB in a thin layer of
spraypaint. After coating, allow it to dry – putting it on a hotplate at ∼60 ∘C will
speed this up greatly. It may be desirable to put a piece of foil over it while drying
to prevent dust from landing and sticking to the surface. Put a second coat on and
allow that to dry as well. It should be dry to the touch, but leaving it out longer (or
baking) is recommended to get as much of the spraypaint solvent out as possible. If
the PCB is copper-clad on both sides, and it is necessary to keep the backside copper
as a ground-plane, cover the backside with two coats of spraypaint as well.
The next step is to design the pattern in Inkscape or another favorite graphic
design program. Save the design as a PDF for laser cutter. This format works well
because the laser cutter acts as a printer and the most compatible format for raster
printing is PDF. In this example, the design shown in Fig. 7-12 will be patterned onto
the copper. In this process, the spraypaint acts as a etch resist and the laser cutter
ablates the spraypaint off where the design is black. As a result, areas of the design
that are black will have their copper exposed and subsequently etched. Areas that
are white on the design will be leftover as bare copper.

125
Figure 7-10: (left) Bare copper PCB board used as the blank substrate for PCB
patterning. (right) The same copper PCB board, covered with two coats of black
spraypaint.

The next step is to send the design file PDF to the laser cutter printer. The
settings used for the Epilog Mini laser cutter are:

∙ Raster mode

∙ 400 dpi (this can be increased, but not usefully)

∙ Power 100 %

∙ Speed 40 %

As laser cutter is printing (shown in Fig. 7-11), it should be possible to see the
pattern take form. Once it’s done, the pattern should be fully printed but there will
be burned spraypaint left over in the exposed areas, which is the result of the laser
ablation not fully removing all of the spraypaint. This leftover spraypaint needs to
be removed that so it does not affect the etching. This removal process is done but
this by using a Q-tip that’s very slightly damp with isopropyl alcohol. Run the q-tip

126
Figure 7-11: Laser cutter in the process of patterning the spraypaint on the surface
of the PCB board.

along the various traces and areas until the copper start to shine. We had to be
careful about pushing too hard or using too much isopropyl alcohol; both the leftover
and untouched spraypaint are soluble and pushing too hard may wipe away the finer
details of the pattern. This difference between the cleaned and uncleaned sections is
shown in Fig. 7-13. As a result, left with a very high-contrast pattern, shiny copper
and black paint.

Figure 7-12: PCB design made in Inkscape. Areas which are black will be “printed”
by the laser cutter, exposing the bare copper and allowing those areas to be etched.
Areas in white will be copper in the finished PCB.

The next step is to cut the patterns out of the PCB and drop them into a solution

127
Figure 7-13: The laser-patterned PCB, parts of which have been cleaned with a Q-tip
soaked in isopropyl alcohol. Before etching, the entire pattern should be as shiny as
the original bare copper was before spraypaint application.

of ferric chloride (FeCl). Warm the FeCl solution on a hotplate set to 80 ∘C. We
purchased the FeCl from a local electronics store, and it took about 15 min to start
seeing the etch, and about 30 min to fully etch the copper. The PCB etch progress
needs to be checked often, the timing can vary significantly even between identical
PCBs due to some of the spraypaint slowly dissolving and clogging up the solution.
The final step is to wash away the spraypaint. The paint we used was very soluble
in acetone, so a little bit of scrubbing with a clean-wipe and acetone removed it
completely. The final result is shown in Fig. 7-14.

128
Figure 7-14: The final product from the in-house PCB fabrication process using the
laser cutter.

129
130
Chapter 8

Conclusion and outlook

Superconducting devices are presently being used in a diverse set of applications

in an equally diverse set of fields. Broadly speaking, devices which are made from
superconductors fall into two categories: phase-based devices and magnitude-based
devices. Several magnitude-based devices are currently employed as sensors (e.g.
SNSPDs, MKIDs, TESs), and the phase-based Josephson junction has been successful
as a electronics element (e.g. RSFQ).
However, in the realm of superconducting devices there is still much left to do.
For instance, although the Josephson junction is both fast and low-power, it cannot
be easily interfaced with external technologies which require high input voltages (e.g.
CMOS). And although devices like the SNSPD and TES are excellent sensors, they
are often amplified by noisy, non-superconducting sources.
To help fill these gaps in functionality, this thesis described the development of
several new devices made by nano-patterning superconducting thin films into 2D
geometries. Superconducting thin films form an ideal platform on which to take
advantage of both the phase and magnitude in a superconductor: in a thin film the
magnitude of the superconducting state can be weakened locally (e.g. the nTron),
and the phase can be manipulated directly by injected current flow (e.g. the current-
biased nanoSQUID).
Although this thesis has described a small handful of implementations for the
nTron, yTron, and current-biased nanoSQUID, there are also several applications

131
which should be possible for the near future. For instance, the nTron has operating
characteristics which are complementary to Josephson-junction-based technologies,
and integration could enable those technologies to perform otherwise-difficult tasks
such as memory line-driving and fanout in exascale computing. Additionally, the
ease of fabrication, transistor-like logic-gate design, and extensibility to other mate-
rials mean that small-scale superconducting circuits are now accessible with minimal
fabrication investment and straightforward circuit designs.
The current-biased nanoSQUID may find use as a convenient metrological tool to
measure kinetic inductances in thin films, since accurately measuring the inductance
of the nanoSQUID only requires that a few DC wires. Additionally, it may find use
as a feedback tool for scanning SQUID microscopy.
One of the most potentially interesting uses of the yTron would be to use it as a
three-terminal controllable weak link for single-flux quantum logic. By forming loops
between the gate and drain and source and drain, flux flow into the source-drain loop
can be controlled by flux stored in the gate-drain loop. In future work, the yTron may
be used inline with transition-edge sensors or superconducting nanowire single-photon
detectors to provide extra, high-impedance outputs.

132
Appendix A

Python equipment automation code

This appendix contains Python code for the control and automation of several dif-
ferent scientific instruments, which was critically enabling to the measurement of the
devices described this thesis. Each instrument is given its own Python class, based
on the PyVISA 1.3 instrumentation framework.

133
LeCroy 620Zi Oscilloscope

from pyvisa import visa

import numpy as np
import time
import datetime

class LeCroy620Zi(visa.Instrument):
"""Python class for LeCroy Oscilloscope, written by Adam McCaughan. Most of
these commands
originate from the Automation Command Reference Manual for WaveRunner
Oscilloscopes"""

def __init__(self, args):

super( visa.Instrument, self).__init__(args)
self.write(’COMM_HEADER OFF’) # Get rid of the leading ’VBS ’ crap
self.write(’COMM_FORMAT DEF9,WORD,BIN’) # Set output to 16 bits of
information (a ’word’) per datapoint

def round_up_lockstep(self, x):

""" Some functions on the LeCroy require numbers to be rounded up to
nearest 1,2 or 5
e.g. 1.2e-6 -> 5e-6 and 4.7e0 -> 5e0 """
x_str = ’%0.9e’ % x # Takes 1234.24e-10 -> ’1.234240000e-07’
x_digits = x_str[:-4] # Takes ’1.234240000e-07’ -> ’1.234240000’
x_exp = x_str[-4:]
if float(x_digits) <= 1: locked_x = ’1’
elif float(x_digits) <= 2: locked_x = ’2’
elif float(x_digits) <= 5: locked_x = ’5’
else: locked_x = ’10’
return float(locked_x + x_exp)

