IEEE TRANSACTIONSON ELECTRON DEVICES, VOL. ED-14,NO.

3, MARCH 1967 117

The Resonant Gate Transistor

HARVEY C. NATHANSON, MEMBER, IEEE, WILLIAM E. NEWELL, SENIOR MEMBER, EEE,
ROBERT A. WICKSTROM, AND JOHN RANSFORD DAVIS, JR., MEMBER, IEEE

Abstract-A device is described which permits high- Q frequency =characteristic pull-in voltage
selection tobe incorporated into silicon integrated circuits. It is =channel pinch-off voltage
essentially an electrostatically excited tuning fork employing field-
effect transistor “readout.” The device, which is called the resonant
=threshold voltage
gate transistor(RGT), can bebatch-fabricated in a manner consistent =channel width
with silicon technology. ExperimentalRGT’s with gold vibrating = oxide thickness1
beams operating in the frequency range 1 kHz <fo <loo kHz are =Young’s modulus
described. As an example of size, a 5-kHz device is about 0.1 m m =beam thickness
long (0.040 inch). Experimental units possessing Q’s as high a s 500
= drain signal voltage
and overall input-output voltagegain approaching +lo dB have been
constructed. =sinusoidal input voltage of frequency w
The mechanical and electrical operation of the RGT is analyzed. =sinusoidal output voltage of frequencyw
Expressions are derived for both the beam and the detector char- =resonant frequency
acteristic voltage, the device center frequency, a s well as the device = clamped-gate FET transconductance
gain and gain-stability product. A batch-fabrication procedure for the
RGT is demonstrated andtheory and experiment corroborated. Both
=electronic charge = 1.6 x l O V 9 coulombs
single- and multiple-pole pair band pass filters are fabricated and =drain resistance
discussed. Temperature coefiicients of frequency a s low as 90- =frequency stability factor
150 ppm/”C for the finished
batch-fabricated
device were =sinusoidal resonator deflection
demonstrated. =equilibrium spacing between resonator and
GLOSSARY substrate
60 =equilibrium spacing when Vp = 0
= area esi, eoz, E O =permittivity of silicon, silicon dioxide, and
=capacitance between input force plate and air, respectively
resonator = e0A/lie
P =material
density
=capacitance between force plate and sub-
Ps =substrate resistivity
strate
P = amplificationfactor of equivalent clamped-
=damping constant gate F E T
=force
Pn =channel mobility
=gain-frequency stability product
4 P =bulkFermi level in substrate = (Ef -Ei)
=drain current wo = 2rf0=angular mechanical resonant fre-
=channel pinch-off current quency
=dynamic spring ((constant”
Wr = d K / M = angular resonant
frequency
=mechanical spring constant
=beam length I. INTRODUCTION
=channel length
=equivalent mass and damping constant of
resonator
=substrate doping
E
XPLOITATION of the uniquecapabilities
silicon integratedcircuits in onemajor area-
namely,digital circuits-is at a stagewhere
further progress will be self-sustaining. There are other
of

=quality factor of resonator = K / w D areas, however, which could similarly benefit from these
= beam-induced channel charge
capabilities, but where the application of integratedcir-
=surface depletion region charge cuit technology hasnot progressed as rapidly. The
=built-in oxide charge classic problem which has hindered the entry of inte-
=load resistance grated circuits into many nondigital systemsis the lack
=channel voltage of a compatibletuningelement. T h e difficulties en-
=polarization voltage countered in trying to obtain high-Q tuned integrated
circuits are bv now well known [11. L A

Manuscript received October 3 , 1966; revised December 19, 1966 A practicalintegratedtuning device must satisfy
The work reported here was supported in part by theU. S. Air Force
Avionics Laboratory, Electronics Technology Division, Wright- various constraints such as small size, capability of
Patterson AFB, under Contract AF-33(615)-3442. high-Q, and the possibility of batch-fabrication. Reason-
The authors are with the Westinghouse Research Laboratories,
Pittsburgh, Pa. manufacturing
able tolerances areand
inherent
also an
118 IEEE TRANSACTIOKS O K ELECTRON
MARCH
DEVICES. 1967

unavoidable problem which must be giventhorough Figure 1 showsonepracticalgeometry which is

consideration in all high-Q tuned circuits, as may readily presently being investigated. A description of detailed
be seen from a simple example. If a Q of 100 is required device operation has been published elsewhere [4]. In
in a typical application, the corresponding passband is Fig. 1 a metal beam electrode, clamped on one end to
only 1 percent of the resonant frequency. Therefore, a an insulating oxide, is fabricatedparalleltoandsus-
variation of the order of 1 percentin any parameter pended over the surface of a silicon slice. Underneath the
which significantly influences theresonant frequency beam at its end is an insulated input force plate.
will shift the desired signal completely out of the pass- Voltages applied to this plate exert electrostatic forces
band. Conventional tuned circuits overcome the toler- on the beam electrode causing it to vibrate. Onlyat the
ance problem throughthe choice of onlyverystable mechanical resonance frequency of the beam is the vibra-
components for the critical elements and the incorpora- tion appreciable. Vibrations of the beam are detected
tion of adjustable components (e.g., the trimmer capaci- as variations of field-effect induced charge in the channel
tors on a n I Ftransformer) to permitperiodic compensa- region of an MOS-type detector underneath the middIe
tion for any residual drift. On the other hand, diffused of the beam. Output is extracted at the “drain” of the
components in silicon integrated circuits are noted for device. A polarizationvoltage Vp controlsboththe
relatively poor stability and the incorporationof adjust- magnitude of the signal force and the magnitude of the
able components (to say nothingof the actualprocess of induced detector charge, making the gain of the device
adjustment) defeats many of the desired advantages of proportional to Vp2. Figure 2 shows a photomicrograph
integrated circuits. Therefore, i t is mandatory that the of anactualRGThaving aclamped-clamped beam
moststableandmostcontrollableparameters of an resonator 0.5 mm in length, along with the tuning curve
integrated circuitbe used as thecritical tuning elements. of this structure. The fundamental resonant frequency
The toleranceproblem is furtheraggravated if the of this device is about 60 kHz, and Q’s of 100 to 500
resonant feedback loop includes an active element be- have been obtained.
cause Q is then also extremely sensitive to component In this paper, a simplified and generalized model of
tolerances. In fact, a givenpercentagevariation in a the RGT will beanalyzed,andthe work to date on
component often causes a variation in Q which is of the practicaf devices will be summarized. The presentation
order of Q times as great [2]. is directed to those who seek to understand the opera-
The use of a passive mechanical resonator seems to tion of the RGT aswell as its capabilities and limitations
offer the most promising solution which can satisfy all in integrated circuits.
of these constraints [3 1. The Q of these resonators is
determined by losses and should be relativelystable.
11. THEORETICAL
The resonant frequency is determined primarily geo- by
metrical dimensions, which are among the easiest A . Electromechanical Analysis of a
parameters to control accurately in integrated circuits. Simelified Device Model
Therefore, although the tolerance problem is not Once the basictechnology of batch-fabrication of
avoided, it is at least minimized. suspended,mechanically resonant beams is mastered,
The resonant gate transistor (RGT) is a new, novel, a wide variety of geometries will be feasible for the
mechanically resonant tuning device which is compati- RGT. Formaximum usefulness, the initialanalysis
ble with silicon integrated circuits and which promises should therefore be as general as possible, even if some
to solve the tuning problem over the frequency range accuracy must be sacrificed. The first simplification de-
from about 0.5 kHzto 1 MHz [4], [SI. Thethree pends on the fact that many resonant structurescan be
essential elements in the RGT are asfollows: adequately represented by a lumped mass and spring
1) An input transducer to convert the input elec- having the same resonant frequency, the same effective
tricalsignal into a mechanical force. Thepresent spring constant, and the same Q.
RGT utilizeselectrostaticattraction,althoughde- A second important consideration is that the mathe-
vices using the piezoelectric effect, magnetostatic matics of this initial analysis should be kept as simple
attraction, or other effects instead can be imagined. as possible to avoid obscuring the basic simplicity of the
2) A mechanical resonator which is sufficiently iso- RGT. Theworst mathematical complicationin an accu-
lated from its surroundings so that the desired Q can rate analysis arises because, during beam deflection, the
be obtained. A variety of singly or multiply resonant spacingbetween the resonator and the substrate is a
suspended beam resonators are possible. function of position along the beam. Therefore, an effec-
3 ) An output transducer tosense the motion of the tive value of spacing which is independent of position
mechanicalresonator and generate a corresponding has been used in the analysis.
electrical signal. The present RGT uses a field-effect The resulting model of theRGT which has been
transistor (FET).The possible use of a bipolar analyzed is shown in Fig. 3 together with the required
transistorhas been analyzed (see Appendix I), and circuit connections. The analysis begins with the equa-
other types of transducers can be conceived. tion for the electrostatic force on the vibrating mass:
NATHANSON ET AL.: RESONANT GATE TRANSISTOR 119

