Physics Of Silicon Image Sensors Stefanov instant download
Physics Of Silicon Image Sensors Stefanov instant download
download
https://ebookbell.com/product/physics-of-silicon-image-sensors-
stefanov-47229712
https://ebookbell.com/product/evolution-of-silicon-sensor-technology-
in-particle-physics-1st-edition-frank-hartmann-auth-4208334
https://ebookbell.com/product/evolution-of-silicon-sensor-technology-
in-particle-physics-2nd-edition-frank-hartmann-auth-6789892
https://ebookbell.com/product/physics-and-technology-of-
amorphouscrystalline-heterostructure-silicon-solar-cells-1st-edition-
wilfried-van-sark-2455344
https://ebookbell.com/product/physics-of-molecular-and-cellular-
processes-krastan-b-blagoev-46081122
Physics Of Cancer Volume 1 Interplay Between Tumor Biology
Inflammation And Cell Mechanics 2nd Edition Claudia Mierke
https://ebookbell.com/product/physics-of-cancer-volume-1-interplay-
between-tumor-biology-inflammation-and-cell-mechanics-2nd-edition-
claudia-mierke-46123770
https://ebookbell.com/product/physics-of-cancer-volume-2-cellular-and-
microenvironmental-efects-2nd-edition-claudia-mierke-46123772
https://ebookbell.com/product/physics-of-flow-in-porous-media-jens-
feder-eirik-grude-flekky-46196878
https://ebookbell.com/product/physics-of-and-science-with-xray-
freeelectron-lasers-j-hastings-46655456
https://ebookbell.com/product/physics-of-the-atmosphere-climatology-
and-environmental-monitoring-modern-problems-of-atmospheric-physics-
climatology-and-environmental-monitoring-1st-edition-robert-
zakinyan-47383436
CMOS Image Sensors
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system
or transmitted in any form or by any means, electronic, mechanical, photocopying, recording
or otherwise, without the prior permission of the publisher, or as expressly permitted by law or
under terms agreed with the appropriate rights organization. Multiple copying is permitted in
accordance with the terms of licences issued by the Copyright Licensing Agency, the Copyright
Clearance Centre and other reproduction rights organizations.
Permission to make use of IOP Publishing content other than as set out above may be sought
at [email protected].
Konstantin D Stefanov has asserted their right to be identified as the author of this work in
accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
DOI 10.1088/978-0-7503-3235-4
Version: 20221101
IOP ebooks
British Library Cataloguing-in-Publication Data: A catalogue record for this book is available
from the British Library.
IOP Publishing, No.2 The Distillery, Glassfields, Avon Street, Bristol, BS2 0GR, UK
US Office: IOP Publishing, Inc., 190 North Independence Mall West, Suite 601, Philadelphia,
PA 19106, USA
To David Burt, from whom I have learnt many of the things in this book,
and to my family, for their support and encouragement.
Contents
Preface xi
Acknowledgement xiii
Author biography xiv
List of frequently used abbreviations xv
Table of common symbols and units xvi
vii
CMOS Image Sensors
viii
CMOS Image Sensors
5 Characterisation 5-1
5.1 Introduction 5-1
5.2 Readout modes 5-2
5.3 Principles of EO characterisation 5-4
5.4 Photoresponse, non-uniformity and nonlinearity 5-7
5.5 Photon transfer curve 5-17
5.5.1 Principles 5-17
5.5.2 Frame differencing 5-21
5.5.3 System gain, CVF and noise 5-24
5.5.4 Nonlinear PTC 5-27
5.5.5 PTC from dark current 5-30
ix
CMOS Image Sensors
6 Electronics 6-1
6.1 On-chip electronics 6-1
6.1.1 Architecture 6-1
6.1.2 Column buffers 6-2
6.1.3 Column amplifiers 6-3
6.1.4 CDS circuits 6-7
6.1.5 Row drivers 6-12
6.1.6 Pixel addressing 6-14
6.1.7 Analogue switches and multiplexers 6-17
6.1.8 Output amplifier 6-19
6.2 Off-chip electronics 6-21
6.2.1 General requirements 6-21
6.2.2 Signal amplifiers 6-23
6.2.3 Power supplies 6-29
6.2.4 Bias circuits 6-30
6.2.5 Noise measurements 6-32
Chapter summary 6-34
References 6-35
x
Preface
Image sensors are fascinating devices that straddle the boundary between semi-
conductor physics and electronic engineering. Nowadays, they are used in almost
everything, come in bewildering varieties, and are mostly made with complementary
metal-oxide semiconductor (CMOS) technology, like the billions of other integrated
circuits (IC) manufactured every year. And they look good! They are among the
handful of IC types which can be easily seen through their transparent glass cover.
To understand how they work, we need to know a fair bit of semiconductor
physics, especially when dealing with high performance imagers designed for science
applications. However, this is not always enough; there are a lot of electronic circuits
inside a CMOS image sensor (CIS), and even the simple ones can show subtle
behaviour and throw up surprises. Without claiming to cover everything, this book
strives to cover both the semiconductor physics and the essential electronics found
inside a CMOS image sensor.
A relatively small number of concepts from semiconductor physics form the
backbone of image sensors’ operation—depletion, drift, diffusion, recombination,
charge conversion, to name a few. Knowing them well provides the foundations to
understand practically all image sensors and helps with the more complex structures
that exist or are yet to be invented.
It is impossible to imagine doing any serious work into image sensors without
semiconductor technology CAD (TCAD). Very often it is the only way to ‘see’ what
is happening inside, and this book offers many examples of device simulations. A
successful TCAD simulation is a good result, but does not guarantee that something
will work. However, if something doesn’t work in TCAD, it’s probably not going to
work in silicon either.
To make full use of this book, some basic knowledge of electronics is essential.
Knowing what an amplifier is, being familiar with gain, bandwidth, and noise, can
be very advantageous. Freely available electronic simulation tools, such as SPICE,
are great for designing and verifying the performance of various circuits.
Throughout my career, my hobby in electronics has helped me enormously when
working with image sensors. Building my first electronic circuit at around 14 years of
age, I was fascinated but I only had a faint understanding of how it worked. In a few
years, electronics gradually started to make more sense, and after learning semi-
conductor physics at university it was clear to me that this is what I wanted to do.
I have often found that many important ‘bread and butter’ topics in image sensors
and their operation are difficult to find in books and papers, and sometimes are not
there at all. Some of those I have only been able to find out in discussions with more
experienced colleagues who have been longer in the field. This book is an attempt to
put some of this ‘unofficial knowledge’, some of which could be simply due to my
ignorance, in one place.
This book is intended to be used as a tutorial and has many examples, taken
mostly from practice. Solved examples are essential for proper understanding of the
theory and bring ‘life’ to the formulas. They also help with appreciating the
xi
CMOS Image Sensors
xii
Acknowledgement
The stimulating research environment at the Centre for Electronic Imaging (CEI) at
the Open University is one of the main reasons for the existence of this book. Many
of the topics and ideas presented here came up from discussions with my colleagues,
students, industrial collaborators, and external partners, and by the challenges and
the difficult questions they often had.
I am grateful for the inspiration and the knowledge I have gained thanks to David
Burt, Andrew Holland, Chris Damerell, Ray Bell, Jérôme Pratlong, Paul Jerram,
Doug Jordan, Neil Murray, Pete Turner, Dave Barry, David Hall, Matthew Soman,
Julian Heymes, Martin Prest, James Ivory, Chiaki Crews, Nathan Bush, Steve
Bowring and Giulio Villani.
Finally, I would like to thank IOP Publishing, and in particular John Navas for
overseeing the production and for making this book possible.