134
def vbs_ask(self,message):
vbs_msg = ’VBS? \’return = %s\’’ % message
# print ’Sending command: ’ + vbs_msg
return self.ask(vbs_msg)

def vbs_write(self,message):
vbs_msg = ’VBS \’%s\’’ % message
# print ’Sending command: ’ + vbs_msg
self.write(vbs_msg)

def reset(self):
self.write(’*RST’)
self.write(’COMM_HEADER OFF’) # Get rid of the leading ’VBS ’ crap
self.write(’COMM_FORMAT DEF9,WORD,BIN’) # Set output to 16 bits of
information (a word) per sample
time.sleep(1)

def clear_sweeps(self):
self.vbs_write(’app.ClearSweeps’) #
time.sleep(0.2) # Necessary to allow the scope time to reset all values

def view_channel(self, channel = ’C1’, view = True):

if channel[0] == ’C’: # If it’s C1, C3, etc
self.vbs_write(’app.Acquisition.%s.View = %s’ % (channel, view))
elif channel[0] == ’F’: # If it’s F1, F2...
self.vbs_write(’app.Math.%s.View = %s’ % (channel, view))

def set_coupling(self, channel = ’C1’, coupling = ’DC1M’):

""" Coupling should be either AC1M, DC1M, DC50, or Gnd """
self.vbs_write(’app.Acquisition.%s.Coupling = "%s"’ % (channel, coupling))

135
def get_coupling(self, channel = ’C1’):
""" Coupling should be either AC1M, DC1M, DC50, or Gnd """
return self.vbs_ask(’app.Acquisition.%s.Coupling’ % (channel))

def set_bandwidth(self, channel = ’C1’, bandwidth = ’Full’):

""" Bandwidth should be either 1GHz, 200MHz, 20MHz, 3GHz, 4GHz, Full """
self.vbs_write(’app.Acquisition.%s.BandwidthLimit = "%s"’ % (channel,
bandwidth))

def set_vertical_scale(self, channel = ’C1’, volts_per_div = 1, volt_offset =

0):
# Lecroy only allows digits 1, 2, and 5. e.g. 5e-6 is acceptable, 4e-6 is
not
volts_per_div = self.round_up_lockstep(volts_per_div)
self.vbs_write(’app.Acquisition.%s.VerScale = %0.0e’ % (channel,
volts_per_div))
self.vbs_write(’app.Acquisition.%s.VerOffset = %0.0e’ % (channel,
volt_offset))

def find_vertical_scale(self, channel = ’C1’):

self.vbs_write(’app.Acquisition.%s.FindScale’ % channel)

def set_horizontal_scale(self, time_per_div = 1e-6, time_offset = 0):

self.vbs_write(’app.Acquisition.Horizontal.HorScale = %0.6e’ %
time_per_div)
self.vbs_write(’app.Acquisition.Horizontal.HorOffset = %0.6e’ %
time_offset)

def set_memory_samples(self, num_samples = 1e6):

self.vbs_write(’app.Acquisition.Horizontal.MaxSamples = %0.3e’ %
num_samples)

136
def set_trigger(self, source = ’C1’, volt_level = 0.1, slope = ’positive’):
""" Slope should be "Either" / "Negative" / "Positive" """
self.vbs_write(’app.Acquisition.Trigger.Source = "%s"’ % source)
self.vbs_write(’app.Acquisition.Trigger.%s.Level = %0.4e’ % (source,
volt_level))
self.vbs_write(’app.Acquisition.Trigger.%s.Slope = "%s"’ % (source, slope)
)

def set_trigger_mode(self, trigger_mode = ’Normal’):

""" trigger_mode should be set to Auto/Normal/Single/Stop """
self.vbs_write(’app.Acquisition.TriggerMode = "%s"’ % trigger_mode)

def set_persistence(self, channel = ’C1’, persistence = False, monochrome =

False):
self.vbs_write(’app.Display.LockPersistence = "PerTrace"’)
self.vbs_write(’app.Acquisition.%s.Persisted = %s’ % (channel, persistence
))
self.vbs_write(’app.Acquisition.%s.PersistenceMonochrome = %s’ % (channel,
monochrome))

def label_channel(self, channel = ’C1’, label = ’Channel 1 label text’):

if (label == ’’) or (label == False) or (label == None):
self.vbs_write(’app.Acquisition.%s.ViewLabels = False’ % channel)
else:
self.vbs_write(’app.Acquisition.%s.LabelsText = "%s"’ % (channel,
label))
self.vbs_write(’app.Acquisition.%s.ViewLabels = True’ % channel)

def set_display_gridmode(self, gridmode = ’Auto’):

""" gridmode should be Auto / Dual / Octal / Quad / Single / XY / XYDual /

137
 XYSingle """
self.vbs_write(’app.Display.GridMode = "%s"’ % gridmode)

def set_parameter(self, parameter = ’P1’, param_engine = ’Maximum’, source1 =

’C1’, source2 = None, show_table=True):
""" Possible param_engine values listed in a table on page 1-151 of the
automation manual.
Some sample param_engine values are:
Frequency / LevelAtX / Fall / Maximum / Mean / Median / Minimum /
PeakToPeak """
self.vbs_write(’app.Measure.ShowMeasure = %s’ % show_table)
self.vbs_write(’app.Measure.%s.ParamEngine = "%s"’ % (parameter,
param_engine))
if source1 is not None:
self.vbs_write(’app.Measure.%s.Source1 = "%s"’ % (parameter, source1))
if source2 is not None:
self.vbs_write(’app.Measure.%s.Source2 = "%s"’ % (parameter, source2))
self.vbs_write(’app.Measure.%s.View = True’ % parameter)

def set_math(self, math_channel = ’F1’, operator = ’AbsoluteValue’, source1 =

’C1’, source2 = None):
""" Possible operator values listed in a table on page 1-151 of the
automation manual.
Sample values include: Average / Trend / Histogram / FFT / Integral / etc
"""
self.vbs_write(’app.Math.%s.Operator1 = "%s"’ % (math_channel, operator))
if source1 is not None:
self.vbs_write(’app.Math.%s.Source1 = "%s"’ % (math_channel, source1))
if source2 is not None:
self.vbs_write(’app.Math.%s.Source1 = "%s"’ % (math_channel, source2))

def get_parameter_value(self, parameter = ’P1’):

138
 return float(self.vbs_ask(’app.Measure.%s.Out.Result.Value’ % parameter))

def get_trigger_mode(self):
return self.vbs_ask(’app.Acquisition.TriggerMode’)

def get_wf_data(self,channel=’C1’): # e.g. channel = C1 or F3 etc

ADDR_VGAIN = 156
ADDR_VOFFSET = 160
ADDR_HINTERVAL = 176
ADDR_HOFFSET = 180
databytes = self.ask(channel + ’:WAVEFORM? DAT1’) # Contains waveform data
databytes = databytes[16:] # Strips leading header data
if len(databytes) == 0:
return np.array([]), np.array([])
if len(databytes) % 2 == 1: # Sometimes accidentally returns an extra byte
databytes = databytes[:-1]
data = np.frombuffer(databytes, np.int16)
# DESC contains info about vertical/horiz scaling, etc,
# see P280, Appendix II Remote Control Manual for byte addresses
desc = self.ask(channel + ’:WAVEFORM? DESC’)
desc = desc[16:]
vgain = float(np.frombuffer(desc[ADDR_VGAIN:ADDR_VGAIN+4], np.float32))
voffset = float(np.frombuffer(desc[ADDR_VOFFSET:ADDR_VOFFSET+4], np.
float32))
hinterval = float(np.frombuffer(desc[ADDR_HINTERVAL:ADDR_HINTERVAL+4], np.
float32))
hoffset = float(np.frombuffer(desc[ADDR_HOFFSET:ADDR_HOFFSET+8], np.double
))
num_samples = len(data)
x = np.array(range(num_samples))*hinterval + hoffset
# yscale = float(2**16)/(vf_stop - vf_start) # Scale the data, it’s
output as ints from -2^16 to 2^16
y = data*vgain - voffset

139
 return x,y

def get_single_trace(self, channel = ’C1’):