Voltage

'..
Silicon substrate

Fig. 1. Geometry and circuit connections of

an RGT with a C-F resonant beam.
Oxide

"ol:"I'ei"yb,;;;@ L

.( 4
Fig. 3. Simplified model of the RGT.

Fig. 4. Equivalent circuit for the RGT.

voltage. This component leads to a dependence of 6 , on

Vp which will be discussed later. The third term is very
small in normal linear operation and will be neglected.
The second term is the signal component of force:
VpeoA
F sin w t = ___ ei, sin wt.
.-, This force causes a corresponding sinusoidal deflection
of the beamresonator, theamplitudeand phase of
which are given by the standard resonance equation:
F
-
K
6= (3)
W2 1 w
1 - -+j- -
wv2 Q wr

The deflection of the resonator is sensed by the surface

FET, which will be discussed in detail in the next sec-
tion. Suffice it to say that the variation in the polariza-
tion field caused by the vibration of a gate held a t con-
stant voltage i s equivalent to that of a conventional
(b)
FET in which thegate is fixed in position butthe
voltage changes. Therefore,
Fig. 2. Photograph and typical frequency response of
an RGT with a C-C beam resonator.

electrostatic force =
1
- - (Vp + ei, sin ot)2
acfb
-
2 86, The equivalent circuit which represents (2) to (4) is
shown in Fig. 4. The overall voltage transfer function
a t resonance with matched load is then

The first term in the bracket in (1) represents the con-

stant component of force caused bythe polarization
120 IEEE TRAKSACTIONS ON ELECTROXDEVICES, MARCH 1967

, -
b-- e "0 0.2 0.4 0.6 0.8 1.0
60
= Normalized Equilibrlum Spacing
"'
_ i Normalized Polarization Voltage
"PI
Fig. 5. Variation in beam displacement with polarization
voltage for the simplified model of the RGT. Fig. 6 . Variationinresonantfrequency with polarizationvoltage
for the simplified model of the RGT.

Away from resonance, the transfer function decreases Differentiating (6) gives
in the same way as that of a conventional L C tank
circuit.
Equation (5) can be simplified, but first we must re-
turn to consider the constant component of force in (1).
The equilibrium position of the resonator occurs when
this force is counterbalanced by the restoring force of
the spring: The variation in K alsocauses a shift in the actual
1 EOAVP' resonant frequency wr away from the mechanical reso-
net force = - ___ - KO(& - 6,) = 0. (6) nant frequency wo. According to ( 9 ) and (lo),
2 6,2

Equation (6) gives 6, as a function of V p . However, _- Wr 65

WO
there is a maximum value of Vp above which the spring
force is unable to maintain equilibrium, and the spacing
immediately goes to zero. The maximumvalue of V p
we define as the pull-in voltage V p ~Setting
. the deriva-
tive of (6) equalto zero shows that Vp = V p I when The approximateform of (11) resultsfrom using (9).
6 , = 2 6 0 / 3 . Therefore, from (6), Both the exact and the approximate forms of (11) are
plotted in Fig. 6. Although the resonant frequency ap-
proaches zero as the polarizationvoltageapproaches
V P I the
, practical tuning rangeis at most about 15 per-
cent below wo. This variationin resonant frequency with
Using ( 7 ) , (6) may be put into the normalized form voltage may be either an advantage or a disadvantage-
it provides a means for fine tuning,but necessitates
voltageregulation if ultimate frequency stability is
required.
The quantitative frequency stability with respect to
A plot of (8) is shown in Fig. 5. In normal operation,
the polarization voltage may be defined as
V p / V p r has a value in the vicinity of 0.5, and 6,/6,,= 1.
Equation (8) can then be approximated by the simpler 1 27
equation:
#=
VP am, SQ w o
Q----
vp
VPI
= ,+/:(l - $) when-
6e

60
= 1. (9)
WY dVP

Another result of the interactionbetween the electro-

static force and the spring force is that the dynamic The factor Q is included to relate the stability to the
spring "constant" is not constant, but variesywith V p . bandwidthratherthantothecenterfrequency.For
NATHANSON ET AL.: RESONANT GATE TRANSISTOR 121

1o-Q.e

fo= 0.1615 50- m

10-4-
I
100-
Y = IO" N/mZ
Assumed Values:
%= 2 x 103./5

Fig. 7. Nomograph for C-F (cantilever) beam resonator.

a typical
~.case where Vp/ V p r = 0.5 and Q = 100, for a matched load is dependent only on the amplifica-
$ = -0.135. Thus a I-percent variation in Vp would tion factor of the FET:
cause the resonant frequency to change by 7.4 percent
of a bandwidth (= 1/0.135).
Now that the pull-in voltage has been derived, the
equation for the transfer function at matched load can P
be further simplified. Substituting (7) into ( 5 ) gives _-
- for RL = rd. (14)
2
Because thegate-to-substratespacing in the RGT is
considerably greater than in a conventional FET, the
= - j - ( - ) 4Q P VP for- VP amplification factor is considerably smaller, as will be
<< 1. (13) discussed later.
27 Vpr VPI
The factor in (13) which makes a near-unity-gain
Equation (13) shows thefundamentalfactors which seem feasible for the RGT is Q. Depending on the appli-
determine thegain capabilities of the RGT. T h e j factor cation, Q may be determined by other requirements such
indicates that at resonance there is a 90" phase shift as bandwidth. However, for most applications in which
between the input and the output voltages. Because of the RGT is attractive, (j's of at least 50 are required. If
thisphaseshift,conventional 0" or 180" feedback unity gain is not feasible, the attenuation can be com-
changes the resonant frequency without changing the Q pensated by one or more extra stages of amplification.
or tending toward instability.If i t is desirableto change Moreimportantthan gain is the noise figure of the
Q bymeans of feedback, the loop must include an device. Since the overall noise figure of the RGT with
additional 90" of phase shift. amplification can never be better than the RGT alone,
Theconstantfactor 4/27in (13) resultsfrom the it is most important that surface noise and signal feed-
particular simplified geometry which has been assumed, through in the device be kept as low as possible.
and would be different for eachpracticalgeometry. In Fig. 7 we have generated a nomograph from the
However, it is useful as it standsfor estimating the order equations derived in this section. This nomograph per-
of magnitude of the gain capabilityof the device. mitsthedetermination of the centerfrequency of a
Thethirdfactor in (13) indicates that the gain is cantilever fo, the characteristic pull-in voltage V p r , and
strongly dqpendent on the polarization voltage. Typi- the gravitational beam deflection 6, of the RGT versus
cally V p / V p y = : 0 . 5 , so this factor is of the order of 0.25. the three RGT parameters L , d , and 6,.
Note that$here isa tradeoff between gain and frequency As an example of the use of these nomographs, a line
stability which entersintothe choice of the proper drawn through the points d = 10 microns and L = 0.050
polarizatiofivoltage.Increasing V P increases the gain inch in Fig. 7 intersects the fo axis at thepoint
but decreases the frequencystability.From (12) and fO-2 kHz. Extending the line to the right and pivoting
(13) it is seen that the gain-frequency stability product it at the indicated pivot line, we see that a beam-to-
122 MARCH
DEVICES,
ELECTRON
TRANSACTIONS
ON IEEE 1967