Konstantin Stefanov
August 2022
xiii
Author biography
Konstantin D Stefanov
Konstantin D Stefanov was born in Rousse, Bulgaria, and has
received his MSc in applied physics from Sofia University
‘St. Kliment Ohridski’. He received his PhD degree in physics
from Saga University, Japan, in 2001. As a research scientist at the
Rutherford Appleton Laboratory in Oxfordshire, UK, Dr Stefanov
has worked on the development of CCD and CMOS sensors for
particle physics. Since 2012 he has been working at the Centre for
Electronic Imaging at the Open University, Milton Keynes, UK,
where he is developing CMOS image sensors for scientific and space applications. His
research interests are in the areas of physics, technology, and design of CMOS image
sensors for science applications, semiconductor device simulations, device character-
isation, radiation damage effects, detector electronics and data acquisition systems. He
has published over 80 research papers, has co-written two book chapters and holds
several patents on CMOS image sensors.
xiv
List of frequently used
abbreviations
AC Alternating current
ADC Analogue-to-digital converter
ADU Analogue-to-digital unit
APS Active pixel sensor
BSI Back-side illumination
CCD Charge coupled device
CDS Correlated double sampling
CG Conversion gain
CIS CMOS image sensor
CMOS Complementary metal-oxide-semiconductor
CVF Charge-to-voltage conversion factor
DC Direct current
DN Digital number
DR Dynamic range
DSNU Dark signal non-uniformity
DUT Device under test
EO Electro-optical
ENC Equivalent noise charge
ETF Electrical transfer function
FPN Fixed pattern noise
FSI Front-side illumination
FWC Full well capacity
HDR High dynamic range
IR Infrared
LED Light emitting diode
MOS Metal-oxide-semiconductor
MOSFET Metal-oxide-semiconductor field effect transistor
MTF Modulation transfer function
MVC Mean-variance curve
NIR Near infrared
NMOS N-channel MOSFET
PD Photodiode
PMOS P-channel MOSFET
PPD Pinned photodiode
PRNU Photo response non-uniformity
PTC Photon transfer curve
RMS Root mean square
RTN Random telegraph noise
RTS Random telegraph signal
SF Source follower
SNR Signal-to-noise ratio
QE Quantum efficiency
UV Ultraviolet
xv
Table of common symbols
and units
xvi
CMOS Image Sensors
xvii
IOP Publishing
Chapter 1
The fundamentals
in low light conditions the photogenerated currents can be extremely small and be
counted in few electrons per second. Such currents are impossible to measure directly,
therefore the method we use is to integrate the charge over a certain time on a collection
element because this creates a signal that is much easier to measure. Most image
sensors, including the ones described in this book, are of the integrating type.
1-2
CMOS Image Sensors
1240
E ph = (1.2)
λ
Using (1.2) we can calculate that a photon with energy equal to the bandgap of
silicon (E ph = Eg = 1.12 eV at 300 K) has a wavelength of 1107 nm, and this is
commonly referred to as the cut-off wavelength. The bandgap increases slightly at
lower temperatures [2] leading to shorter cut-off wavelength and weaker absorption
in the near-IR band.
In silicon, as an indirect bandgap semiconductor, the excitation of a valence
electron into the conduction band requires that lattice vibrations (phonons) are
involved to obey both energy and momentum conservation laws [2]. As the photon
energy increases, the amount of momentum transfer must increase too, and the
energy required to generate one electron–hole pair gradually increases. Because of
this, a photon with energy equal to double the silicon bandgap (2.24 eV, or 554 nm)
still generates one electron–hole pair, and not two.
The band structure of silicon has a direct bandgap of 3.1 eV [3] (400 nm) as well,
allowing an electron–hole pair to be created directly, without the assistance of a
phonon. However, it is thanks to its indirect bandgap that silicon is sensitive to
visible light; if it only had the 3.1 eV direct bandgap it would have been sensitive
only to wavelengths shorter than 400 nm and unusable for mainstream imaging.
A single electron–hole pair, corresponding to internal quantum yield of unity, is
created up to a photon energy equal to three times the bandgap (3.36 eV, or 369 nm) [4].
Above this energy two e–h pairs begin to be created, but at a very low rate. Multiple
e–h pair creation becomes significant only when the photon energy exceeds 4 eV
(310 nm) [5, 6].
As the photon energy increases further, the ionisation energy E w needed for the
creation of one e–h pair reaches a peak of approximately 4.5 eV [4, 5]. For E ph > 10 eV
the pair creation energy E w levels off to around 3.65 eV, as shown in figure 1.2.
An incoming beam of photons with flux Φ0 (number of photons per unit area per
second) and energy higher than the bandgap is gradually absorbed in the
Figure 1.2. Electron–hole pair creation energy in silicon at 140 and 300 K (data from [1]). The dashed line
marks E w = E ph .
1-3
CMOS Image Sensors
Figure 1.3. Photon absorption in silicon for low energy photons from near-UV to near-IR at 300 K. The data
is from [7].
semiconductor. Ignoring any reflections, the flux Φ(x ) at a distance x away from the
illuminated surface is given by the Beer–Lambert law:
Φ(x ) = Φ0e−αx (1.3)
where α is the absorption coefficient, typically measured in units of cm–1. At distance
x0 = 1/α the incoming photon flux is attenuated by 1/e , which means that 63% of the
light has been absorbed. The distance x0 is called absorption length and is often
more practical to use than the absorption coefficient because it allows straightfor-
ward comparison with the dimensions used in image sensors.
The absorption length depends strongly on the wavelength of light and changes
by a factor of 50 between the lower end (400 nm) and the top end (700 nm) of the
visible light range1, as shown in figure 1.3 and table 1.1. As the bandgap increases at
low temperatures the absorption length increases too, especially at near-IR wave-
lengths [7]. It is worth noting that photon absorption does not depend on the doping
concentration or the free carrier concentration (either electrons or holes) in silicon
for most practical cases.
Very often we would like to know what the silicon thickness should be to achieve
certain level of photon absorption.
Example 1.1. Calculate the silicon thickness for 95% photon absorption for light
with 400, 700 and 900 nm wavelength.
Solution: 95% absorption means that only 5% of the light is left. From formula (1.3)
we have e−αx = 0.05 and therefore x = −ln (0.05)/α = 3.0/α . From table 1.1 we get
1/α = x0 = 0.105 μm for 400 nm wavelength, therefore the thickness is x = 0.31 μm.
For 700 nm we have 1/α = 5.263 and 15.8 μm silicon thickness; and for 900 nm
x = 97.9 μm.
1
In the CIE (The International Commission on Illumination) luminous efficiency functions [8] the wavelength
range of visibility is 380–700 nm.
1-4
CMOS Image Sensors
300 0.006
350 0.010
400 0.105
450 0.392
500 0.901
550 1.565
600 2.415
650 3.559
700 5.263
750 7.692
800 11.77
850 18.69
900 32.68
950 63.69
1000 156.3
1050 613.5
1100 2857
The light attenuation with distance for the wavelengths used in example 1.1 is
plotted in figure 1.4 and illustrates the huge differences in silicon thickness required
for the same absorption. This example shows the extremes of the visible range; in
practice silicon thickness of 5 μm is often sufficient for visible light imagers because it
allows acceptable absorption in the red end of the spectrum around 600–650 nm.
The fact that in the visible wavelength range each photon creates one electron–
hole pair can be used to calculate the total light-generated charge in a volume of
silicon. This charge, if collected, is the electrical output of the image sensor.
Knowing the incident optical power Pph (measured in watts) and the photon energy
we can calculate the number of e–h pairs ∆Ne−h generated per unit time Δt based on
the energy conservation law, simply as this:
ΔNe−h Pph
= (1.4)
Δt E ph
As we can see the number of generated e–h pairs is inversely proportional to the
photon energy, therefore lower energy photons, corresponding to near-IR and red
light generate more carriers at the same optical power. While this is true, e–h pair
generation requires that the photons are absorbed; for those long wavelengths the
silicon must be very thick to ensure full absorption as figure 1.4 tells us.