""" Sets scope to "single" trigger mode to acquire one trace, then waits
until the trigger has happened
(indicated by the trigger mode changing to "Stopped"). Returns blank
lists if no trigger occurs within 1 second """
n = 0; x = np.array([]); y = np.array([])
self.set_trigger_mode(trigger_mode = ’Single’)
while self.get_trigger_mode() == ’Single’ or n > 1e10:
time.sleep(1e-4)
n = n+1
x,y = self.get_wf_data(channel=channel)
return x,y

def get_num_sweeps(self,channel=’F1’): # For use with histograms, trends, etc

return int(self.vbs_ask(’app.Math.%s.Out.Result.Sweeps’ % channel))

def setup_math_trend(self, math_channel = ’F1’, source = ’P1’, num_values = 10

e3):
self.set_math(math_channel = math_channel, operator = ’Trend’, source1 =
source)
self.vbs_write(’app.Math.%s.Operator1Setup.Values = %s’ % (math_channel,
num_values))
self.view_channel(channel = math_channel, view = True)

def setup_math_wf_average(self, math_channel = ’F1’, source = ’C1’, num_sweeps

= 100):
self.set_math(math_channel = math_channel, operator = ’Average’, source1 =

140
 source)
self.vbs_write(’app.Math.%s.Operator1Setup.Sweeps = %s’ % (math_channel,
num_sweeps))
self.view_channel(channel = math_channel, view = True)

def setup_math_histogram(self, math_channel = ’F1’, source = ’P1’, num_values

= 10e3,
num_bins = 100, center = 0, width_per_div = 1,
auto_scale = True):
self.set_math(math_channel = math_channel, operator = ’Histogram’, source1
= source)
self.vbs_write(’app.Math.%s.Operator1Setup.Values = %s’ % (math_channel,
num_values))
self.vbs_write(’app.Math.%s.Operator1Setup.AutoFindScale = %s’ % (
math_channel, auto_scale))
self.vbs_write(’app.Math.%s.Operator1Setup.Bins = %s’ % (math_channel,
num_bins))
self.vbs_write(’app.Math.%s.Operator1Setup.Center = %s’ % (math_channel,
center))
width_per_div = self.round_up_lockstep(width_per_div)
self.vbs_write(’app.Math.%s.Operator1Setup.HorScale = %s’ % (math_channel,
width_per_div))
self.view_channel(channel = math_channel, view = True)

def collect_sweeps(self, channel = ’F1’, num_sweeps = 1000):

self.clear_sweeps()
time.sleep(0.1)
while (self.get_num_sweeps(channel = channel) < num_sweeps+1):
time.sleep(0.1)
x, ic_values = self.get_wf_data(channel=channel)
while len(ic_values) < num_sweeps:
time.sleep(0.05)
x, ic_values = self.get_wf_data(channel=channel)
return ic_values[:num_sweeps] # will occasionally return 1-2 more than

141
 num_sweeps

def save_screenshot(self, file_path = None, white_background = True):

if file_path == None:
time_str = datetime.datetime.now().strftime(’%Y-%m-%d %H-%M-%S’)
file_path = time_str + ’ lecroy screenshot.png’

# lecroy.ask(’HCSU?’) # Asks what the current hard copy setup is (useful

for reference)
if white_background: self.write(’HCSU BCKG,WHITE’) # Sets background to
white
if white_background is False: self.write(’HCSU BCKG,BLACK’) # Sets
background to black
self.write(’HCSU DEV,PNG’)
self.write(’HCSU FORMAT,LANDSCAPE’)
self.write(’HCSU DEST,REMOTE’)
self.write(’HCSU AREA,DSOWINDOW’)
self.write(’SCREEN_DUMP’) # Takes the screenshot
newFileBytes = self.read()

with open(file_path, "wb") as newFile:

newFileByteArray = bytearray(newFileBytes)
newFile.write(newFileByteArray)

return file_path

#### Sample script for PyVisa 1.3

# lecroy_ip = ’192.168.1.100’
# lecroy = LeCroy620Zi("TCPIP::%s::INSTR" % lecroy_ip)
# x,y = lecroy.get_wf_data(’F1’)
# from matplotlib import pyplot as plt
# plt.plot(x,y)

142
# plt.show()

143
Agilent 8153A tunable laser power meter

from pyvisa import visa

from time import sleep
import numpy as np

class Agilent8153A(visa.Instrument):
"""Python class for Agilent 8153A power meter, written by Adam McCaughan"""
def __init__(self, args):
super( visa.Instrument, self).__init__(args)
# Anything else here that needs to happen on initialization
self.timeout = 1 # Set timeout to 1 second

def reset(self):
self.write(’*RST’)

def setup_basic(self, lambda_nm = 1550, averaging_time = 0.1):

self.write(’*RST’)
self.write(’INIT1:CONT 1’)
self.set_averaging_time(averaging_time = averaging_time) # Sets averaging
time, 20ms < value < 3600s
self.set_wavelength(lambda_nm = lambda_nm)
# self.write(’OUTP 1’)

def set_averaging_time(self, averaging_time = 0.1):

self.write(’SENS1:POW:ATIME %0.3e’ % averaging_time) # Sets averaging time
, 20ms < value < 3600s

def read_power(self):
power = float(self.ask(’READ1:POW?’)) # Returns power in watts
return power

144
 def set_wavelength(self, lambda_nm):
self.write(’SENS1:POW:WAVE %0.6e’ % (lambda_nm*1e-9))

# pm = Agilent8153A(’GPIB0::22’)

145
Agilent 33250A 80 MHz Arbitrary Waveform Generator

from pyvisa import visa

import numpy as np

class Agilent33250a(visa.GpibInstrument):
"""Python class for Agilent 33250a 80MHz Frequency Generator, written by Adam
McCaughan"""
def reset(self):
self.write(’*RST’)

def set_sin(self, freq=1000, vpp=0.1, voffset=0):

# In a string, %0.6e converts a number to scientific notation like
# print ’%.6e’ %(1234.56789) outputs ’1.234568e+03’
self.write(’APPL:SIN %0.6e HZ, %0.6e VPP, %0.6e V’ % (freq,vpp,voffset))

def set_pulse(self, freq=1000, vlow=0.0, vhigh=1.0, width = 100e-6, edge_time

= 1e-6):
vpp = vhigh-vlow
voffset = vpp/2
self.write(’APPL:PULS %0.6e HZ, %0.6e VPP, %0.6e V’ % (freq,vpp,voffset))
self.write(’PULS:WIDT %0.6e’ % (width))
self.write(’PULS:TRAN %0.6e’ % (edge_time))

def set_freq(self, freq=1000):

self.write(’FREQ %0.6e’ % (freq))

def set_vpp(self, vpp=0.1):

self.write(’VOLT %0.6e’ % (vpp))

def set_voffset(self, voffset = 0.0):

self.write(’VOLT:OFFS %0.6e’ % (voffset))

146
def set_vhighlow(self, vlow=0.0, vhigh=1.0):
if vhigh > vlow:
self.set_vpp(vhigh-vlow)
self.set_voffset((vhigh+vlow)/2.0)
self.set_polarity(inverted = False)
elif vhigh < vlow:
self.set_vpp(vlow-vhigh)
self.set_voffset((vhigh+vlow)/2.0)
self.set_polarity(inverted = True)

def set_output(self,output=False):
if output is True: self.write(’OUTPUT ON’)
else: self.write(’OUTPUT OFF’)

def set_load(self, high_z=False):

if high_z is True: self.write(’OUTP:LOAD INF’)
else: self.write(’OUTP:LOAD 50’)

def set_polarity(self, inverted = False):

if inverted is True: self.write(’OUTP:POL INV’)
else: self.write(’OUTP:POL NORM’)

def set_trigger(self, external_trigger = False, delay = 0.0):

if external_trigger: self.write(’TRIG:SOUR EXT’ )
else: self.write(’TRIG:SOUR IMM’ )
self.write(’TRIG:DEL %s’ % (delay)) # Delay in seconds

def trigger_now(self):
self.write(’*TRG’)

def set_burst_mode(self, burst_enable = True, num_cycles = 1, phase = 0):

if burst_enable:
self.write(’BURS:STAT ON’) # Enables burst state
self.write(’BURS:NCYC %s’ % (num_cycles))