substratespacing of 5 microns results in apredicted channel of the MOS transistor, modulating the channel
V p y of about 1 2 volts. conductance and therefore the output of the device.
Closer inspection of the implicit relationshipsbetween Note that by choosing an MOS region containing a
the variables in Fig. 7 reveals that for fixed values of fo normally on channel region, it is not necessary to have
and 6 0 , a thinner,shorter beam will resultboth in a alargeinverting field on the beamelectrode, which
smaller device and in a lower value of V p r and therefore might place an unnecessary restriction on the quiescent
in a lower requiredvalue of Vp needed to achieve a mechanical operating point of the beam electrode.
given overall device gain. For instance, using the above In the order for the operating “p” of the RGT to be
example where f o= 2 kHz, A0 = 5 microns, and L = 0.020 high, it is desirable that the V I characteristic of the
inch, we can reduce l l p ~to about 4.5 volts while still detector be pentode-like, i.e., that it“pinches off.” Since
keepingfo-2 kHz. Therefore, depending on the geome- the effective surface gate of thedetector(the beam
tries which prove most practical, RGT’s are potentially electrode) is several microns away from the silicon sur-
useful on relatively low voltages. Furthermore, if the face, gate-controlled pinch-off occurs at a high voltage.
spacingcan be sufficiently reduced to yield a pull-in In order to obtain low-voltage pinch-off in these com-
voltage of one volt or less, it is conceivable that the paratively “gateless” structures,substantialsubstrate
RGT can be made to operatesolely on contact potential pinch-off must be employed. I t will be shown that for
differences between the various electrodes of the struc- normal oxide surface charge obtained during steam oxi-
ture. For instance, if the input plate is aluminum and dation of p-type silicon, values of substrate resistivities
the beam is gold, the difference in vacuum work func- in the range 5-15 ohm-cm P can result in quite usable
tions for these two materials (-0.5 volt)provides an pentode-like detector characteristics.
effective bias of Vp-0.5 volt in (2) for the electrostatic We have already discussed the equivalent amplifica-
force. A similar work function difference exists for the tion factor p of the surface detector in dealing with the
gold beam-silicon couple encountered at the detector. gain [see (13) ] and gain-stability product [see (14) ] of
In conclusion, in this section we have presented the the RGT. In this section, we derive the fixed-gate trans-
simple design theory of the RGT. Throughout the dis- conductance g,, of the detector. The quantity g,, has
cussion we have assumed that the detector is a surface been foundexperimentallyto be somewhat easier to
FET of voltage gain p. We have derived expressions for predict and control than the product gmord=p because
the device gain and the gain-stability product based on of the marked dependence of rd on nonuniformities and
this value of p. lateral fringing effects at the semiconductor surface.
For circuit applications it is often sufficient to con- Since we assume the detector region to be “normally
sider the surface detectorasnothing more than a on” and reasonably far away from the beam electrode,
‘(black box.” However, for device design purposes, we a derivation of g,, involves the combined effect of both
treat the “substrate-controlled”detector on the RGT fixed oxide charge and substrate pinch-off.
with greater detail in Section 11-B. We derive general We assume a detection region channel W units wide
expressions for pinch-off voltage, transconductance, and by L, units long. Following a recent derivation of MOS
amplification factor for this detector, and illustratethese characteristics where substrate pinch-off was predomi-
expressions with a simple overall devicedesign example. nant [7 1, we write the mobile charge in the channel as
The surface-controlled detector represents probably the the algebraic sum of three components: 1) a charge com-
simplest, yet most effective way (see Appendix I) of ponent induced by the beam field, QB (coulombs/cm2);
detecting beam vibration consistent with silicon-based 2) a charge component due to the built-inoxide charge,
technology. Q s ; and 3 ) a subtractive charge due toionized depletion
region acceptors, QD:
B. The Field-Efect Detector
The mechanical vibration 6 of theresonant beam QmobiIe = QB - QD 4-Qs (coulombs/cm2). (15)
electrode as a function of input frequency is given by Relatingchargedensitiestoappropriatemechanisms
( 3 ) . There are a number of ways in which this vibration [ 7 ] ,we find
can be used to effect a voltage output. Since the beam
is already polarized to a voltage Vp to provide an input
force, vibrations of the beam electrode produce a vary-
ing field perpendiculartothe semiconductorsurface,
QmobiIe = €0 r p :.“‘I -
suggestingsome form of electrostatic“readout.”For 20NA
this reason, detection of beam movement in the RGT is - esq,/+ [Vu(.) -I- 2 d f QS (16)
accomplished by field-effect modulation of the channel
of a conventional, “normally on7’type of MOS transis- where Va(x)is the channel voltage,a function of position
tor [ 6 ] . Movement of the beam at constant voltage YP in the region between source and drain. Other terms are
exertsavariablesurface field perpendiculartothe defined in the Glossary.
NATHANSON ET AL.: RESONANT GATE TRANSISTOR 123

Substrate Resiiivity pS,Ohm cm

Fig. 8. Detector pinch-off voltage as a function of oxide charge and

substrate resistivity (assuming pp-500 cm2.volt-1.sec-').
Beam-to-Substrate Spacing 50. LJ
Following Shockley's gradual channel approximation Fig. 9. Fixed-beam detector transconductance as a function of
18J we set the drain current
IDequal to pinch-off voltage and beam-to-substrate spacing (assuming
W/L,= 10).

QmobiledV. Low values of V p o permit pentode-likeoperation and

thereforerelatively high detector gain without exces-
Thus, sively large supply voltages.
Wehave also includedinFig. 8 a curvemarked
'(channel strength." We have obtained this curve from
(18) by setting VPO=2+p-1 volt (a good estimate for
lightly doped substrate material). I t can be shown that
this bottom curve represents in a simplified form the
oxide channel strength in ohm .cm as originally defined
by Atalla e t al. [9]. The equation of this curve is given,
from (18), as
where, for simplicity, we define the thresholdvoltage 4es4
channel strength = ___ ohm-cm. (19)
v r f = 2 - -6,.
C S i J T
PlpQS2

I t will be remembered that the channel strength of an

€0
MOS system containing a positive oxide charge +Qs is
We allow thebeam-to-substratedistance 6, to ap- defined as the minimum p-typesubstrateresistivity
proach infinity in (17 ) , and calculate the valueof drain which will just be inverted by the presence of Qs. For
voltage at which aI,/a Vo = 0. This is the substratecon- instance, the channel strength of a system containing
trolled pinch-off voltage VPOsince the beam, being so 2 XlQ1l charges/cm2 is about 5 ohm .ern @-type.Equa-
far away, cannot itself cause pinch-off. This simplified tion (19) is important in the fabrication of the RGT in
situation is encountered when 6 , is greater than a few that for a given process-dependent Qs value, it gives the
bulk extrinsic Debye lengths. minimum value of substrate resistivity consistent with
Setting dI=/I3V D = O in (17) and assuming a t pinch-off a normally on RGT detector.
Vp0>>2+p, we find Returning to our calculation of g,,, we assume that
Qs2
Ir,, is independent of Vp within operating limits, and,
lim [vD (arDpvD=o] = ~ p =o 7volts. (18) in addition, that Vp>>Vpo. We find from (17) that g,,,
Go-+- 2qCd~ the device "fixed-beam" transconductance, equals
Equation (18) relates the pinch-off voltage V p g to the arpo- w vpo
surface oxide charge Qs (coulombs/cm2) and bulk g, = __ pn - €0 -mho. (20)
doping N A (cm-3). In Fig. 8 we plot Qs versus p s , the dVP L C 6,
silicon substrate resistivity for parameters of pinch-off In Fig. 9 we plot the transconductance of the detector
voltage. We assume a $-type substrate where , u p = 500 versus the beam-to-substrate spacing for various prac-
cm2.volt-l -s-l. As can be seen in this figure, common tical values of pinch-off voltage. As can be seen for VPO
values of inversion layer content (2-5 X10" charges. near 5 volts, ,g, is on the order of 1-2 pmho, a relatively
cm-3 require p-type substrate resistivities in the range low value due mainly to the large value of 6 o required
124 TRANSACTIONS
IEEE ON
MARCH
DEVICES,
ELECTRON 1967