1-5
CMOS Image Sensors
Figure 1.4. Photon absorption in silicon for violet (λ = 400 nm), red (λ = 700 nm) and for 900 nm light in the
near-infrared.
Example 1.2. Calculate the number of e–h pairs generated per second by red light
(λ = 650 nm) with an irradiance (power per unit area) of Ee = 1 W m−2 in a square
pixel with a size a = 10 μm. Assume that the light is fully absorbed in the pixel’s
volume.
Solution: From equation (1.2) we find that for λ = 650 nm the photon energy is
E ph = 1.91 eV. The energy deposited in the pixel per second is the irradiance multiplied
by the pixel area a 2 . The number of generated e–h pairs per second is the energy
deposited in the pixel per second, divided by the energy to create one e–h pair:
ΔNe−h E a2 1 × 10 × 10−6 × 10 × 10−6
= e = = 3.27 × 108 s−1
Δt qE ph 1.6 × 10−19 × 1.91
Here we have multiplied E ph by the elementary charge q to convert the photon
energy from eV to Joules. Irradiance of 1 W m−2 at 650 nm corresponds to
approximately 68.3 lux, or a dimly lit room. This example shows that even meagre
illumination manages to create a third of a billion e–h pairs every second in a tiny
10 μm pixel.
If all the e–h pairs were collected, a steady state photocurrent Iph will flow; it is
given by multiplying formula (1.4) by the elementary charge to convert the number
or e–h pairs per second to coulombs per second, which is current:
qΔNe−h
Iph = (1.5)
Δt
1-6
CMOS Image Sensors
Example 1.3. Calculate the photocurrent flowing in the pixel in example 1.2.
Solution: Multiplying the answer by the elementary charge gives:
qΔNe−h
Iph = = 1.6 × 10−19 × 3.27 × 108 = 5.235 × 10−11A = 52.4 pA
Δt
1.2.2 Ionisation
Silicon makes an excellent sensor material not just for the near-IR, visible and UV
light, but also for much more energetic photons, such as x-rays. The absorption
length covering photon energies from 1.2 eV to 10 keV in figure 1.5 shows what
happens at the higher end of silicon’s usable range: similarly to the near-IR end,
silicon becomes transparent beyond photon energy of about 10 keV. The absorption
above 100 eV shows discontinuous absorption edges at the energy levels of the L-
shell (≈100 eV) and the K-shell (1839 eV) of the silicon atom. As the photon energy
exceeds the binding energy of a shell, the electrons occupying it can be excited and
the photon absorption sharply goes up. This corresponds to a stepwise decrease in
the absorption length, most clearly seen at the K-edge.
For photons with E ph > ≈50 eV the ionisation energy is E w = 3.65 eV at 300 K
[11] and is nearly constant. This allows us to calculate the number of electron–hole
pairs Ne−h generated by x-rays and gamma-rays using this simple expression:
E ph
Ne−h = (1.8)
Ew
1-7
CMOS Image Sensors
Figure 1.5. Photon absorption due to photoeffect in silicon for a wider energy range. Data for <5 eV from [7],
5 – 20000 eV from [9], and >30 eV from [10].
The ionisation energy is temperature dependant; this is due to the reduction of the
bandgap as the temperature increases [1, 12]. X-rays in the range 1–10 keV, emitted
by radioactive sources and x-ray fluorescence from various materials are particularly
useful for sensor characterisation. They are widely used for calibration due to the
well-known x-ray energies and the amount of charge created in silicon by them.
Also, the initial charge cloud created by low energy x-rays is very compact [13, 14]
and this allows the charge to be considered a point source.
One very popular calibration source is the 55Fe isotope which decays via electron
capture to manganese (55Mn) with a half-life of 2.737 years. 55Mn emits character-
istic K-shell x-rays with energies 5.89 keV (Mn-Kα) and 6.49 keV (Mn-Kβ), with
probabilities of 24.4% and 2.9%, correspondingly [15]. Figure 1.6 shows an example
of a 55Fe spectrum obtained by a CMOS image sensor.
The absorption length for 5.9 keV photons in silicon is approximately 28 μm [10].
This length is much larger than the depth of the active silicon in the typical optical
sensor, therefore only a small fraction of the incoming x-rays is absorbed and
converted to charge.
1-8
CMOS Image Sensors
Figure 1.6. 55Fe spectrum obtained by a CMOS image sensor. The peak at 0 ADU is due to pixels without x-ray
signal, and the continuum leading to the x-ray peaks is caused by charge collected by more than one pixel.
At much higher photon energies two other mechanisms overtake the photoeffect
and begin to play an increasing role—Compton effect and electron–positron pair
creation [16]. Besides the detection of optical and x-ray photons, silicon is widely
used for the detection of high energy, ionising particles. Traversing the material,
charged particles lose energy due to several mechanisms, and most of that energy
loss is due to ionisation. The ionisation loss is very high at low particle energies but
decreases and flattens off at higher energies in a logarithmic dependence [16].
Particles with energies at and above the plateau, which usually lies at hundreds of
MeV, are called minimum ionising particles (MIP).
The average number of the created electron–hole pairs is calculated by the most
probable energy loss divided by the ionisation energy as in (1.8). The energy loss due to
ionisation has large statistical fluctuations. For particles which lose only a small part of
their energy in the material (i.e. the material is ‘thin’) the energy loss is described by the
Landau distribution [17]. In a couple of microns of silicon there is a significant
probability that a traversing high energy particle will cause no ionisation at all [18].
At the same time, the probability that the energy loss can far exceed the most probable
value is also significant, due to the long tail in the Landau distribution.
The most probable number of e–h pairs increases with the thickness, as shown in
figure 1.7, and above 10 μm the energy loss begins to approach the Landau distribution.
Figure 1.7. Most probable number of electron–hole pairs per micrometre of track for high energy charged
particles, data from [17].
Left on its own, the charge will simply diffuse out, never to be seen again.
Diffusion is fundamental in nature and always occurs when there is a difference in
carrier concentration. The charge generated in pixels receiving more illumination
will diffuse towards pixels receiving less illumination, until we get nearly uniform
charge ‘blob’ everywhere. Obviously, this is not what we want to happen in an image
sensor.
We need a charge collecting element—something that is electrically attractive to
either electrons or holes (but obviously cannot be attractive to both). To make the
charge move in a particular direction for collection we need to create an electric field;
within it the charge experiences an electrostatic force and begins to accelerate in a
direction opposite to the field (for electrons) and along the field (for holes). This
charge movement in the presence of electric field is called drift and is the primary
mechanism for charge collection. During drift the charge continues to diffuse due to
its concentration gradient, regardless of the presence of any electric field; this is
unavoidable but not always undesirable.
1-10
CMOS Image Sensors
Whenever electrons and holes are created, for example by illumination with light,
the excess carrier concentration will decay back to equilibrium after the source of e–
h pair generation is turned off. In silicon the dominant physical mechanism for the
decay is trap-assisted recombination. Direct e–h recombination occurs too, but at a
much smaller rate. The rate of decay towards the equilibrium concentration,
expressed as the change of carrier concentration per unit time U = Δn /Δt , is called
recombination rate. The simplest possible mathematical description of this process is
to assume that the recombination rate U is proportional to the excess carrier
concentration. For example, if the electron concentration in p-type silicon is np and
the equilibrium concentration is np0 , the recombination rate can be written as
U ∝ (np − np0 ). Since proportionality is assumed, we need a proportionality constant
in units of seconds, so that the recombination rate is measured in units of carrier
concentration over time (cm−3 s−1 in semiconductor physics). This constant is called
the carrier lifetime (τn for electrons, τp for holes) and can be thought of as the
characteristic time over which the carrier concentration decreases. Now, the
recombination rate can be written in its familiar form:
dnp n p − n p0
U=− = (1.9)
dt τn
We have added a negative sign in (1.9) because due to recombination the carrier
concentration decreases (dnp /dt < 0) when np − np0 > 0, and U > 0. Because np0 is
constant, we can write that
d (n p − n p 0 ) n p − n p0
=− (1.10)
dt τn
and after separating the variables we can integrate both sides:
d (n p − n p 0 ) 1
∫ n p − n p0
=−
τn
∫ dt (1.11)
t
ln(np − np0) + const = − (1.12)
τn
The final equation can be written by using the initial conditions: at t = 0 the
excess electron concentration is np(0) − np0 and at t → ∞ naturally np − np0 = 0,
with only the equilibrium concentration np0 left. Therefore, the constant in equation
(1.12) must equal −ln(np(0) − np0 ) and we arrive at the time dependence of np:
t
np(t ) = np0 + (np(0) − np0)e− τn (1.13)
Equation (1.13) tells us that the electron lifetime τn is the characteristic time over
which the excess carrier concentration decreases by 1/e , i.e. only 37% of the excess
carriers remain. After three times the lifetime only 5% of the initial charge will be
left.