147
 self.write(’BURS:PHAS %s’ % (phase)) # Phase in degrees

else:
self.write(’BURS:STAT OFF’) # Disables burst state

def set_arb_wf(self, t = [0.0, 1e-3], v = [0.0,1.0], name = ’ARB_PY’):

""" Input voltage values will be scaled to +/-1.0, you can then adjust the
overall
amplitude using the set_vpp function. The 33250a does not allow the input
of time for each
point, so we instead use interpolation here to create waveform of 2^14
equally-spaced
points, after which you can use set_freq to get the desired freq"""

t = np.array(t); v = np.array(v)

v = v-min(v); v = 2*v/max(v); v = v-1

temp = self.timeout; self.timeout = 60
t_interp = np.linspace(t[0],t[-1],2**14) # Can be up to 2**14 long
v_interp = np.interp(t_interp, t, v)

data_strings = [’%0.3f’ % x for x in v_interp]

data_msg = ’, ’.join(data_strings)

self.write(’DATA VOLATILE, ’ + data_msg) # Form of "DATA VOLATILE, 1, .67,

.33, 0, -.33", p200 user’s guide
name = name[0:8].upper()
self.write(’DATA:COPY %s, VOLATILE’ % name)
self.write(’APPL:USER’) # Set output to ARB
self.write(’FUNC:USER %s’ % name) # Select the waveform in the volatile
memory
self.write(’APPL:USER’)
self.timeout = temp
# self.write(’FUNC USER’) # Output the selected waveform

148
def setup_heartbeat_wf(self):
heartbeat_t = [0.0, 4.0/8, 5.0/8, 6.0/8, 7.0/8, 8.0/8]
heartbeat_v = [0.0, 0.0, 1.0, 0.0, -1.0, 0.0]
freq_gen.set_arb_wf(t = heartbeat_t, v = heartbeat_v, name = ’HEARTBEA’)

def select_arb_wf(self, name = ’HEARTBEA’):

name = name[0:8].upper()
self.write(’APPL:USER’) # Set output to ARB
self.write(’FUNC:USER %s’ % name)
self.write(’APPL:USER’) # Set output to ARB

149
Agilent 53131A Universal Counter

from pyvisa import visa

import numpy as np
from time import sleep
from matplotlib import pyplot as plt

class Agilent53131a(visa.GpibInstrument):
"""Python class for Agilent 53131a counter, written by Adam McCaughan
Use like c = Agilent53131a(’GPIB0::3’)"""

def reset(self):
self.write(’*RST’)

def basic_setup(self):
self.write(’*RST’)
self.write(’*CLS’)

self.write(’:EVEN:LEV:AUTO OFF’) # Turn off auto trigger level

self.write(’:EVEN:LEV -0.200V’) # Set trigger level
self.write(’:EVEN:SLOP NEG’) # Or POS. Trigger on negative slope
self.write(’:EVEN:HYST:REL 0’) # Set hysteresis (?)
self.write(’:INP:COUP AC’) # Or DC. Input coupling
self.write(’:INP:IMP 50’) # Set input impedance to 50ohms
self.write(’:INP:FILT OFF’) # Turn off 100kHz lowpass filter
self.write(’:FUNC "TOT 1"’) # Totalize on channel 1
self.write(’:TOT:ARM:STAR:SOUR IMM’) # Set start source to immediate (run
on command)
self.write(’:TOT:ARM:STOP:SOUR TIM’) # Set stop source to time (wait
certain time)
self.write(’:TOT:ARM:STOP:TIM 0.1’) # Set stop time to 100 ms
self.write(’:INP:ATT 1’) # Or 10. Set attenuation factor

150
def set_trigger(self, trigger_voltage = -0.075, trigger_slope = None):
if trigger_slope is ’POS’ or trigger_slope is ’NEG’:
self.write(’:EVEN:SLOP %s’ % trigger_slope) # Or POS. Trigger on
negative slope
self.write(’:EVEN:LEV %0.3fV’ % trigger_voltage) # Set trigger level

def count_rate(self, counting_time = 0.1):

self.write(’:TOT:ARM:STOP:TIM %0.3f’ % counting_time) # Set stop time to #
of seconds
dcr = self.ask(’:READ?’)
dcr = float(dcr)/counting_time
# time.sleep(counting_time + 0.1)
return dcr

def counts_vs_time(self, trigger_voltage= -0.075, counting_time=0.1,

total_time=2):
self.set_trigger(trigger_voltage)
num_tests = int(total_time/counting_time)
dcr = []
t = []
start_time = time.time()
for n in range(num_tests):
dcr.append(self.get_dcr(counting_time))
t.append(time.time() - start_time)

return t, dcr

def scan_trigger_voltage(self, voltage_range=[-0.2,0.2], counting_time=0.1,

num_pts=40):
v = np.linspace(voltage_range[0],voltage_range[1],num_pts)
dcr = []
for trigger_voltage in v:
self.set_trigger(trigger_voltage)

151
 dcr.append(self.count_rate(counting_time))
print ’Trigger voltage = %0.3f / Count rate %0.1f’ % (
trigger_voltage, dcr[-1])
return v, np.array(dcr)/float(counting_time)

152
Anritsu MG9638A Tunable Laser Source

from pyvisa import visa

class AnritsuMG9638A(visa.Instrument):
"""Python class for Antritsu M9638 tunable laser source, written by Adam
McCaughan. Adapted from
Mihir’s MATLAB code"""
def __init__(self, args):
super( visa.Instrument, self).__init__(args)
# Anything else here that needs to happen on initialization
self.timeout = 1 # Set timeout to 1 second

def reset(self):
self.write(’*RST’)

def setup_basic(self):
self.write(’MCW’) # Set laser to CW mode
self.write(’COH ON’) # Set laser to coherent mode
self.set_power_unit(’mW’)
self.set_wavelength(1550)

def set_power_unit(self, power_unit):

""" power_unit should be either dBm, mW, or uW """
self.write(’POWU %s’ % power_unit)

def get_power_unit(self, power_unit):

return self.ask(’POWU?’)

def get_wavelength(self):
wavelength_str = self.ask(’OUTW?’)
return float(wavelength_str)

153
def set_output(self, output = False):
self.write(’OUTP %d’ % output)

def get_output(self):
return bool(self.ask(’OUTP?’))

def set_power(self, power, power_unit = ’uW’):

""" power_unit should be either dBm, mW, or uW """
power_str = self.write(’POW %0.3d%s’ % (power, power_unit))

def get_power(self):
power_str = self.ask(’POW?’)
return float(power_str)

def set_wavelength(self, lambda_nm):

self.write(’WCNT %0.3fNM’ % lambda_nm)

154
HP 8722C Network Analyzer

from pyvisa import visa

from time import sleep
import numpy as np

class HP8722C(visa.GpibInstrument):
"""Python class for HP 8722C Network Analyzer, written by Adam McCaughan"""
def reset(self):
self.write(’*RST’)
self.timeout = 5

def freq_range(self, f_start = 0.1e9, f_stop = 1.0e9, f_center = None, f_span

= None, num_pts = 401):

if (f_center is not None) and (f_span is not None):

f_start = f_center - f_span/2.0
f_stop = f_center + f_span/2.0

self.write(’POIN %0.0i;’ % num_pts)

self.write(’STAR %0.6e;’ % f_start)
self.write(’STOP %0.6e;’ % f_stop)

def fixed_freq(self, f = 10e6):

self.write(’CWFREQ%0.6eHZ;’ % f)

def power(self, power = -20):

if (power < -60) or (power > 0):
print ’Out of range value’; return
if power >= -5: power_range = ’01’ # Only available below 26 GHz
elif power >= -20: power_range = ’02’
elif power >= -35: power_range = ’05’
elif power >= -50: power_range = ’08’