The impedance level of the device is inversely propor- At the present stage of RGT technology it is hard to
tional to the pinch-off current of the detector. Using the predict what value of Q will obtain in a given device.
value VPOin (18), we obtain from (17 ) , again assuming Experience has shown that in the 1-10 kHz frequency
&>>bulk extrinsic Debye length, range, values of Q in the 50-200 range are typical for
plated gold. Assuming Q-100, we find that gmo-3.4
pmho. If we make the detector width W about 2 of the
beamlength .E, then W=0.005 inch. If the channel
length Loequals about 6 microns, then W / L , = 2 0 . From
The cubic dependence of I p o on Qs indicates the neces- Fig. 9 for ,,g (remembering to adjust for the fact that
sity of close control on the effective oxide charge if one W/L,= 20 by halving the requiredg,,), we see that V P ~
hopes to obtain predictable detector impedance levels must be about 8 volts. From Fig. 8 a substrate resistivity
in the RGT. of 15 ohm-cm $-type will result in a pinch-off of 8 volts
Assuming an expression for output impedance similar when Qs-3 X 1011positive oxide charges/cmZ, a reason-
to Shockley’s 181 it can be shown that ablytypicalvalueobtainedduringoxidation.From

E* = grneyd (2) *
(21), I p O = 2 5 (WJL,)p A or Ipo-0.5 mA. By operating
at the knee of the detector V I characteristic, it can be
shown that the required supply voltage for the 20-kQ
Since in the ordinary MOS transistor Ioad is then about 18 volts, making a common Vp and
v d supply practical.
&,E In the conclusion, Figs. 7 , 8, and 9 permit the design
pz- (reference [6]) of an RGT of a specific frequency,gain, output im-
6siWoz
pedance level, and operatingvoltage. Prime independent
it can be seen that the amplification factor of the RGT. variables are beam dimensions d and L, 60, N A , Qs, and
is on the order [(1/3) .( Wo,/6,J] times as small as the ,u W /L, (we assume a gold beam). In the next section we
of an MOS. For a typical Woz=2000 and 6 0 ~ 5 000
0 show these design equations to be in good agreement
k , a factor of 75 reduction in L./ over an MOS of equi- with experiment.
valentchannellength is predicted.However, (13) in-
dicates that the mechanically resonant beam system es- I1 I . EXPERIMENTAL
sentially amplifies the input voltage by a factor of Q at In this section, we describe a method for batch-
resonance, making overall gains on the order 0.1-1.0 or fabricatingRGT’sconsistent with silicon integrated
greater still possible. technology. We describethe behavior of a typical device
We illustrate an example of the design of an RGT. and discuss the range of parameters we have achieved
Suppose we wish to fabricate a 5-kHz RGT which will at present. We confirm some theoretical predictions of
exhibit a voltage gain of - 20 dB (loss) at a polarization Section I1 and, finally, discuss recent experimental re-
voltage of Vp= +15voltswith anoutput loadim- sults on multiple-beam coupled-mode devices.
pedance of 20 kS1. In order to achieve reasonable fre-
quency stability [see (14)], we will, in addition, require A . Fabrication of Resonant Beam Structure
that ( Vp/ Vp,)2 = 0.1 under operating conditions. Fabrication of the detectionregion of the RGTis con-
FromFig. 7 we find that, amongothers, a beam ventional. After the N+ source-drain regions are dif-
0.020 inch long by 4 microns thick will resonate a t fused and an insulatingoxide regrown, contact windows
about 5 kHz. Note that this choice of L and d is some- are opened in the oxide [Fig. lO(I)]. In Fig. 1O(II) a
what arbitrary. Any attempt at optimumdesignin- metal bilayer is deposited over the slice. The important
volvesacomplicatedrelationshipbetween the beam properties of this layer are that it 1) must adhere well
aspect ratio and its loss mechanism. Optimization will to the oxide, and 2) should be easily electroplated by
not be treated in this paper. If we choose V p r w 4 7 volts, the metal from which the cantilever beam will be made.
Fig. 7 indicates that 6 0 must be ~ 8 . microns,
5 giving In Fig. lO(II1) the input and tuning gates, as well as the
(Vp/Vp1)~=0.1. Now for an insertion loss of 20 d B with source-drain contact land areas, are formed via the de-
R L= 2 0 K , we find, from (13), position of a thinmetal flash. In Fig. lO(IV),via a
photoresist process, we deposit the metal spacer layer
onto which we will plate the beam electrode. The thick-
ness of this layer will correspond closely to the nominal
It- beam-to-substrate distance l i 0 in the final device. At this
RL
stage a photoresist layer is spun over the slice and ex-
Assuming &<<rd, we find that for a 20-dB insertion loss, posed such that theresist is removed down to bare metal
the relation between g, and Q is given by everywhere the beam anditspadareto be plated.
There is a marked flexibility in the process at this stage.
P 340 pmho Either one or a number of beam areas of varying shapes
and lengthscan be photo-delineated.Suspended me-
NATHANSON ET AL.: RESONANT GATE TRANSISTOR 125

Fig. 11. Batch-fabricated C-F RGT’s

(L=O.O17 inch).

@ Strip
frequencies of these devices, where Lg0.017 inch, de-
pend, of course, on beam thickness. By varying d , f < sin
the range from 7 through 24 kHz have been obtained.
As an example, a group of 12-kHz devices using this
S
geometry operated with a V P= + 3 0 volts and showed a
gain of -8 dB at a 30-kS2 load impedance. Q’s of about
Fig. 10. Resonant beamfabrication process. 60-IO0 were in evidence.
In an attempt maketo higher-frequency RGT’s which
approach the l/lOth-MHz rangeof operation, clamped-
chanical webs for coupling purposes can also be fabri- clampeddevices as showninFig. 2 were fabricated.
cated at this time, as will be discussed further on in this Thesedeviceshave gold beams of length 0.019 inch
section. In Fig. 1O(V) an appropriate metal is plated in with thicknesses in the range of 3 to 8 microns. The de-
the delineated areasto form the beam electrode.A1though tector in these devices is located at the center of the
we have used gold because of its easy plateability and beamelectrode, which is anantinode for thefunda-
relative absence of residual stress, a number of different mental vibration. On either side of the MOS detector
metals or alloys are available, somehaving lower in- are input force plates, which are connected together by
ternal loss and lower temperature coefficients of Young’s a shortingbandasshown in Fig. 2. A grounded N+
modulus. After plating, the top resist layer is sprayed- guard ring, an extension of the source contact, prevents
stripped, followed by a quick metal spacer etch, then a direct feedthrough of input gate signals to the output
pad-contact etch. A final rinse in a nonpolar organicsol- drain contact. Feedthroughcan occur because of capaci-
vent completes the process,resultingin a suspended tive coupling of the input gate to the relatively high
beamelectrode[Fig.lO(VI)]. sheet resistant channel inversion layer.
In a typicalrun, a 10 000 thermal oxide is grown
B . Experimental Measurements upon 10 ohmacm P chemically polished (111) silicon.
B-1) GePzeral Device Characteristics: We have thus far The oxide is opened for source-drain diffusion, and a
fabricated two general categories of devices, those with POC13, 2.5-micron, n-type diffusion is carried out at
clamped-freebeams and those with clamped-clamped 1100°C. Oxide is regrown over the n-type regions and,
beams. The latter exhibit a center frequency 6.27 times at this stage, channel content may be adjusted by bak-
higher than the former for equal beam length and thick- ing the slices in an appropriate ambient a t 350°C for
ness. times on the order of one-half hour. Channel pinch-off
The generalcharacteristics of the clamped-freede- currents in the range 0.05 through 0.5 mA are typically
vices were discussed elsewhere [SI. Insummary,the employed. At this stage, the device is ready for canti-
C-F devices previously reported had a gold beam elec- lever formation, as describedin Section 111-A.
trode 0.038 inch long and about 10 microns thick. Spac- In Fig. 1 2 we plot the output voltage of a clamped-
ingfrom the substrate was about 8-12 microns. In a clamped 30-kHz RGT versus input frequency a t a con-
typicalrun of thesedevices, the following average stant input voltage of one volt. A voltage insertion loss
parameters were observed: f o = 2.8 kHz, Q= 90, and of about -20 dB is observed a t resonance. A bandwidth
gain = -20 to -40 d B at V p = 8 5 volts. I t is interesting of 450 Hz implies a Q of 67. From the nomograph in
to note that of a run of 23 devices fabricated a t the Fig. 7, an fo of 30 kHz indicates, for an L =0.019-inch
same time on the same chip, a mean center frequency beam, a gold thickness (assuming theplated gold
deviation of +_8percent was observed. This reproduci- possesses bulkmechanicalproperties) of about 3.4
bility of center frequency is presently encouraging. microns. This implies a weight gain of 2.8 mg during the
In Fig. 11 we show a section of a slice of some morere- gold-platingoperation on a one-inch-diam slice. (The
cently fabricated C-F RGT’susing a tapered cantilever plated-to-nonplatedarearatio of the cantilever mask
design. An auxiliary MOS transistor is fabricated as a used in making these devices was 0.085.) An observed
buffer output amplifier beside each unit. Fundamental weight gain of 3.3 mg after plating the slice from which
126 PEEE TRAXSACTIOXS O?: ELECTROX DEVICES, MARCH 196’1