1-11
CMOS Image Sensors
1.3.2 Recombination
This is a good place to answer an important question: why don’t the electrons and
the holes recombine immediately after they are generated? After all, they are created
together and close to each other, and it would be reasonable to expect that they
should recombine at high rate. Fortunately, such direct (band-to-band) recombina-
tion in silicon is very rare because it is an indirect bandgap semiconductor. Carrier
lifetime controlled by band-to-band recombination is very long; in high purity
silicon the electron and hole lifetimes can be many milliseconds. This long lifetime
allows the charge to diffuse out a long distance from the place it was generated
without recombining, unless it is quickly collected with the help of an electric field.
Figure 1.8 shows two direct, and very rare band-to-band recombination mech-
anisms: radiative with the emission of a photon, and Auger recombination where the
excess energy is transferred to another carrier, such as a hole in highly doped p-type
silicon.
The third one, trap-assisted recombination via a mid-band trap, is by far the
dominant mechanism in silicon, described by the Shockley–Read–Hall (SRH)
theory [20]. Traps are produced by imperfections or impurities in the crystal lattice,
which introduce energy levels deep in the bandgap. Traps take part in both capture
and emission of carriers and are also called generation-recombination centres.
The recombination rate from a trap with concentration Nt and energy level Et
above the valence band is given by
σnσpvth(pn − n i2 )Nt
U= (1.14)
σn⎡n + ni exp
⎣ ( Et − Ei
kT )⎤⎦ + σ ⎡⎣p + n exp(−
p i
Et − Ei
kT )⎤⎦
Here Ei is the intrinsic Fermi level (approximately the mid-band energy level);
k is the Boltzmann’s constant;
T is the absolute temperature;
vth is the carrier thermal velocity;
ni is the intrinsic carrier concentration;
n and p are the electron and hole concentrations, respectively;
σn and σp are the electron and hole capture cross-sections, respectively.
1-12
CMOS Image Sensors
Comparing with (1.9) we see that the term multiplying the concentration differ-
ence is the inverse of the electron lifetime (also called recombination lifetime)
1
τn = (1.17)
σnvthNt
Typical capture cross-sections are in the range 10−16 to 10−14 cm2.
Example 1.5. Calculate the electron lifetime due to a mid-band trap (Et = Ei ) with
σn = 10−15 cm2 and concentration 1012 cm−3, using that the electron thermal velocity
is vth = 1.4 × 107 cm s−1.
Solution: Using (1.17) we get
1
τn = −15
= 71.4 μs
10 × 1.4 × 107 × 1012
To put this into perspective, trap concentration of 1012 cm−3 corresponds to an
average of one trap per cubic micrometre, or one trap per 50 billion Si atoms.
Measurements show that in high quality, lightly doped (<1015 cm−3) silicon the
minority carrier lifetime can be tens of milliseconds [22]. Considering the calculation
in example 1.5, this implies that the trap density responsible in the SRH model must
be less than 1010 cm−3, or one trap per 5 trillion atoms. As the dopant concentration
increases above 1016 cm−3 the lifetime begins to decrease and this is taken into
account as concentration-dependent SRH lifetime [23].
At high carrier density, such as along a dense ionisation track or in solar cells,
Auger and band-to-band radiative recombination can begin to limit the lifetime even
in low-doped silicon. Auger recombination involves a direct recombination between
1-13
CMOS Image Sensors
an electron and a hole, with the excess energy transferred to another electron or hole.
The Auger lifetime for high excess carrier concentration in a lightly doped semi-
conductor is [22]:
1
τAuger = (1.18)
CaΔn 2
Here Ca is the ambipolar Auger coefficient (1.66 × 10−30 cm6 s−1 in silicon [24])
and ∆n = np − nn0 or ∆n = pn − pp0 is the excess carrier concentration.
The lifetime due to band-to-band radiative recombination is given by
1
τrad = (1.19)
BΔn
where B = 4.7 × 10−15 cm3 s−1 is the radiative coefficient in silicon at 300 K [25]. It is
easy to see that the Auger and radiative lifetimes are much larger than τSRH and can
become comparable to the SRH lifetime only at excess concentration above
1016–1017 cm−3. In a typical image sensor, the excess carrier concentration due to
illumination rarely exceeds 108–1010 cm−3, therefore the direct recombination
mechanisms have negligible influence.
The total lifetime τtot can be calculated from Matthiessen’s rule for the three
recombination processes as in [19]
1 1 1 1
= + + (1.20)
τtot τSRH τAuger τrad
Equation (1.20) is analogous to the one used to calculate the resistance of parallel
resistors; physically it means that the different recombination mechanisms work
independently and in parallel.
1.3.3 Drift
We are going to consider a hypothetical collection element without specifying what
it is and how it is made; then in the following section we will talk about two real
charge collection elements—the pn junction and the MOS capacitor.
Figure 1.9 shows our hypothetical collection element. From the surface down to
depth d there is a constant electric field E , and below d the field is zero. This may
look artificial but is not far off from reality.
In this structure only the electrons are collected, and the holes are discarded never
to be seen again, as is typical for most image sensors. Electrons are preferred because
they move much faster than the holes, resulting in shorter collection times. Holes are
allowed to diffuse until they reach the backside substrate electrode, or they
recombine after travelling a long distance away from the charge collection element.
Within a region having an electric field E electrons experience a force F = −qE
and begin to accelerate. Holes experience the same force but with the opposite sign
and move in the other direction. As mentioned previosly, this movement under the
influence of an electric field is called drift. In semiconductors it is experimentally
1-14
CMOS Image Sensors
Figure 1.9. Charge collection of electrons created by a light beam and experiencing drift and diffusion.
observed that at low electric fields (<∼10 4 Vcm−1) the charge carriers acquire drift
velocity vd proportional to the electric field E :
vd = μE (1.21)
The proportionality factor μ is called mobility and is not a constant—it decreases
as the temperature and the doping concentration increase [26]. The electron mobility
μn at low fields is about three times higher than the hole mobility μp ; the values in
low-doped silicon at 300 K are μn = 1400 cm2 V−1 s−1 and μp = 470 cm2 V−1 s−1 [26].
The higher electron mobility and velocity is one of the main reasons why we prefer
collecting and transferring electrons, rather than holes.
The drift velocity decreases below the value calculated by (1.21) at higher electric
fields (above 10 4 V cm−1), and stops increasing altogether (i.e. saturates) at
E > 105 V cm−1. The saturation drift velocity vdsat for both electrons and holes is
approximately 107 cm s−1.
Drift velocity is directional and determined by the applied electric field; it is also
superimposed on the thermal carrier velocity caused by their random movement in
the crystal lattice. The thermal velocity for electrons is given by
3kT
vth = (1.22)
m 0*
where k is the Boltzmann constant, T is the absolute temperature and m 0* = 0.26m 0
is the effective electron mass [2]. At 300 K formula (1.22) gives vth = 2.3 × 107 cm s−1.