155
 elif power >= -60: power_range = ’10’

self.write(’PRAN%s’ % power_range)
self.write(’POWE %0.0d’ % power) # Sets power (in dBm)

def s_mode(self, s_mode = ’S11’):

self.write(’%s;’ % s_mode)

def format_polar(self):
self.write(’POLA’) # Set to polar coordinates

def format_logarithmic(self):
self.write(’LOGM’) # Set to polar coordinates

def run_sweep_ri(self):
""" Runs a sweep using whatever settings are currently on the NA and
returns the real
and imaginary components of each data point """
self.format_polar() # Set to polar coordinates
f_start = float(self.ask(’STAR?;’))
f_span = float(self.ask(’SPAN?;’))
f_stop = float(self.ask(’STOP?;’))
num_pts = int(float(self.ask(’POIN?;’)))

print ’Sweeping from %0.0d MHz to %0.0d MHz, with %0.0d points’ % (f_start
/1e6, f_stop/1e6, num_pts)

temp = self.timeout; self.timeout = 20

completion = self.ask(’OPC?;SING;’) # Runs a SINGle sweep, and waits for
the OPeration to Complete
self.timeout = temp

self.write(’FORM4;’) # Make the data output in ASCII

data = self.ask_for_values(’OUTPFORM;’)

156
 F = np.linspace(f_start, f_stop, num_pts, endpoint = True)
R = data[::2] # Every other element starting with element 0
I = data[1::2]
self.write(’CONT’)
return F, R, I

def run_sweep_mag(self):
""" Runs a sweep using whatever settings are currently on the NA and
returns the real
and imaginary components of each data point """
self.format_logarithmic() # Set to logarithimic coordinates
f_start = float(self.ask(’STAR?;’))
f_span = float(self.ask(’SPAN?;’))
f_stop = float(self.ask(’STOP?;’))
num_pts = int(float(self.ask(’POIN?;’)))

print ’Sweeping from %0.0d MHz to %0.0d MHz, with %0.0d points’ % (f_start
/1e6, f_stop/1e6, num_pts)

temp = self.timeout; self.timeout = 20

completion = self.ask(’OPC?;SING;’) # Runs a SINGle sweep, and waits for
the OPeration to Complete
self.timeout = temp

self.write(’FORM4;’) # Make the data output in ASCII

data = self.ask_for_values(’OUTPFORM;’)

f = np.linspace(f_start, f_stop, num_pts, endpoint = True)

F = f[::1]
M = data[::2] # Every other element starting with element 0
self.write(’CONT’)
return F, M

157
JDS Fitel HA9 Optical Attenuator

from pyvisa import visa

import numpy as np
import time

class JDSHA9(visa.Instrument):
"""Python class for JDS HJA9 Optical Attenuator, written by Adam McCaughan."""

def __init__(self, args):

super( visa.Instrument, self).__init__(args)
self.timeout = 2

def set_attenuation_db(self, attenuation_db = 10):

self.write((’ATT %0.1f dB’ % attenuation_db))

def set_beam_block(self, beam_block = True):

self.write((’D %0.0f’ % beam_block))

# att = JDSHA9(’GPIB0::10’)

158
Keithley 2400 Sourcemeter

from pyvisa import visa

class Keithley2400(visa.Instrument):
"""Python class for Keithley 2400 Sourcemeter, written by Adam McCaughan"""
def __init__(self, args):
super( visa.Instrument, self).__init__(args)
# Anything else here that needs to happen on initialization
self.timeout = 1 # Set timeout to 1 second

def reset(self):
self.write(’*RST’)

def setup_read_volt(self):
self.write(’*RST’)
self.write(’:SOUR:FUNC CURR’)
self.write(’:SOUR:CURR:LEVEL 0E-6’)
self.write(’SENS:FUNC \"VOLT\"’)

def setup_4W_source_I_read_V(self):
self.write(’*RST’)
self.write(’:SOUR:FUNC CURR’) # Set operation mode to: source current
self.write(’:SOUR:CURR:LEVEL 0E-6’) # Set current level to 0 uA
self.write(’:SYST:RSEN 1’) # Turn off "Remote Sensing" aka 4-wire
measurement mode
self.write(’SENS:FUNC \"VOLT\", \"CURR\"’) # Have it output

def setup_2W_source_I_read_V(self):
self.write(’*RST’)

159
 self.write(’:SOUR:FUNC CURR’) # Set operation mode to: source current
self.write(’:SOUR:CURR:LEVEL 0E-6’) # Set current level to 0 uA
self.write(’:SYST:RSEN 0’) # Turn off "Remote Sensing" aka 4-wire
measurement mode
self.write(’SENS:FUNC \"VOLT\", \"CURR\"’) # Have it output

def setup_2W_source_V_read_I(self):
self.write(’*RST’)
self.write(’:SOUR:FUNC VOLT’) # Set operation mode to: source voltage
self.write(’:SOUR:VOLT:LEVEL 0E-3’) # Set voltage level to 0 mV
self.write(’:SYST:RSEN 0’) # Turn off "Remote Sensing" aka 4-wire
measurement mode
self.write(’SENS:FUNC \"VOLT\", \"CURR\"’) # Have it output

def set_output(self, output = False):

if output is True: self.write(’:OUTP ON’)
if output is False: self.write(’:OUTP OFF’)

def set_measurement_time(self, plc_cycles = 1.0):

""" plc_cycles Sets integration time Keithley. Each cycle corresponds to
1/60th of a second. Default is 1.0, Max is 10.0. Min is 0.01. See
Keithley
manual p18-70 for more details """
self.write(’:NPLCycles %0.2f’ % plc_cycles)

def disable_remote(self):
""" Simulates the pressing of the "LOCAL" button on the Keithley
which will take the keithley out of remote mode """
self.write(’:SYST:KEY 23’)

def set_compliance_i(self, compliance_i = 10e-6):

160
 self.write(’:SENS:CURR:PROT %0.3e’ % compliance_i)

def set_compliance_v(self, compliance_v = 10e-6):

self.write(’:SENS:VOLT:PROT %0.3e’ % compliance_v)

def set_current(self, current = 0e-6):

self.write(’:SOUR:CURR:LEVEL %0.4e’ % current) # Set current level

def set_voltage(self, voltage = 0e-6):

self.write(’:SOUR:VOLT:LEVEL %0.4e’ % voltage) # Set current level

def read_voltage_and_current(self):
read_str = self.ask(’:READ?’)
# See page 18-51 of manual, returns: voltage, current, resistance,
timestamp, status info
# Returns something like ’5.275894E-05,-1.508318E-06,+9.910000E
+37,+2.562604E+03,+3.994000E+04’
data = read_str.split(’,’)
voltage, current = float(data[0]), float(data[1])
return voltage, current

def read_current(self, current = 0e-6):

voltage, current = self.read_voltage_and_current()
return current

def read_voltage(self):
voltage, current = self.read_voltage_and_current()
return voltage

161
SRS SIM928 adjustable voltage source

from pyvisa import visa

class SIM928(visa.Instrument):
"""Python class for SRS SIM928 Isolated Voltage Source inside a SIM900
mainframe, written by Adam McCaughan"""
def __init__(self, sim900port, args):
super( visa.Instrument, self).__init__(args)
self.sim900port = sim900port
# Anything else here that needs to happen on initialization
def write_simport(self, message):
write_str = ’SNDT ’ + str(self.sim900port) + ’,\"’ + message + ’\"’
# print write_str
self.write(write_str) # Format of ’SNDT 4,\"GAIN 10\"’
def ask_simport(self, message):
write_str = ’SNDT ’ + str(self.sim900port) + ’,\"’ + message + ’\"’
return self.ask(write_str) # Format of ’SNDT 4,\"GAIN 10\"’
def reset(self):
self.write_simport(’*RST’)
def set_voltage(self, voltage=0.0):
# In a string, %0.4e converts a number to scientific notation
self.write_simport(’VOLT %0.4e’ %(voltage))
def set_output(self, output=False):
if output==True:
self.write_simport(’OPON’)
else:
self.write_simport(’OPOF’) # Only uses "OPOF" or "OPON": "OPOFF" does
not work

162
Bibliography

[1] M. W. Brenner, D. Roy, N. Shah, and A. Bezryadin. Dynamics of superconduct-

ing nanowires shunted with an external resistor. Phys. Rev. B, 85(22):224507,
June 2012.