this 30-kHz devicecameindicateseither a lack of

platinguniformityoverthe whole slice, or thatthe
acousticalvelocity 4-p of the plated gold is only
about 85 percent of the accepted bulk gold value.
Operation of C-C RGT’s at higher-order modes has
also been demonstrated. For instance, a 21-kHz RGT
has been showntoexhibittuningpeaks a t 21, 72.4,
135.2, 215, and 410 kHz, in fairagreementwith the
modal ratios 1:2.7:5.4:8.9 . . . predicted by Rayleigh
[lo] for C-C beams. Deviations from the Rayleigh
ratios are duein part tochanges in the effective value of
V p lassociated with the different modes. This leads to a
different percentage “detuning” a t constant V pfor each
mode [see ( l l ) ] , modifying theexperimentallyob-
servedratios. Because of thesymmetricgeometry of
the C-C RGT series shown in Fig. 2 , gains in the third
mode (5.4 fO) usually exceed the gainobserved in
second-mode operation (2.7 fo). A nonoptirnized “gain”
of -17 dB has been seen in the third mode at a fre-
quency of 250 kHz in a L = 0.019-inch device.
B-2) Amplitude of Beam Vibration: I t is possible to
obtain a figure for the sinusoidal beam vibration neces- I I \ I ! I l l l ! l
10
sary to obtain a given output voltage from these C-C 29 30 31
Input Frequency, KHZ
devices. This is important since nonlinearities may re-
sult if, for reasonable output voltages (say, 100 mV), the Fig. 12. Response characteristic of a 30 kHz C-C RGT.
amplitude of beam vibration at resonance is an appre-
ciable percentageof 60. ,g should be about 5.2 pmho. The calculated amplifica-
Consider a 40-kHz C-C RGT which possesses a drain tion factor is then
resistance of105 ohms, a beam-to-substrate spacing of 5
p = ,,g rd = (5.2 X 10-6)(105)-= O S 2 (26)
microns, and an insertion loss of about 6 dB when oper-
ated in a normal manner with V P =60 volts, Vo =60 which compares favorably with the experimentally ob-
volts, and R L = 3 6 K. Now, by applying a nonresonant served value of p = 0.7 seen above.
ac input signal directly to the cantilever beam (while 23-3) Variation of Gain and Resonant Frequency with
holding thedc beambias a t +60 voltstomaintain Polarization Voltage: Equations (11) and (13) show that
quiescent operation), itis possible to obtain the value of the resonant frequency and gain of an RGT can be ex-
the amplification factor for the “fixed-gate” MOS tran- pected to varywith polarization voltage. These relation-
sistor [see (4) 1. The measurement is made atf<<fo. ships were substantiated by measurements on an RGT
For this particular device, a “fixed-gate” p of 0.7 was which had a resonant frequency near 32 kHz. The ge-
obtained at 0.1 through 5 kHz. From (4) it can be seen ometry of this device is as shown in Fig. 2, where the
that the beam movement 6 necessary to obtain an out- beam dimensions are length = 19 mils, width = O S mil,
put of 100 mV is given simply by and thickness =r_ 3 microns. The beam-to-substrate spac-
ing is approximately 4 microns. Based on these values,
+
a=-- rd RL 6, eout = 100 3. 36 5 X 1 the nomograph in Fig. 7 predicts a pull-in voltage of
about 75 volts.
RL ~ V P 36 0.7 X 60 lo Experimental measurements were made using a con-
= 450 A. (25) stant input signal of 0.965 volt and a 40 000-ohm load
resistor. The results are plotted in Fig.13. I t can be seen
The inherently small displacements of the beam elec- that both resonant frequency and gain vary approxi-
trode in this 40-kHz deviceindicate that creep and mately parabolically with polarization voltage, as pre-
fatigue processes which occur in the flexure of ordinary dicted by (11) and (13). Quantitative correlation of the
metals should not be severe in the RGT. Only long-term theoryandthemeasurementsdepends on the pull-in
life tests (which have not yet been made) can confirm voltage, which is somewhat risky to obtain experiment-
this hypothesis. ally because of possible damage to thedevice. However,
AS a check on the experimentally determined value of if the pull-in voltage is 7 5 volts, (11) predicts a 10-
p, we note that the MOS detector on this RGT had a percent shift in frequency at a polarization voltage of
pinch-off voltage of V p o = 7 volts.Correcting for the
fact that W / L , in this device is 21, we find from Fig. 9 75 X 0.822 = 61 volts. (27)
thatthe theoretical
“fixed-beam”transconductance
NATHANSON ET AL.: RESONANT GATE TRANSISTOR 127

5- I I I I I I I

I
I
I
4-

Gain-Siabilit)
Product
3-

2-

1-

A Q = 57.2
/-
0
& P I
20
I I
40
I I
6l
I I 0 0- _-- I
80 0
v P , volts vp = rolartzauon voltage

Fig. 13. Experimental variation of output voltage and resonant Fig. 14. Experimental variation of gain, frequency stability,and
frequency with polarization voltage. (Outputandcenter fre- gain-stability product with polarization voltage. [Gain-stability
quency versus V, for #148 C-C RGT.) product for 32.3-kHz RGT versus V, (#148).]