This is similar to the experimentally observed saturation velocity vdsat , and is an
indication that the simple proportionality in formula (1.21) is valid only when
vd ≪ vth .
1-15
CMOS Image Sensors
Example 1.6. Calculate the electron drift velocity for E = 1000 V cm−1 and electron
mobility μn = 1400 cm2 V−1 s−1 (value for Si at 300 K and for low doping
concentration), and compare it to the thermal velocity vth .
Solution: From formula (1.21) we get
vd = μnE = 1400 × 1000 = 1.4 × 106 s−1
This velocity is about 20 times lower than the random thermal electron velocity.
Knowing the drift velocity allows us to calculate the time it takes to collect the
charge. For simplicity we can consider that vd is much smaller than the saturation
velocity. The travel distance x is simply the velocity multiplied by the time:
x = vdt = μnEt (1.23)
The charge collection time t is the device thickness divided by the drift velocity,
and substituting the drift velocity from (1.21) we arrive at:
d d d2
t= ≅ = (1.24)
vd μnE μnV
Here we have used that the electric field is the applied voltage V divided by the
thickness d . This is an approximate and simple, but very useful formula; we will
refine it further in the following sections.
Example 1.7. Calculate the charge collection time in silicon with thickness d = 5 μm
and applied voltage across itV = 1 V. The electron mobility is μn = 1400 cm2 V−1 s−1.
Compare with the charge collection time under velocity saturation (vdsat ≅ 107 s−1).
Solution: First, we need to see how the electron drift velocity compares to the
saturation velocity. Using (1.21)
V 1
vd = μnE = μn = 1400 × = 2.8 × 106 s−1
d 5 × 10−4
we see that it is about a factor of 3 lower than vdsat , therefore formula (1.24) can be
used and gives:
(5 × 10−4)2
t= = 0.18 ns
1400 × 1
The collection time under velocity saturation is calculated as
d 5 × 10−4
t= sat = = 0.05 ns
vd 107
1-16
CMOS Image Sensors
This example shows that for the typical sensor thicknesses the charge collection
time by drift is very short and may become an issue only if very fast operation is
required. However, in a much thicker sensor, made so that it can have higher
absorption at near-IR wavelengths, the charge collection time could be substantially
longer due to the quadratic dependence on the device thickness.
1.3.4 Diffusion
Charge generated in a field-free semiconductor diffuses out from the point at which
it is created and can travel large distances. It can reach a region with an electric field,
where it is swept away, or it can recombine with the assistance of bulk or surface
traps.
If we generate a point-like sphere of electron–hole pairs, they will expand
stochastically in a cloud described by the Gaussian distribution. After time t , the
RMS cloud radius rn for electrons and rp for holes in one dimension is given by:
rn = 2Dnt (1.25)
rp = 2Dpt (1.26)
where Dn is the diffusion coefficient for electrons and Dp for holes, measured in cm2 s−1.
The diffusion coefficients are connected to the mobility via the Einstein relationship:
kT
Dn = μn (1.27)
q
kT
Dp = μp (1.28)
q
The diffusion radii are the standard deviations of the charge density spread; from
the properties of the Gaussian distribution, we know that 95% of the charge is
contained within radius 2rn for electrons and 2rp for holes.
Figure 1.10 shows the diffusion spread with time of a point-like charge generated
at t = 0 without any recombination. The electron density is described by a Gaussian
with standard deviation given by (1.25), with the peak moving by a distance x
determined by (1.23) when the electric field is not zero.
Electrons spread 3 times faster than holes due to their diffusion coefficient being
three times higher; the values can be calculated from (1.29) and (1.30) and are
Dn = 36 cm2 s–1, Dp = 12 cm2 s–1 at 300 K.
During diffusion the charge carrier concentration decreases because they spread
out; at the same time their concentration decreases also because they are subjected to
various recombination processes, with their combined influence reflected in the
carrier lifetime. The longest distance the carriers can travel is naturally limited by
their lifetime and is called diffusion length. It is an important parameter in
semiconductors and enters numerous formulas describing image sensor operation.
The diffusion length L n for electrons is defined by:
1-17
CMOS Image Sensors
Figure 1.10. Electron spread only due to diffusion (at zero electric field) and due to drift and diffusion with
electric field = 500 V cm−1 (after [2], p 55).
Ln = Dnτn (1.29)
and a similar expression can be written for hole diffusion length Lp. The diffusion
length is usually much larger than the typical pixel sizes due to the long carrier lifetime.
Example 1.8. Calculate the diffusion length for electrons in silicon with electron
lifetime τn = 1 ms (fairly typical for low-doped, high quality epitaxial silicon). The
diffusion coefficient is Dn = 36 cm2 s−1.
Solution:
Ln = 36 × 10−3 = 1897 μm
This is a really long distance; some image sensors are physically smaller than this!
1-18
CMOS Image Sensors
Since charge travels so well on its own, can we collect it using only diffusion? The
short answer is ‘No’—collection entirely by diffusion can lead to significant losses
because the charge is not going to stay in the confines of the pixel where it was
generated. Diffusion is isotropic, and charge moves in all directions. To capture it,
we need the collection element to surround the charge on all sides, or a special
structure that forces the charge to go in one predominant direction.
Drift is a far better choice for charge collection because it is directional towards
the source of electric field. It is also much faster and minimises the chance of charge
loss. The distance travelled under drift (1.23) is proportional to time, while the
diffusion radius (1.25) increases much slower as a square root. Also, electrons move
three times faster than holes in electric field and not only 3 times faster as in
diffusion.
Example 1.9. Calculate the electron collection time if the charge moves only by
diffusion for the values in example 1.7, assuming that the charge is forced to travel in
one direction, and there is no charge loss.
Solution: Using equation (1.25) we can calculate the time it takes the electrons to
travel the longest distance, equal to the depth of the sensor rn = d = 5 μ m :
rn2 (5 × 10−4)2
t= = = 3.5 ns
2Dn 2 × 36
Because of the stochastic nature of the diffusion this time can be considered as a
time constant; three time constants would suffice for 95% charge collection, and
equals to 10.5 ns. This time is two orders of magnitude longer than in charge
collection by drift.
1-19
CMOS Image Sensors
Figure 1.11. Charge transfer in image sensors: (a) single transfer; (b) multiple transfers with charge collection
elements capable of charge transfer; (c) multiple transfer with dedicated transfer elements which do not collect
charge.
Figures 1.11(b) and 1.11(c) show examples where the charge travels larger
distances, and the transfer is done is many steps. The transfer elements can either
be the same as the charge collection elements, as in full frame CCDs, or they can be
dedicated to charge transport only, as in interline transfer CCDs.
1-20
CMOS Image Sensors
Figure 1.12. Charge conversion types: (a) continuous current on a resistor; (b) on the effective capacitance of
the charge collection element; (c) using a charge-sensitive amplifier; (d) non-destructive by electrostatic
induction.
1-21
CMOS Image Sensors
Sensitive Amplifier (CSA). Due to the negative feedback the CSA ‘moves’ the
incoming charge to the feedback capacitor and the voltage step (1.30) appears across
it, while the input stays at nearly constant voltage. The output signal can be large
because the feedback capacitor can be made very small.
Figure 1.12(d) illustrates the fourth method of charge conversion discussed here,
where the charge is sensed non-destructively by capacitive coupling. The output is
the top electrode of a floating capacitor, separated by an insulator from the structure
underneath. The collection element has capacitance to both the output and to
ground, so the equivalent circuit has two capacitors connected in series. After the
switch is disconnected, the collected charge couples by electrostatic induction to the
output electrode and produces a voltage change across the effective capacitance.
This method does not interfere with the collected charge because there is no
connection to it.
The common feature between the charge conversion methods in figures 1.12(b)
and (c) is that the charge is converted to voltage on a capacitance. This may not be a
physical capacitor as the feedback capacitor in figure 1.12(c), but an effective
capacitance to ground formed by the structure of the charge collection element.