[2] D. S. Holmes, A. L. Ripple, and M. A. Manheimer. Energy-Efficient Super-

conducting Computing–Power Budgets and Requirements. IEEE Trans. Appl.
Supercond., 23(3):1701610–1701610, June 2013.

[3] Robert H. Hadfield. Single-photon detectors for optical quantum information

applications. Nat. Photonics, 3(12):696–705, December 2009.

[4] Jonas Zmuidzinas. Superconducting Microresonators: Physics and Applica-

tions. Annu. Rev. Condens. Matter Phys., 3(1):169–214, March 2012.

[5] Karsten Sternickel and Alex I Braginski. Biomagnetism using SQUIDs: status
and perspectives. Supercond. Sci. Technol., 19(3):S160–S171, March 2006.

[6] John Clarke and Frank K. Wilhelm. Superconducting quantum bits. Nature,
453(7198):1031–1042, June 2008.

[7] Don M. Boroson, Bryan S. Robinson, Daniel V. Murphy, Dennis A. Burianek,

Farzana Khatri, Joseph M. Kovalik, Zoran Sodnik, and Donald M. Cornwell.
Overview and results of the Lunar Laser Communication Demonstration. In
Hamid Hemmati and Don M. Boroson, editors, SPIE Photonics West 2014-
LASE Lasers Sources, volume 8971, page 89710S, March 2014.

[8] Jonas Zmuidzinas and Paul L Richards. Superconducting Detectors and Mixers
for Millimeter and Submillimeter Astrophysics. October, 92(10), 2004.

[9] Peter K. Day, Henry G. LeDuc, Benjamin A. Mazin, Anastasios Vayonakis, and
Jonas Zmuidzinas. A broadband superconducting detector suitable for use in
large arrays. Nature, 425(6960):817–821, October 2003.

[10] W. J. Skocpol, M. R. Beasley, and M. Tinkham. Phase-slip centers and nonequi-

librium processes in superconducting tin microbridges. J. Low Temp. Phys.,
16(1-2):145–167, July 1974.

[11] T. M. Klapwijk. Proximity effect from an Andreev perspective. J. Supercond.,

17(5):593–611, October 2004.

163
[12] Cihan Kurter, Alexander P. Zhuravel, Alexey V. Ustinov, and Steven M. An-
lage. Microscopic examination of hot spots giving rise to nonlinearity in super-
conducting resonators. Phys. Rev. B, 84(10):104515, September 2011.

[13] T. Orlando, J. Mooij, L. Tian, C. van der Wal, L. Levitov, S. Lloyd, and J. Mazo.
Superconducting persistent-current qubit. Phys. Rev. B, 60(22):15398–15413,
1999.

[14] J. J. A. Baselmans, A. F. Morpurgo, B. J. van Wees, and T. M. Klapwijk.

Reversing the direction of the supercurrent in a controllable Josephson junction.
Nature, 397(6714):43–45, January 1999.

[15] S. Nagasawa, S. Tahara, H. Numata, and S. Tsuchida. A miniaturized vortex

transitional memory cell for a Josephson high-speed RAM. Electron Devices
Meet. 1992. IEDM ’92. Tech. Dig. Int., pages 793–796, 1992.

[16] B.D. Josephson. Possible new effects in superconductive tunnelling. Phys. Lett.,
1(7):251–253, 1962.

[17] K. K. Likharev and V. K. Semenov. RSFQ logic/memory family: a new

Josephson-junction technology for sub-terahertz-clock-frequency digital sys-
tems. IEEE Trans. Appiled Supercond., 1(1):3–28, March 1991.

[18] Konstantin K. Likharev. Superconductor digital electronics. Phys. C Super-

cond., 482:6–18, November 2012.

[19] W. Chen, A.V. Rylyakov, V. Patel, J.E. Lukens, and K.K. Likharev. Rapid
single flux quantum T-flip flop operating up to 770 GHz. IEEE Trans. Appiled
Supercond., 9(2):3212–3215, June 1999.

[20] S Pagano and A Barone. Josephson junctions. Supercond. Sci. Technol.,

10(12):904–908, December 1997.

[21] P. de Vegvar, T. Fulton, W. Mallison, and R. Miller. Mesoscopic Transport in

Tunable Andreev Interferometers. Phys. Rev. Lett., 73(10):1416–1419, Septem-
ber 1994.

[22] P Cadden-Zimansky, J Wei, and V Chandrasekhar. Coherent nonlocal correla-

tions in Andreev interferometers. New J. Phys., 14(4):043004, April 2012.

[23] Christopher Checkley. Andreev Interferometry of Flux Qubits Driven By Radio

Frequency Field. PhD thesis, 2010.

[24] Alex D. Semenov, Gregory N. Gol’tsman, and Alexander A. Korneev. Quan-

tum detection by current carrying superconducting film. Phys. C Supercond.,
351(4):349–356, April 2001.

164
[25] Faraz Najafi, Jacob Mower, Nicholas C Harris, Francesco Bellei, Andrew Dane,
Catherine Lee, Xiaolong Hu, Prashanta Kharel, Francesco Marsili, Solomon
Assefa, Karl K Berggren, and Dirk Englund. On-chip detection of non-classical
light by scalable integration of single-photon detectors. Nat. Commun., 6:5873,
January 2015.

[26] Nathan R. Gemmell, Aongus McCarthy, Baochang Liu, Michael G. Tanner,

Sander D. Dorenbos, Valery Zwiller, Michael S. Patterson, Gerald S. Buller,
Brian C. Wilson, and Robert H. Hadfield. Singlet oxygen luminescence detection
with a fiber-coupled superconducting nanowire single-photon detector. Opt.
Express, 21(4):5005, February 2013.

[27] M.K. Mc Manus, J.A. Kash, S.E. Steen, S Polonsky, J.C. Tsang, D.R. Knebel,
and W Huott. PICA: Backside failure analysis of CMOS circuits using Pi-
cosecond Imaging Circuit Analysis. Microelectron. Reliab., 40(8-10):1353–1358,
August 2000.

[28] Matthew E. Grein, Andrew J. Kerman, Eric A. Dauler, Oleg Shatrovoy,

Richard J. Molnar, Danna Rosenberg, Jung Yoon, Catherine E. DeVoe,
Daniel V. Murphy, Bryan S. Robinson, and Don M. Boroson. Design of a
ground-based optical receiver for the lunar laser communications demonstra-
tion. In 2011 Int. Conf. Sp. Opt. Syst. Appl., pages 78–82. IEEE, May 2011.

[29] J Zmuidzinas. Superconducting Microresonator Detectors. pages 1–12, 2008.

[30] S. J. C. Yates, A. M. Baryshev, J. J. A. Baselmans, B. Klein, and R. Gusten.

Fast Fourier transform spectrometer readout for large arrays of microwave ki-
netic inductance detectors. Appl. Phys. Lett., 95(4):042504, 2009.

[31] W. Gallagher. Three-terminal superconducting devices. IEEE Trans. Magn.,

21(2):709–716, March 1985.

[32] D. Buck. The Cryotron-A Superconductive Computer Component. Proc. IRE,

44(4):482–493, April 1956.

[33] A. F. Morpurgo, T. M. Klapwijk, and B. J. van Wees. Hot electron tunable

supercurrent. Appl. Phys. Lett., 72(8):966, 1998.

[34] Seung-Beck Lee, Gregory D Hutchinson, David a Williams, David G Hasko,

and Haroon Ahmed. Superconducting nanotransistor based digital logic gates.
Nanotechnology, 14(2):188–191, February 2003.

[35] R. Sprik, W. J. Gallagher, S. I. Raider, B. Bumble, and C.-C. Chi. Transient

response of quasiparticle injected superconducting links. Appl. Phys. Lett.,
55(5):489, 1989.

[36] Ting Wah Wong, J. T C Yeh, and D. N. Langenberg. Quasiparticle-injection-

induced superconducting weak links. Phys. Rev. Lett., 37(3):150–153, July 1976.