Figure 13 shows that the 10-percent shift actually oc- RGT [for instance, (13)] and the experimental result is
curred a t 47.5 volts. In view of the simplifications in- the fact that, in the simplified model presented in Sec-
volved in the theoretical model, and the inaccuracies in- tion 11, no account is taken of where the input gate and
volved in measuring the beam thickness and the beam- detector are situated under the cantilever beam. Obvi-
to-substrate spacing, this agreement is quite satisfac- ously, if theinputgate ismadequiteshortand is
tory. crowded very close to the clamped end of the beam, the
In Fig. 14 the gain, stability, and gain-stability prod- gain of the RGT will be severely reduced although this
uct have been plotted, in arbitrary units, for the above reduction will not explicitly appear in the gain expres-
device. In calculating I $1
[in (12) 3 we have used the ap- sion [see (13)]. In the simplified analysis it has been
proximate form -(1/2@(1 -fr/fo). This formas- assumed that both the input plate and the detector are
sumes parabolic
detuning. I t is exactonly when located at the antinode of the cantilever. This assump-
Vp<<VpIand is simpler to use inasmuch as it does not tion avoids the difficulties inherent in the distributed
involve a possibly inaccurate graphical determination nature of the beam deflection, but can lead to inaccu-
of the termdfT/d Vp. racies .of factors of 2 or more when precise comparisons
For Vp greater than 20 volts, the G S product is seen between experiment and theory are made. We can, of
to be constant to about + 3 percent over 1.5 decades in course, define “gain-reduction factor” which depends on
gain as predicted by (14). Quantitatively, the average gate and detector geometry in a complicated manner.
value of the G S productobtainedexperimentally for The calculation of this factor will not be discussed in
this device is this paper. Suffice it to say, the factoris less than 1 and
GS = I eOnt/ein1., I 4 I = 0.006.
is constant for a given device geometry. I t is about 0.5
for the C-F device illustrated in Fig. 11. Measurements
When we calculate p from G S takingintoaccount of the ratio of the beam versus input gatedevice gain in
the presence of a parallel load reactance of - j 35 ,$a the C-C RGT in Figs. 2 and 14 indicate a geometrical
associated with the measuring equipment, we find that factor of 0.25-0.15 for these devices. In other words, the
the experimental value of p for the RGT detector is on fact that the input plates in this device are off to the
the order of 0.03. This valueof p is about a factor of 4-5 side represents a loss of about 12 dB over the ideal gain
less than thetheoretical value of p calculated from either expression, which accounts for thefactor of 4dis-
g,,r, [in (20)] or p = ~ ~ L ~ [in
/ ~ (2211.
~ i 6 ~ crepancy of p in Fig. 14.
It is presently thought that the reason for the dis- The “geometrical factor” can best be calculated with
crepancy between the theoretical voltage gain for the new geometriesusingbeam deflection theory. This
128 TRAKSACTlONS
IEEE OX ELECTRON DEVICES, MARCH 1967

"factor" represents an important detail describing de-

vice utility. However, it is the basic agreement of the
qualitativebehavior of the RGT withthe simplified
V "
120 PPWOC

IN) P P W Y
-3 i
vp t 12 volt
=

theory presented in this section which indicates that the

V p + 23 Volts
=
role of Vp on devicebehavior is satisfactorilyunder-
stood.
B-4) TemperatureSensitivity of ResonantFrequency:
One primary reason for going to a mechanical resonator
for achieving high-Q integrated tuning is the fact that
since a mechanical resonance is passive, it cannot oscil-
late with small changes in beam parameters. We con-
trastthestability of a mechanicalsystemwith the
stability of anactivefeedbackRC filter where high
selectivity is obtainedonly by working close tothe
verge of oscillation [2]. Here, smallchanges in loop
parameters brought on by temperature fluctuations can
easilyquench Q or, if the loop gainapproachesone,
initiate self-oscillation.
Experiments on RGT's have shown gain variations as
small as 1 percent over a 50°C range, indicating that L I
30 40
1
50
I
CO
I
70
I
80
1
gross variations of Q with temperature are probably not Device Temperature. OC
of primary importance at this stage of device develop-
Fig. 15. Experimental variation in resonant frequency with
ment. temperature, at various polarization voltages.
More important is the variation of Young's modulus
of gold with temperature. Values of (1/Y) d Y/dT of 3 ) A plot of ppm/"C versus Vp does not seem to
of about 240 ppm/"Chave beenreported for bulk extrapolate through the origin at Vp = 0 (see insert),
gold [111. Sincefo cc 47, the fundamentallower limit on but approaches about 100 ppm/"C.
frequencystability in the RGT shouldbe about 120
ppm/"C when gold cantileversare used. There is no We divide the observed variation of ppm/"C with Vp
guarantee of course that plated gold will be as good as into 1) a constant variation, -100 ppm/"C, associated
bulk gold from a stability standpoint. Better cantilever withtheintrinsicproperties of the gold beam itself
materials and/or temperature compensation of V pcan, (probablyduetoYoung'smodulus)andindependent
however, substantially improvethis figure. The pos- of Vp;and 2 ) an "electrostatic" contribution to instabil-
sibility of electroplating beams of low-temperature co- ity varying as Vpn where n = 2 at low Vp.The exact
efficient alloys such as Ni Span-C1 is still an open ques- reason for a temperature instability varying as Vp2 is
tion because of the control in percent Ni content needed not known a t present. It is thought that eitherreversible
to hold (1/ Y)d Y/dT at below 10 ppm/"C(-27 percent stress relief resulting in changes in 6, or the stronger
A $ percent). variation of 6 , (and therefore fo with Yas V p approaches
In Fig. 15 we illustrate a representative series of fo V P ~may ) be causing this effect. Further experimenta-
versus temperature curves made on a nominally %kHz tion is necessary to ascertain the dominant mechanism.
C-F RGT with a gold-plated cantilever about L-0.017 In any case, it is important to note that when we
inch long and d-4 microns thick. Gold pIating was calculate Vpl for the above structure [48+ volts using
carried out at 50°C using a slightlyalkalineplating (27)], we find that when (Vp/Vp1)-0.5, the temper-
solution and dc plating techniques. We first note the ature coefficient of frequency is about 160 ppm/"C.
gross decrease inf0 with Vp, independent of temperature. This value is already within range of the stability of
This is j u s t the electrostatic "tuning" effect expressed evaporated thin-film resistors used in the production of
by (11). In addition, for a given value of Vp,the vari- thin-film oscillators on glass [12]. Preliminaryresults
ation of f o with T is illustrated over the range 30"sT are thereforeencouraging, and seIection of a better
5 80°C. beam material is in process.
In Table I we have tabulated some possible metals
Some generalizations can be made from Fig. 15: which could be employed as the beamelectrode. The
1) C-F devicesinvestigatedto date decrease in metals are listed in order of increasingsensitivity of
center frequency with increasing temperature. Young's modulus to temperature in ppm/OC. Note that
2) The percentage decrease info depends on Vp- of the metals which promiserelative ease of plating,
higher values of V p result in higher temperature co- gold is about a factor of 2 more sensitive to temperature
efficients. than, say, palladium. Current work is therefore being
directedtowardperfecting a process for obtaining
1 International Nickel Co., Inc. palladium beams for RGT's in the hope of reducing the
NATHANSON E T AL.: RESONANT GATE TRANSISTOR 129

TABLE I
CHANGESOF YOUKC'S MODULUSWITH TEMPERATURE
IN ppm/"C FOR VARIOUS
MATERIALS

dY 1dY
Y kg/mm2 - -kg/mm2/Co - - -ppm/CO
dT Yd T output
-1ZkHz
tungsten 41 500 4.2 101 -2VPP
palladium 12 400 1.3 105
molybdenum 33 600 4.3 128
tantalum 19 000 2.5 132
platinum 17 300 2.3 133
beryllium 29 000 4.1 141
iridium 54 000 10.6 196
gold 7 900 1.9 240
iron 21 500 5.6 260
rhodium 38 600 10.5 272
copper 13 100 3.9 298
manganese 20 000 6.3 315
silver 8 200 3.4 415
zinc 9 400 4.0 425
a!umi?um 7 200 3.1 430
zlrconlum 7 000 3.4 485
titanium 10 500 6.4 610
tin 5 500 7.7 1400
~ ~~