The conversion to voltage is characterised by the conversion gain Gc , expressed as
the voltage change at the output per one collected electron. Using (1.30), the
conversion gain is written as:
q
Gc = (1.31)
C
where q is the elementary charge and C is the conversion capacitance. The term
charge to voltage factor (CVF) is also frequently used as a synonym for the
conversion gain.
1.6 pn junction
The pn junction is arguably the most important element in image sensor technology.
It can be used for both charge collection and charge-to-voltage conversion and is
used in practically all image sensors. Many excellent books describing the pn
junction have been written (for example, [2] and [21]) and we are going to spend a
great deal on it too, due to its importance to image sensors.
2
Nobody is forming junctions by bringing p- and n-type silicon bars together.
1-22
CMOS Image Sensors
them neutral, and this creates a region of fixed space charge—positively charged
donor atoms on the n-side and negatively charged acceptor atoms on the p-side. This
fixed charge creates a built-in electric field which counteracts the diffusion of both
majority carriers. The same field, however, is accelerating the minority holes in the
n-side and the minority electrons in the p-side towards the opposite sides of the
junction. As the space charge region grows, the electric field increases, slowing down
the diffusion until an equilibrium is reached. The two opposing currents balance
each other for both electrons and holes, the net flow becomes zero and the width of
the space charge region settles to xn on the n-side to xp on the p-side as shown in
figure 1.13.
The concentration of majority electrons and holes in the space charge region is
nearly zero due to the electric field which forces them away. The space charge region
is depleted from majority carriers and this is why it is more often called the depletion
region. The depletion behaves essentially as an insulator, separating the electrically
neutral (and conducting due to the large carrier concentration) n and p regions on
both sides.
1-23
CMOS Image Sensors
It is this electric field that makes the pn junction so useful in image sensors.
Electrons and holes generated by the photoeffect get separated upon entering the
depletion region and create current, which can be measured. Also, only a negligible
current will flow in the absence of light due to the depletion behaving as an insulator.
This is what was described in the previous sections when we discussed the idealised
charge collection.
In equilibrium the whole of the pn junction must remain electrically neutral
regardless of the fixed space charge. If we consider an abrupt junction, where the
doping concentrations ND (in the n-side) and NA (in the p-side) are uniform and
change in a stepwise fashion at the junction at x = 0, the condition of electrical
neutrality can be written as
NDxn = NAxp (1.32)
Equation (1.32) mathematically means that in the space charge region the number
of donor and acceptor atoms per unit area is equal. Despite its idealistic appearance,
the abrupt junction is a very good approximation to real pn junctions found in image
sensors.
From the condition of thermal equilibrium, it follows that the Fermi level
through the device is constant. Therefore, the valence and the conduction bands
must bend so that the Fermi level stays flat, creating the built-in potential Vbi in
figure 1.13. Because the electric field is the gradient of the potential distribution, this
is just another way to describe the presence of electric field in the space charge
region. The built-in voltage can be calculated from the uniformity of the Fermi level
and is [2]
kT ⎛ NAND ⎞
Vbi = ln⎜ ⎟ (1.33)
q ⎝ n i2 ⎠
1-24
CMOS Image Sensors
qNAxp qNDxn
E max = = (1.36)
ε0εSi ε0εSi
The field becomes zero in the neutral silicon at x ⩽ −xn and x ⩾ xp.
Furthermore, we can calculate the potential distribution in the junction by
integrating (1.34) and (1.35), using the boundary conditions Vn( −xn ) = Vbi and
Vp(xp ) = 0. This gives the expected quadratic dependence on distance, since we are
integrating a linearly changing electric field:
qND
Vn(x ) = Vbi − ( x + xn ) 2 (1.37)
2ε0εSi
qNA
Vp(x ) = (x − xp)2 (1.38)
2ε0εSi
Here we must clarify an important point about the built-in potential—it does not
appear across the terminals of the junction. Anybody who has measured the voltage
across a pn diode with a voltmeter will testify that the voltage is zero, unless the
diode is illuminated. Obviously the pn junction is not a battery! The reason we do
not see the built-in voltage is the contact potential between the silicon and the
electrodes used to connect it to the outside world. The contact potentials between the
metal electrodes on the n and the p-side precisely cancel the built-in voltage, and
there is no potential difference between the two external electrodes. If that were not
the case, a shorted diode would generate a continuous current through itself, which
obviously does not happen.
The total width of the depleted region W = xn + xp can be found from (1.37) and
(1.38) by using that Vn(0) = Vp(0), which gives
qND 2 qNA 2
Vbi − xn = xp (1.39)
2ε0εSi 2ε0εSi
From here, using (1.32) we can write
q (NA + ND)ND 2 q (NA + ND)NA 2
Vbi = xn = xp (1.40)
2ε0εSi NA 2ε0εSi ND
and finally, we get
2ε0εSi NAVbi
xn = (1.41)
q (NA + ND)ND
2ε0εSi NDVbi
xp = (1.42)
q (NA + ND)NA
1-25
CMOS Image Sensors
2ε0εSi ⎛ NA + ND ⎞
W= ⎜ Vbi
⎟ (1.43)
q ⎝ NAND ⎠
Most of the pn junctions used in image sensors are one-sided (also called
asymmetric), i.e. one of the dopants has much higher concentration than the other.
Typical examples are n+p junctions used as photodiodes and for charge conversion.
In a one-sided n+p junction ND ≫ NA , which leads to almost all of the depletion
being on the p-side because xp ≫ xn . From (1.42) and (1.43) we see that xp ≅ W and
the expression for the depletion width simplifies to
2ε0εSi
W= Vbi (1.44)
qNA
Why are we using one-sided junctions? If one dopant has significantly higher
concentration than the other, for example by a factor of 100, the depletion width is
determined entirely by the low-doped side, and only its concentration has to be
precisely controlled. The built-in voltage depends on both dopant concentrations,
but thanks to the logarithmic dependence in (1.33) it has much weaker effect on the
depletion width.
Due to the asymmetric doping, the depletion width is almost entirely contained in
the p-side of the junction, and so is the electric field.
Example 1.10. Calculate the depletion depth of a silicon one-sided n+p junction with
NA = 6.7 × 1014 cm−3 (resistivity ρ = 20 Ωcm ) and ND = 1018 cm−3 for Vr = 0
and Vr = 5 V and 300 K. Use that ni = 1.45 × 1010 cm−3, ε0 = 8.85 × 10−14 F cm−1
and εSi = 11.9. Also calculate Emax and xn at Vr = 5 V.
Solution: First, from (1.33) we calculate Vbi :
1-26
CMOS Image Sensors
This example shows how much smaller the depletion is on the n-side in a one-
sided n+p junction compared to the p-side; in this case xn is entirely negligible.
The characteristic right-triangular shape of the electric field and its drop to zero at
the edge of the depletion are important features with implications to charge
collection.
Figure 1.14 shows the calculated potential and electric field from example 1.10
using (1.34)–(1.38), and a simulation using commercial TCAD software [27].
It is useful to compare the analytic formulae with a finite element device
simulation using the same parameters; this provides a necessary cross-check even
if both cannot be made exactly the same. The matching in figure 1.14 is good,
considering the approximations in the analytical calculation and the discrete
structure of the simulation model.
The approximation of ideal abrupt pn junction assumes that the space charge
region and the majority carrier concentration have infinitely sharp edges. This is of
course an idealisation and is the reason why the electrical field falls to zero at both
ends of the depletion. In practice infinitely sharp edges do not exist and the majority
carrier concentration changes smoothly.
The finite element analysis (FEA) TCAD simulation solves the Poisson equation
without this assumption and reveals the fine details at the edge of the depletion.
Plotted on a logarithmic scale, the electric field in figure 1.15 continues to be above
zero for about a micron beyond the calculated edge of the space charge region. This
field is small but can make a sizeable effect on the charge collection time due to drift
being much faster than diffusion, as discussed in section 1.3.3.