165
[37] S. Faris, S. Raider, W. Gallagher, and R. Drake. Quiteron. IEEE Trans. Magn.,
19(3):1293–1295, May 1983.

[38] Tatsushi Akazaki, Hideaki Takayanagi, Junsaku Nitta, and Takatomo

Enoki. A Josephson field effect transistor using an InAs-inserted-channel
In0.52Al0.48As/In0.53Ga0.47As inverted modulation-doped structure. Appl.
Phys. Lett., 68(3):418, 1996.

[39] Huigao Duan, Hailong Hu, Karthik Kumar, Zexiang Shen, and Joel K W Yang.
Direct and Reliable Patterning of Plasmonic Nanostructures with Sub-10-nm
Gaps. ACS Nano, 5(9):7593–7600, September 2011.

[40] Joel K. W. Yang and Karl K. Berggren. Using high-contrast salty development
of hydrogen silsesquioxane for sub-10-nm half-pitch lithography. J. Vac. Sci.
Technol. B Microelectron. Nanom. Struct., 25(6):2025, 2007.

[41] Francesco Marsili, Faraz Najafi, Eric Dauler, Francesco Bellei, Xiaolong Hu,
Maria Csete, Richard J Molnar, and Karl K Berggren. Single-photon detectors
based on ultranarrow superconducting nanowires. Nano Lett., 11(5):2048–53,
May 2011.

[42] C. P. Foley and H. Hilgenkamp. Why NanoSQUIDs are important: an intro-

duction to the focus issue. Supercond. Sci. Technol., 22(6):064001, June 2009.

[43] Amit Finkler, Yehonathan Segev, Yuri Myasoedov, Michael L. Rappaport, Lior
Ne’eman, Denis Vasyukov, Eli Zeldov, Martin E. Huber, Jens Martin, and Amir
Yacoby. Self-Aligned Nanoscale SQUID on a Tip. Nano Lett., 10(3):1046–1049,
March 2010.

[44] Denis Vasyukov, Yonathan Anahory, Lior Embon, Dorri Halbertal, Jo Cuppens,
Lior Neeman, Amit Finkler, Yehonathan Segev, Yuri Myasoedov, Michael L
Rappaport, Martin E Huber, and Eli Zeldov. A scanning superconducting quan-
tum interference device with single electron spin sensitivity. Nat. Nanotechnol.,
8(September):639–44, 2013.

[45] Quentin P. Herr, Anna Y. Herr, Oliver T. Oberg, and Alexander G. Ioannidis.
Ultra-low-power superconductor logic. J. Appl. Phys., 109(10):103903, 2011.

[46] M. Tinkham. Introduction to Superconductivity: Second Edition. Dover Books

on Physics. Dover Publications, 2004.

[47] Terry P. Orlando and Kevin A. Delin. Foundations of Applied Superconductivity.

Reading, MA, 1990.

[48] G. N. Gol’tsman, O. Okunev, G. Chulkova, A. Lipatov, A. Semenov,

K. Smirnov, B. Voronov, A. Dzardanov, C. Williams, and Roman Sobolewski.
Picosecond superconducting single-photon optical detector. Appl. Phys. Lett.,
79(6):705, 2001.

166
[49] Kristine M Rosfjord, Joel K W Yang, Eric A Dauler, Andrew J Kerman, Vikas
Anant, Boris M Voronov, Gregory N Gol’tsman, and Karl K Berggren. Nanowire
single-photon detector with an integrated optical cavity and anti-reflection coat-
ing. Opt. Express, 14(2):527, 2006.
[50] F Marsili, V B Verma, J A Stern, S Harrington, A E Lita, T Gerrits,
I Vayshenker, B Baek, M D Shaw, R P Mirin, and S W Nam. Detecting sin-
gle infrared photons with 93% system efficiency. Nat. Photonics, 7(3):210–214,
February 2013.
[51] Andrew J. Kerman, Eric A. Dauler, William E. Keicher, Joel K. W. Yang,
Karl K. Berggren, G. Gol’tsman, and B. Voronov. Kinetic-inductance-limited
reset time of superconducting nanowire photon counters. Appl. Phys. Lett.,
88(11):111116, 2006.
[52] Andrew Kerman, Joel Yang, Richard Molnar, Eric Dauler, and Karl Berggren.
Electrothermal feedback in superconducting nanowire single-photon detectors.
Phys. Rev. B, 79(10):1–4, March 2009.
[53] S Berg. Fundamental understanding and modeling of reactive sputtering pro-
cesses. Thin Solid Films, 476:215–230, 2005.
[54] J.K.W. Yang, A.J. Kerman, E.A. Dauler, V. Anant, K.M. Rosfjord, and K.K.
Berggren. Modeling the Electrical and Thermal Response of Superconducting
Nanowire Single-Photon Detectors. IEEE Trans. Appl. Supercond., 17(2):581–
585, June 2007.
[55] A. Semenov, R. Nebosis, Yu. Gousev, M. Heusinger, and K. Renk. Analysis of
the nonequilibrium photoresponse of superconducting films to pulsed radiation
by use of a two-temperature model. Phys. Rev. B, 52(1):581–590, July 1995.
[56] K.D. Irwin, S.W. Nam, B. Cabrera, B. Chugg, G.S. Park, R.P. Welty, and
J.M. Martinis. A self-biasing cryogenic particle detector utilizing electrothermal
feedback and a SQUID readout. IEEE Trans. Appl. Supercond., 5(2):2690–2693,
1995.
[57] John R Clem. Current-Induced Suppression of Superconductivity in Thin Films
due to Simple Geometric Effects. Physics (College. Park. Md)., pages 1–10,
2011.
[58] F Gross, BS Chandrasekhar, and D Einzel. Anomalous temperature dependence
of the magnetic field penetration depth in superconducting UBe13. Zeitschrift
für Phys. B . . . , 188:175–188, 1986.
[59] M. Tinkham, J. Free, C. Lau, and N. Markovic. Hysteretic I-V curves of super-
conducting nanowires. Phys. Rev. B, 68(13):1–7, October 2003.
[60] W. J. Skocpol, M. R. Beasley, and M. Tinkham. Self-heating hotspots in su-
perconducting thin-film microbridges. J. Appl. Phys., 45(9):4054, 1974.

167
[61] Albert Schmid. A time dependent Ginzburg-Landau equation and its appli-
cation to the problem of resistivity in the mixed state. Phys. der Kondens.
Mater., 5(4):302–317, 1966.

[62] Klaus D. Usadel. Generalized diffusion equation for superconducting alloys.

Phys. Rev. Lett., 25(8):507–509, 1970.

[63] L. N. Bulaevskii, M. J. Graf, C. D. Batista, and V. G. Kogan. Vortex-induced

dissipation in narrow current-biased thin-film superconducting strips. Phys.
Rev. B, 83(14):144526, April 2011.

[64] Simon K H Lam and Sabaratnasingam Gnanarajan. Hysteretic behaviour of

nanoSQUIDs–prospective application as trapped-vortex memory. Supercond.
Sci. Technol., 22(6):064005, June 2009.

[65] AH Dayem and JJ Wiegand. Behavior of Thin-Film Superconducting Bridges

in a Microwave Field. Phys. Rev., 155(2):419–428, March 1967.

[66] H. Fink, V. Grünfeld, and A. López. Quantum-interference device without

Josephson junctions. Phys. Rev. B, 35(1):35–37, January 1987.

[67] Aico G. P. Troeman, Hendrie Derking, Bert Borger, Johannes Pleikies, Dick
Veldhuis, and Hans Hilgenkamp. NanoSQUIDs Based on Niobium Constric-
tions. Nano Lett., 7(7):2152–2156, July 2007.

[68] K. Hasselbach, C. Veauvy, and D. Mailly. MicroSQUID magnetometry and

magnetic imaging. Phys. C Supercond. its Appl., 332:140–147, 2000.

[69] Simon K H Lam, John R Clem, and Wenrong Yang. A nanoscale SQUID
operating at high magnetic fields. Nanotechnology, 22(45):455501, December
2011.