inherentfrequencyshift of the device toabout 50

PPm/"C-
B-5) Integrated Oscillator: Thus far only the simple
open-loop bandpass properties of the RGT have been
mentioned. In this brief section,oneparticularly im-
portant area of application is described-namely, the
construction of monolithic audio oscillators.
I t will be recalled that the integration of audio-fre-
quency RC oscillators on a silicon slice using conven-
tional components has been impractical because of the
high RC products needed to obtain the necessary phase
shift to permitoscillation. Using the RGT, a number of
low-frequencymonolithicoscillators are feasible. The
only supplementary componentsneeded are the load re-
sistors of the FET. The relatively low value of the load
resistors and the elimination of any capacitors needed
in the device results in a low-frequency integrated oscil-
lator that is compatiblewithcurrentsemiconductor
fabricationtechniques.
Because theoutput of theRGT is in quadrature
(90") with the device input at resonance [see (3) 1, it is
necessary to introduce an additional 270" phase shift
around the device to get it to oscillate at its open-loop
frequency. A phase shift of 180° is easy to obtain. How-
ever, the additional90" phase shift is just as hard toob- (c>
tain, circuitwise, for the RGT as isitto obtain ina small Fig. 16. Twin RGT oscillator requiring no external 90" phase shift
network. (a) Circuit. (b) Ouput waveform. ( c ) Twin RGT oscil-
volume for more conventional RC oscillators. lator mounted on TO-5 header.
One simple solution compatible with monolithic tech-
nology is to use two RGT's tuned to the same frequency.
Since both devices exhibita 90" phase shiftat resonance, valuesquitecompatible withintegratedcircuitsare
two devices in series plus 180" additional phase shift can needed. We illustrate the output waveform of the oscil-
result in closed-loop oscillatory behavior. For the 180" lator in Fig. 16(b) and present a photograph of the final
phase shift, we take advantage of the phase reversal at oscillator encapsulated in a TO-5 header in Fig. 16(c).
the output when the signal is applied directly on the This device oscillated at about 12 kHz with a voltage
cantilever rather than on the input force plate. output -2 volts peak-to-peak. Higher-frequency oscil-
In Fig. 16(a) is illustrated a simple circuit employing lators of this type (47 kHz) have exhibited output volt-
two RGT's selected to be about 12 kHz at the quiescent ages of up to 15 volts peak-to-peak a t reasonably low
voltages indicated. Note that only external resistors of distortion level.
130 IEEE TRANSACTIONS 05 ELECTROS DEVICES, MARCH 1967

An obviousproblemassociated with the above ap-

proach is the necessity of fabricating the two cantilever
frequencies to within a bandwidth of each other. Since Beam 2 -,

the two cantilevers of each oscillator are side by side

throughout the entire fabrication process, i t seems rea-
sonable to assume that they both will be exposed to the
same treatment. Once techniques improve to the point
where the beams are well matched, there should be no
reason why complex oscillators cannot be implemented nput Gate
by double-RGT techniques.
B-6) CoupledMultipoleFilters: The bandpasschar-
acteristics of the RGT are determined almost entirely
by the resonant modes of the mechanical structure of
the device. Simple cantilever beams possessing a single
fundamental resonance have been studied thus far in
this paper.In many situations,a more complex bandpass
characteristic is desirable. A telemetry filter requiring
both wide bandwidth and steep rejection slopes is one
example. By designing the shapeof the resonant member
in the RGT to resonate at two or more closely spaced
frequencies [13] we can implement sophisticated band-
pass behavior in a batch-fabricated manner.
One simple system exhibiting doubly tuned behavior
consists of two identical, loosely coupled C-F cantilever
beams. Such a system will “split” into two frequencies,
and can exhibituseful pass response.
We have fabricated C-F twin-beam RGT’s contain-
ing a coupling web near the clamped end, as diagramed
in Fig. 17(a). Under Beam 1 are two input electrodes
used to excite the beam electrostatically. Vibrations of
Beam 1 are mechanically transmitted to Beam 2 via the
coupling web, causing Beam 2 to vibrate. Under Beam
2 is the channel of the FET detector, the source and
drain contacts of which appear in the figure. Since the
input electrodes are under Beam 1 while the channel is
under Beam 2, only collective vibrational modes result
in signal transmission from input to output. When the
beams are tuned to exactly the same frequency, the two
modes of collective vibration occur 1) when both beams
go up and down together and2) when one goes up while
the other goes down. The modalfrequency is higher
under condition 2) than under 1) because the coupling Fig. 17. Doubly resonant C-F bandpass RGT. (a) Device g e m -
etry. (b) Photograph of an actual device (L=9.5 mils). (c) Fre-
web is stretched and acts as an additional spring tending quency response (experimental), where fcenter=22 kHz,band-
to “stiffen” the system. If the web lengthisadjusted
I
widths300Hz.
properly, the two frequencies associated with modes 1)
and 2) can be made about fu/Q apart. Smooth-topped
bandpass properties are then possible.
In Fig. 17 (b) we present a finished coupled RGT. This
device hastwoidenticalbeams of length =9.5 mils,
thickness-4microns,andbeamwidth ~ 0 . 5mil. The
coupling web in these devices is0.5 mil wide and 0.3 mil
long measured fromthe clamped end. Beam-to-substrate
spacings 6 0 of about 10 microns for these devices have
resulted in values of Vpr of about 300 volts.
In Fig. 17(c) we illustrate theresponse of one of these
devices when it hasbeen properly tuned. A doubIe-peak
characteristic is in evidence. Peak frequencies 275 Hz Fig. 18. Doubly resonant bandpass RGT, where fcenter-132
apart at 22 kHz correspond to a coupling coefficient of kHz, bandwidth= 1000 Hz.
STOR GATE NATHANSON
RESONANT E T AL.: 131

K = 1.25 percent. Measurement of the Q of the indivi- We have demonstrated that it should be possible to
dual peaks in these devices, obtained by detuning the makeRGT’swithtemperature coefficients of center
structure, gives a Q-130. Thus KQ-1.62. I t canbe frequencyabout half that associatedwithYoung’s
shown classically that a (K@)2= (1.6)2= 2.6 should re- modulus for the beam material. However, if we restrict
sult in a theoretical midband rippleof about 1 dB, which ourselves to the plating of pure materials, 50 ppm/”C
is close tothemidbandripple (1.9 dB)observed in seems to be about the best practical temperature stabil-
Fig. 17 (c). ity obtainable, from TableI.By moresophisticated
These devices were operated at V p = - 210 volts with fabricationtechniques,such as bimetalplating, alloy
RL= 70 kQ and VD = +30 volts. An output of 87 mV in plating,orsimplytheassembly of prepunched low-
the passband with an inputof 580 mV indicates a loss of coefficient beams onto the silicon slice, it may be pos-
- 16 dB at resonance, which is the lowest we have ob- sible to increase the temperature stabilityof the device.
served for a coupled RGT to date. The properly tuned I t is also possible that diode-type voltage compensation
device was also free from spurious resonances. of V pwith temperature may also increase device stabil-
In Fig. 18 we show the response of another type of ity. The latter method is currently being investigated.
coupled RGT. This device is similar in construction to In anycase, the temperature stabilityof the mechanical
the device illustrated in Fig. 17 except that both ends of Q of the RGT already representsa large qualitative im-
the cantileversare affixed tothe oxide surface. The provement in inherent device stability over, say, theRC
center frequency of this essentially C-C device is about feedback approach to tuning. Ultimate limits of both Q
132 kHz with a bandpass of about I000 Hz. andfo stabilityin the RGT are still not understood.
I t should be pointed out that the shapeof the band- As far as reproducibility and predictability of center
pass characteristic in these devices was quite dependent frequency is concerned, much work still has to be done.
on VP.By changing V Pa few tens of volts, it was pos- We are encouraged by the approximately 20-30 percent
sible to tune theresponse of the device from 1) two well- spread in fo over any given slice, considering the labor-
separated low-gain peaks, through 2) a region where, as atory methods of fabrication we are using. On the other
the peaks came together, double-peak high-gain band- hand, no attempt has been made to predetermine the
pass behavior occurred, through a region where 3) the mean center frequency of the resonators on a given slice
peaks began to diverge again with another accompany- to say 5 percent, by current-time integration of plating
ing decrease in gain. I t is felt that the variation of the current. Theproblem of fabricating predictable and pre-
bandpass characteristics of the device with Vp is due to cise beamfrequencies is compounded when the semi-
a differential electrostatic tuning caused by the lack of conductordiecontains a number of differentbeams,
electrostatic ground-plane symmetry for the two beams especially when the beams have different aspect ratios.
in the structure illustrated in Fig. 1 7 (a). The very high More work on attempting to hold the center frequency
values of Q in the structure make tuning difficult since as constant as possible over the entire slice must be
when K must be small, small differentials in frequency done,not so much for single RGT’s where simple
between the two beams have a large effect on bandpass frequencyrange “bin-type” classification is effective,
shape. In any case, the results of the twin-beam struc- but for multiple beam/frequency RGT’swhere the yield
ture areencouraging, and work will begin on triple-beam drops to zero rapidly as the variance increases.
electrostatically symmetric RGT’s in the near future. Other problems of the RGT which must be solved are
The prospect of being able to batch-fabricate complex as follows:
bandpass filters is an exciting one.
1) The role of residualbeamstress afterplating
IV. DISCUSSION which can result in beam curl, and therefore unpre-
dictable values of 8 0 .
The RGT holds much promise for solving the prob-
2) The roles of annealing and fatigue on long-term
lem of tuning integrated circuits. The device is quite
device behavior.
small, the typical volume of its mechanicalresonator
3 ) Control of Q g , and therefore thedetector im-
being about lop9 cubic inches. Even including the vol-
pedance level.
ume of the necessary silicon chip to contain theFET de-
4) The role of the free oxide surface as far as excess
tector, the devicerepresentsquite a considerablede-
device noise and ultimate V-1 quiescent stability is
crease in volume a t a given frequency and Q over exist-
concerned.
ing state-of-the-art means of achieving high-Q tuning.
5) The role of feedthrough in degrading the ulti-
Since the fabrication process is basicallycompatible
mate dynamic range of the RGT.
with silicon-based technology,
_ _ i t is possible that even-
tually not only size but substantial cost savings maybe
realized once optimumfabricationmethodsarede- V. CONCLUSIONS
We veloped. transistors
batch-fabricated
demonstrated have
Before long-rangepredictions about the usefulness of exhibiting high-Q tuningcharacteristicsintherange
the RGT are made, however,certainproblems associ- 1-100 kHz.Both single andmultiresonantstructures
ated
with
the device must be resolved. have been constructed.
Theory
and
experiment
corre-
132 IEEE TRANSACTIONS ON ELECTRON DEVICES, MARCH 1967