1-27
CMOS Image Sensors
Figure 1.14. Potential and electric field calculated and simulated in example 1.10 for Vr = 5 V.
Figure 1.15. A comparison between the electric field calculated using the abrupt n+p junction approximation in
example 1.10 for Vr = 5 V, and an FEA device simulation which does not make this approximation.
The calculations in this section used the abrupt one-sided pn junction model,
which is a good approximation to most practical devices. This gives the familiar
square root dependence (1.45) of the depletion width on the reverse bias. A different
doping profile would produce a different voltage dependence, with the linearly
graded junction [2] the most notable example, giving W ∝ (Vbi + Vr )1/3.
1-28
CMOS Image Sensors
which gives
qNAd 2
Vfd = − Vbi (1.47)
2ε0εSi
Equation (1.47) tells us that the full depletion voltage is proportional to the
dopant concentration, and this is why it is easier to reach in high resistivity, low-
doped semiconductors.
Figure 1.16. Charge collection in partially depleted (a); and in a fully depleted pn junction (b).
Example 1.11. Calculate the full depletion voltage of the n+p junction in example
1.10 for a device thickness d = 5 μm.
Solution: Using (1.47) and the previously calculated Vbi
1-29
CMOS Image Sensors
For very thick devices the full depletion voltage is large and we can ignore Vbi
because Vfd ≫ Vbi .
We can now refine the treatment of the charge collection time in (1.24) for an
abrupt one-sided pn junction. Figure 1.17 shows the familiar triangular shape of the
electric field, which in full depletion is described by (1.35) with xp = W = d.
qNA
E (x ) = (d − x ) (1.48)
ε0εSi
To avoid the electric field dropping to near zero at the back of the device, the
reverse bias can be increased above that required for full depletion so that the
junction becomes over-depleted. This creates an additional electric field Eod = Vod/d,
where Vod is the voltage top-up above Vfd. Adding Eod to E(x), and using (1.47)
with Vfd ≫ Vbi the electric field becomes
2Vfd V
E (x ) = 2
(d − x ) + od (1.49)
d d
Due to the linear dependence of the electric field the electron velocity vn will
change linearly too, depending on the position in the device according to
vn(x ) = μnE (x ). The travel time can be obtained by solving equation (1.50) as in [28]
dx 2V V
vn(x ) ≡ = −μn⎡ 2fd (d − x ) + od ⎤ (1.50)
dt ⎣ d d ⎦
Here the minus sign takes into account that the electrons move opposite to the
direction of the x-axis, and we ignore velocity saturation. The solution for charge
generated at coordinate x0 < d is:
⎡ ⎤
d2 1
t= ln⎢ ⎥ (1.51)
2μnVfd ⎢ 1 −
⎢
⎣
( 2Vfd
2Vfd + Vod ) x0 ⎥
d ⎥
⎦
Figure 1.17. Electric field in a partially depleted (a), fully depleted (b) and over-depleted (c) abrupt pn junction.
1-30
CMOS Image Sensors
and shows that when charge is generated near the back of the device (x0 ≈ d) the
charge collection time tends to infinity. This is because we are considering only
charge drift and the electric field at the back of the device for Vod = 0 is zero,
therefore an electron will stay there forever and will not be collected. Fortunately,
diffusion is ever-present and takes care of this ‘unphysical’ situation—the charge will
diffuse to regions with non-zero field will be quickly swept away.
The maximum charge collection time tmax for over-depleted junction (Vod > 0)
can be obtained from (1.51) for x0 = d
d2 2V + Vod ⎞
t max = ln⎛ fd
⎜ ⎟ (1.53)
2μnVfd ⎝ Vod ⎠
Comparing (1.53) to (1.24), the differences are a factor of two in the denominator
and the logarithmic term, which can become significant for low over-depletion
voltages. The two formulas give the same result when Vod = 2Vfd/(e2 – 1) = 0.31Vfd.
Example 1.12. Calculate the maximum charge collection time for the pn junction in
example 1.11 for Vod = 1 V , ignoring Vbi .
Solution: Using formula (1.53) we get
(5 × 10−4)2 2 × 11.98 + 1 ⎞
t max = ln⎛ = 24 ps
2 × 1400 × 11.98 ⎝ 1 ⎠
However, this is wrong because we have ignored velocity saturation in the
derivation of the formula. A quick check using (1.49) for x = 0 gives
vn = 6.7 × 107 cm s−1, well above the saturation velocity. Therefore, this
calculation gives too short charge collection time; a better estimation as in
example 1.7 would give around 50 ps.
You may ask the question: do we need a specialised semiconductor structure (e.g.
pn junction or a MOS capacitor) to generate an electric field? Can we just apply
some voltage across silicon to collect the charge? Let’s investigate this, picking a
square pixel with size a = 10 μm in silicon substrate d = 5 μm thick. Applying
voltage V across the pixel will force current to flow because silicon has finite
resistance. This current will look like photogenerated signal and must be reduced as
much as possible; therefore, we have to choose intrinsic (i.e. undoped and pure) silicon,
which has the highest resistivity ρ = 230 kΩcm at room temperature. If we now apply
one volt across the pixel, the current can be calculated from the Ohm’s law:
V V 1
I= = 2
= = 8.7 nA
R ρd / a 230 × 10 × 5 × 10−4 /(10 × 10−4)2
3
1-31
CMOS Image Sensors
Comparing with examples 1.3 and 1.4 we see that this current is many orders of
magnitude higher than the photogenerated current; obviously a pixel made like this
will be a very poor image sensor. It may have a chance in very bright illumination
conditions, or when the sensor is cooled down (to increase the resistivity and reduce
the current), but not as a normal image sensor we are all used to.
1.6.4 Junction capacitance
Besides for charge collection, the pn junction can be used for charge-to-voltage
conversion as in the diagrams shown in figures 1.12(a)–(c). Two of the circuits rely
on the conversion of the photocurrent on an external resistor (figure 1.12(a)), or on
an external capacitor (figure 1.12(c)). The diagram in figure 1.12(b) shows the most
popular use, where the junction capacitance itself is used for the conversion.
As mentioned before, the depletion region is an insulator separating the conducting
p and n-type field-free regions. This is exactly the situation in a parallel plate capacitor
with a distance between the electrodes W and capacitance C = ε0εSiA/W , where A is
the electrode area. Using (1.45) the capacitance of the pn junction is
ε0εSiA ε0εSiqNA
C= =A (1.54)
W 2(Vbi + Vr )
Example 1.13. Calculate the capacitance of the pn junction with an area A = 25 μm2
with the parameters given in example 1.10 for Vr = 1 V .
Solution: From (1.54)
8.85 × 10−14 × 11.9 × 1.6 × 10−19 × 6.7 × 1014
C = 25 × 10−8 × = 1.42 fF
2 × (0.74 + 1)
After the junction has been biased to Vr and left floating by disconnecting the
switch, photogenerated electrons will collect at the cathode and reduce its potential
according to formula (1.30). However, because the junction capacitance depends on
the applied voltage, it is not an ideal parallel plate capacitor. As electrons are
collected, the voltage on the junction goes down, and the capacitance goes up. The
change of the capacitance is nonlinear, and its dependence on the reverse bias can be
found by differentiating (1.54) by Vr :
dC A ε0εSiqNA C
=− 3
=− (1.55)
dVr 2 2(Vbi + Vr ) 2(Vbi + Vr )
The nonlinear capacitance means that the conversion from charge to voltage will
be nonlinear too, which is not what we normally want from an image sensor. As
figure 1.18 shows, the change of capacitance can exceed 20% when the voltage across
the junction changes by 1 V. This change is large but can be counteracted by
connecting the junction in parallel with a larger, linear capacitance, so that the
nonlinearity is much reduced. In image sensors the role of this additional capaci-
tance is performed by the readout circuitry, as well as by actual capacitors.