[70] C H Wu, Y T Chou, W C Kuo, J H Chen, L M Wang, J C Chen, K L Chen, U C

Sou, H C Yang, and J T Jeng. Fabrication and characterization of high-T(c)
YBa(2)Cu(3)O(7-x) nanoSQUIDs made by focused ion beam milling. Nan-
otechnology, 19(31):315304, 2008.

[71] R. Arpaia, M. Arzeo, S. Nawaz, S. Charpentier, F. Lombardi, and T. Bauch.

Ultra low noise YBa2Cu3O7-𝛿 nano superconducting quantum interference de-
vices implementing nanowires. Appl. Phys. Lett., 104(7):0–4, February 2014.

[72] V. V. Moshchalkov, L. Gielen, M. Dhallé, C. Van Haesendonck, and

Y. Bruynseraede. Quantum interference in a mesoscopic superconducting loop.
Nature, 361(6413):617–620, February 1993.

[73] Sebastien Michotte, Damien Lucot, and Dominique Mailly. Fluxoid quantiza-
tion in the critical current of a niobium superconducting loop far below the
critical temperature. Phys. Rev. B, 81(10):100503, March 2010.

168
[74] S. Adam, X. Hallet, L. Piraux, D. Lucot, and D. Mailly. Switching current
modulations induced by vortices rearrangement in mesoscopic superconducting
loops. Phys. Rev. B, 84(10):104512, September 2011.

[75] A. Gubin, K. Il’in, S. Vitusevich, M. Siegel, and N. Klein. Dependence of

magnetic penetration depth on the thickness of superconducting Nb thin films.
Phys. Rev. B, 72(6):1–8, August 2005.

[76] A. Troeman, S. van der Ploeg, E. Il’Ichev, H.-G. Meyer, A. Golubov,

M. Kupriyanov, and H. Hilgenkamp. Temperature dependence measurements
of the supercurrent-phase relationship in niobium nanobridges. Phys. Rev. B,
77(2):024509, January 2008.

[77] K. K. Likharev. Superconducting weak links. Rev. Mod. Phys., 51(1):101–159,

January 1979.

[78] John Clem and V. Kogan. Kinetic impedance and depairing in thin and narrow
superconducting films. Phys. Rev. B, 86(17):174521, November 2012.

[79] Anthony J Annunziata, Daniel F Santavicca, Luigi Frunzio, Gianluigi Catelani,

Michael J Rooks, Aviad Frydman, and Daniel E Prober. Tunable supercon-
ducting nanoinductors. Nanotechnology, 21(44):445202, November 2010.

[80] K. Il’in, M. Siegel, A. Semenov, A. Engel, and H.-W. Hbers. Critical current
of Nb and NbN thin-film structures: The cross-section dependence. Phys. status
solidi, 2(5):1680–1687, March 2005.

[81] S. Michotte, S. Mátéfi-Tempfli, L. Piraux, D. Vodolazov, and F. Peeters. Con-

dition for the occurrence of phase slip centers in superconducting nanowires
under applied current or voltage. Phys. Rev. B, 69(9):094512, March 2004.

[82] Shi-zeng Lin and Lev N Bulaevskii. I-V characteristics of short superconducting
nanowires with different bias and shunt: a dynamic approach. page 5, August
2013.

[83] A. T. Fiory, A. F. Hebard, P. M. Mankiewich, and R. E. Howard. Penetration

depths of high Tc films measured by two-coil mutual inductances. Appl. Phys.
Lett., 52(25):2165, 1988.

[84] Adam N. McCaughan and Karl K. Berggren. A superconducting-nanowire

three-terminal electrothermal device. Nano Lett., 14(10):5748–53, October
2014.

[85] M Ejrnaes, A Casaburi, R Cristiano, N Martucciello, F Mattioli, A Gaggero,

R Leoni, J-C Villégier, and S Pagano. Characterization of superconducting pulse
discriminators based on parallel NbN nanostriplines. Supercond. Sci. Technol.,
24(3):035018, March 2011.

169
[86] AV Gurevich and RG Mints. Self-heating in normal metals and superconductors.
Rev. Mod. Phys., 59(4):941–999, October 1987.

[87] TL Peterson, I. Maartense, and R.R. Biggers. Self-heating hotspot effects in

HTS thin films. IEEE Trans. Appl. Supercond., 5(2):1436–1439, June 1995.

[88] Allen Rothwarf and B. N. Taylor. Measurement of recombination lifetimes in

superconductors. Phys. Rev. Lett., 19(1):27–30, July 1967.

[89] D. E. Prober. Superconducting terahertz mixer using a transition-edge mi-

crobolometer. Appl. Phys. Lett., 62(17):2119, 1993.

[90] Anthony J. Annunziata, Orlando Quaranta, Daniel F. Santavicca, Alessandro

Casaburi, Luigi Frunzio, Mikkel Ejrnaes, Michael J. Rooks, Roberto Cristiano,
Sergio Pagano, Aviad Frydman, and Daniel E. Prober. Reset dynamics and
latching in niobium superconducting nanowire single-photon detectors. J. Appl.
Phys., 108(8):084507, 2010.

[91] F. Najafi, F. Marsili, E. Dauler, R. J. Molnar, and K. K. Berggren. Timing per-

formance of 30-nm-wide superconducting nanowire avalanche photodetectors.
Appl. Phys. Lett., 100(15):152602, 2012.

[92] K. S. Il’in, M. Lindgren, M. Currie, A. D. Semenov, G. N. Gol’tsman, Roman

Sobolewski, S. I. Cherednichenko, and E. M. Gershenzon. Picosecond hot-
electron energy relaxation in NbN superconducting photodetectors. Appl. Phys.
Lett., 76(19):2752, 2000.

[93] M. Beck, M. Klammer, S. Lang, P. Leiderer, V. V. Kabanov, G. N.

Gol’tsman, and J. Demsar. Energy-Gap Dynamics of Superconducting NbN
Thin Films Studied by Time-Resolved Terahertz Spectroscopy. Phys. Rev. Lett.,
107(17):177007, October 2011.

[94] M H Volkmann, A Sahu, C J Fourie, and O a Mukhanov. Implementation of

energy efficient single flux quantum digital circuits with sub-aJ/bit operation.
Supercond. Sci. Technol., 26(1):015002, January 2013.

[95] H. L. Hortensius, E. F. C. Driessen, T. M. Klapwijk, K. K. Berggren, and J. R.

Clem. Critical-current reduction in thin superconducting wires due to current
crowding. Appl. Phys. Lett., 100(18):182602, 2012.

[96] M. Bell, A. Sergeev, V. Mitin, J. Bird, A. Verevkin, and G. Gol’tsman. One-

dimensional resistive states in quasi-two-dimensional superconductors: Experi-
ment and theory. Phys. Rev. B, 76(9):094521, September 2007.

[97] Nayana Shah, David Pekker, and Paul Goldbart. Inherent Stochasticity of
Superconductor-Resistor Switching Behavior in Nanowires. Phys. Rev. Lett.,
101(20):207001, November 2008.

170
 [98] M Sahu. Switching current distributions of superconducting nanowires: Evi-
dence of quantum phase slip events. PhD thesis, 2009.

[99] J. B. Johnson. Thermal Agitation of Electricity in Conductors. Phys. Rev.,

32(1):97–109, July 1928.

[100] Hiroyuki Shibata, Kaoru Shimizu, Hiroki Takesue, and Yasuhiro Tokura. Ulti-
mate low system dark count rate for superconducting nanowire single-photon
detector. pages 3–6, 2015.

[101] Peng Li, Phillip Wu, Yuriy Bomze, Ivan Borzenets, Gleb Finkelstein, and
A. Chang. Switching Currents Limited by Single Phase Slips in One-
Dimensional Superconducting Al Nanowires. Phys. Rev. Lett., 107(13):1–5,
September 2011.

[102] Jack Ekin. Experimental Techniques for Low-Temperature Measurements. Ox-

ford University Press, October 2006.

171