1-32
Random documents with unrelated
content Scribd suggests to you:
"Kiitoksia paljon; mutta antakaa minulle enemmin reikäleivästä
tehty voileipä ja hyvää viilipiimää."
"No, tuopa vain ei näy ymmärtävän, mikä hyvää on! Heti uuniin,
poika, jo on rauta kuuma!"
"Sinä olet liian kuuma, poika parka, liian kuuma. Minä olen
lumikuningas ja kasvatan kaikki alammaiseni jäämöhkäleiksi, ja minä
teen sinutkin jäämurikaksi. Ylilumiukko, pistäppä poika seitsemän
kertaa jääkylmään veteen ja ripusta hänet johonkin naulaan
jäätymään!"
Knut tunsi suussansa pillin ja alkoi puhaltaa puu puu. Kuusi alkoi
haukotella, ojensi oksiansa ja mutisi jotakin päivällisunen
häiritsemisestä sekä heittäytyi pitkäkseen sammalikolle, musertaen
allensa sen suuren kuningaskotkan. Knut hiipi pois oksien välistä,
kuuli metsän kuorsaavan ja läksi juoksemaan pakoon, min enimmän
kerkesi. "Nyt minä kyllä pidän varani", ajatteli Knut. "Näkyypä täällä
olevan vaarallista metsässä; no, astunpahan ennemmin rantatietä
veden vieritse."
"Jospa minulla nyt olisi kaikki litteät kivet, joita olen viskellyt pitkin
veden pintaa, ja ne olisivat oikeita voileipiä, niin miten minä söisin!"
ajatteli Knut itsekseen.
"Eikö ole?" virkkoi Knut ja otti pillin taskustaan. Samassa tunsi hän
suuren aallon vierivän päänsä ylitse, niin että hän kaatui pitkäkseen
vuorelle. Enempää hän ei tiennyt, ennenkuin jälleen heräsi vuorella,
ja silloin aurinko jo oli lännen puolella, niin että arvattavasti oli jo
iltapuoli päivää.
Ylioppilas Keikka,
Kas, tutkivi kreikkaa.
Hyvää yötä! ja soma
Näe armahas oma!
Hotellissa tässä
Taas vierahat mässää.
Lasi punssia mulle!…
Jopa mar sitä tullee!
Pahin oli pitkä Penttu. Hän narrasi Paavoa, ja kun Paavo sattui
näkemään poikain lyövän kiekkoa maantiellä, määki Penttu "pää
pää!", ikäänkuin Paavo olisi ollut pässi. Se oli kylän lapsista hyvin
sukkelaa, mutta niin ei ajatellut pikku Liisu. Hän vei Paavon
kanssansa tupaan ja lohdutteli häntä viilipiimällä, ett'ei hän rupeaisi
itkemään.
Kello oli kohta 8 aamua, mutta vielä oli aivan pimeä ulkona, ilma
oli kylmä, tähdet välkkyivät kirkkaina taivaalla. Silloin Paavo,
ajettaessa jäätä myöten hiljaisten vesien yli, kuuli kaikkein kauneinta
soittoa, kuin kukaan ihminen on kuullut aina ensimmäisestä
jouluaamusta asti, jolloin paimenet kuulivat enkelein laulavan
Betlehemin edustalla. Mitähän se oli? Paavo ei sitä silloin tiennyt,
hänhän ei ollut vielä koskaan kuullut mitään soittoa eikä laulua;
mutta sittemmin hän sen kyllä ymmärsi. Kointähdet siinä ylistivät
Jumalaa.
Sen kuuli Paavo, mutta ymmärsikö hän sen myöskin? Ei, hänellä
oli nälkä, hän ajatteli tuoreita jouluvehnäsiä, jotka niin makealta
tuoksusivat uunissa eilen, ja mietiskeli, annetaankohan hänelle oikein
paljo siankinkkua päivälliseksi. Kun kotiin saavuttiin, paistoi äiti eilistä
puuroa aamiaiseksi, ja Paavo kuuli puuron sanovan puulusikalle: "älä
kaikkea syö, jätä vähän isällekin!"
Kun hänellä oli oma työhuone, arveli hän, ett'eipä hän suinkaan
voisi saada parempaa vaimoa kuin Liisu, jos vain Liisu huoli hänestä
raukasta, joka oli kuuromykkä. Mutta kas, Liisu ei ollutkaan
taipumaton tulemaan hänen vaimoksensa; hän kyllä tunsi Paavon ja
tiesi voivansa tosin saada miehen, joka olisi vähemmin vaiti, mutta ei
parempaa miestä. Hyvin onnellinen pari heistä tulikin. Heidän
lapsensa osasivat kaikki kuulla ja puhua, ja kun heidän
vanhempansa ja sisaruksensa usein kävivät heitä tervehtimässä, oli
Paavon äitillä tapana sanoa pikku pojalle, kun näki paistettua puuroa
pöydällä: "ettekö kuule, kun puuro sanoo lusikalle: älä syö kaikkea,
jätä vähän isällekin!" Silloin nauroi Paavo ja merkitsi, niinkuin hänen
äitinsä oli ennen muinoin merkinnyt hänelle: "sen puhui puuro hyvin
viisaasti!"
No, mitäpä nyt enää olisi kertomista Paavosta? Se, että kun
ihminen oikein sydämmestään jotakin rukoilee Jumalalta Jesuksen
nimeen, niinkuin Jumalan sana meitä opettaa, niin saa hän aina
varmaan uskoa, että Jumala kuulee hänen rukouksensa. Mutta
Jumala ei kuule meitä aina, niinkuin itse ajattelemme, vaan antaa
sen sijaan jotakin vielä parempaa. Eihän ketään kummastuta, että
kuuromykkä poika raukka rukoilee Jumalaa opettamaan häntä
puhumaan ja kuulemaan. Ne ovat kaksi suurta Jumalan lahjaa, joista
kaikkein, jotka ne ovat saaneet, tulee kiittää ja ylistää Luojaansa.
Mutta Jumala antoi sen sijaan Paavolle jotakin, joka oli vielä
parempaa kuin kuulo ja puhe: kuulla äänettömyyden puhuvan, joka
Paavolla merkitsi samaa kuin alinomaa ja minkään häiritsemättä
kuunnella omantunnon ääntä. Kun Paavo sellaisesta Jumalan
armosta tuli hyväksi ihmiseksi, oli se hänelle parempi kuin kuulla ja
puhua kuinka paljon hyvänsä maailmassa ja tulla rentuksi kuin
Penttu.
"Pankaa suunne lukkoon!" torui Sipi Ilonen taas. "Minä sanon, että
nyt tulevat englantilaiset ja kanuunat paukkuvat; ell'ette sitä kuule,
niin ainakin näette kanuunat. Nyt kaalaa pikku Napoleon koko
armeijansa kanssa veteen…"
"No, ell'ette näe vettä, niin kuulette sen ainakin", sanoi Sipi Ilonen
äkeissään. "Huh, miten meri pauhaa ja aallot kohisevat ja laivat
varustautuvat taisteluun! Ja nyt keisari Napoleon sieppaa yhden
linjalaivan kumpaankin kainaloonsa ja marssii rannalle."
"Ei, nyt minä lähden tieheni", sanoi suutarin Leena. "En minä kärsi
tuota nähdä, hän aivan halkeaa."
Riks! raks! Siinä makasi kaappi maassa. Sipi Ilonen aivan hurjistui.
"Kuka uskalsi sysätä minun kaappini kumoon?"
"Ottakaa kiinni varas!" kiljui Sipi Ilonen. "Antti, Antti, vanteet ovat
pudonneet Suurelta Mogulilta ja hänen äitinsä on katkennut aivan
keskeä poikki!"
Sipi Ilosta harmitti, että kuu alinomaa vain katseli häntä eikä
antanut hänen nukkua. "Miksi minua yhä katselet?" kysyi hän.
ebookbell.com