MOS Detection Bipolar Detection

Fig. 19. Geometries of two posible basic detection schemes for the RGT.

late well, indicating a good understanding of the basic c) Noise and stability differences in the two detec-
device properties. Temperature sensitivity of the device tion schemes are neglected.
is shown to be under control. Although more work is
We define the change in capacitive stored charge QO
necessary to permit quantity production of the RGT,
associated with the cantilever moving a distance 6 as
progress has been relatively rapid on the developmentof
suchindividual
devicecharacteristics as gain, fre-
quency range, size, and multipole pass behavior.

APPENDIXI Referring first to the left side of Fig. 19, we note that
COMPARISON OF BIPOLARAND MOS TRANSISTORSthis additional charge appears directly within the MOS
FOR DETECTING CANTILEVER MOVEMENT active region. The output currentis then Qo divided by
I N RESONANT GATE DEVICES thetransittime in the MOS transistor T ~ o s L,2/ =
The vibrations of a polarized, mechanically resonant PLn V P O :
cantilever can be detected either by field-effect modula-
tion of an MOS transistor or, alternately, by using the
cantilever as a vibrating capacitor to modulate the base
current in a bipolar transistor of current gain 0.For In the case of the bipolar transistor, the same charge
equivalent geometries, it is shown in this Appendix that QOenters the base region at a certain rate QO/rO,where
the ratioof output currentin the two detection schemes T O is the period of the resonance frequency (seconds).
is given by Because of the finite base lifetime TD(seconds), however,
I M O S 1 qEgapailieon resonant period (seconds) base recombination occurs at a rate -Qa/7g, where Qsis
---- _- the steady-state basecharge. As a result, the steady-
rbipolar 8, KT transistor base lifetime ' state base charge is given by
indicating that for frequencies below about 10 MHz,
71)
field-effect detection results in a greater outputsignal. Qs = - Qo,
We define comparative geometries in Fig. 19. Under- 70

neath a cantilever Y units in breadth is an active detec- which for TD/rO<<l (low resonant frequency) is con-
tion region W units wide. Inthe case of the MOS siderably less than the value Q0 all of which appears in
transistor,the region W by Y contains a convoluted the channel of the MOS device. The charge Qs moves
channel L units long, the periphery P of which is about across the base of thetransistor with a transittime
N Y(W/2L). In the case of the bipolar transistor, the
Tbipolar=Lc2/2D
active region is simply a vibrating capacitor of area W Y .
The following assumptions are made in this compari-
son : l'bipolar

a) The cantilever ac motion 6, its equilibrium posi-

Dividing (29) by (30), including the effects of area
tion a,, and polarization voltage Vp are the same for
difference implied in Fig. 1, and taking the Einstein re-
both the MOS and the bipolar transistor.
lation D/p = KT/¶, gives
b) The output resistance of bothtypes of tran-
sistors is sufficiently high that the ac output current IMOS 1 ___
QVPO 70
times the load resistance (assumed equal for the two -e- - -. ( 311
systems) solely determines system voltage gain. Ihipolar 8~ kT TD
NATHANSON ET AL.: RESONANT GATE TRANSISTOR 133

Since the relative transit times for the MOS and bi- ACKNOWLEDGMENT
polar transistors differ by a factor of qVpo/KT for The authors would like to acknowledge the assistance
equivalent base widths, a value of V p o on order E G , the of E. A. Halgas, G. J. Machiko, E. M. Black, and H. B.
silicon bandgap, results in an advantageof about 40 to 1 Shafferin thefabricationandevaluation of experi-
in favor of field-effect demodulation. This comparison is mental devices. Helpful discussions on the gold electro-
not entirely fair, however, since conventional transistor plating process have been held with M. P. Lepselter of
base lengths (formed by paralleldouble-diffusion) are the Bell Telephone Laboratories.
typically two to three times thinner than typical MOS
channel lengths (formed by lateral diffusion). However, BIBLIOGRAPHY
the MOS stillhas in its favor a factor of about 5 in References
transit time even when fabrication advantages of the W. E. Newell, “Tuned integrated circuits-a state-of-the-art
bipolar are considered. survey,” Proc. I E E E , vol. 52, pp. 1603-1608, December 1964.
W. E. Newell, “Selectivity and sensitivity in functional blocks,”
The effect of the term T ~ / T Din (31) is to give the MOS Proc. IRE (Correspondence), vol. 50, p.2517, December 1962.
transistor additional advantage a t low frequencies. For W. E. Newell, “Ultrasonics in integratedelectronics,” Proc.
IEEE (SpecialIssue on Ultrasonics), vol.53, pp. 1305-1309,
instance, for a 100-kHzresonantdevice,theterm October 1965.
T ~ / is
T -100
~ for 1=ps base material. H. C. Nathanson and R.A. Wickstrom. “A resonant-gate silicon
surface transistor with high-Q bandpass properties,” Ap@ Phys.
Since it would be necessary to resort to ultrahigh$ Lett., vol. 7, pp. 84-86, August 15, 1965.
transistors in order to compete with the inherent effi- H. C . Nathanson, W. E. Newzll, and R. A. Wickstrom, “Tuning
forks sounda hopeful note, Electronics, vel. 38, pp. 84-87,
ciency of the field-effect detection process, noise and September 20, 1965.
stability problems in both MOS and bipolar detection S. R. Hofstein and F. P. Heiman, “The insulated-gate field-
effect transistor,” Proc. IEEE, vol. 51, _ pp. .1190-1202, Septem-
should be about equivalent. This is because low-current, ber 1963.
high-0 transistors are essentially surface-recombination- H. C. Nathanson. “A high field triode,” Solid-StateElectron.,
vol. 8, pp. 349-363, 1965.”
velocity-controlled and thereforesusceptible tomany W. Shocklev. “A uniDolar field-effect transistor.” Proc. IRE. vel.
of the surface problems of field-effect surface devices. 40, pp; -136211376, fiovember 1952.
M. M. Atalla, E. Tannenbaum, and E. J. Schiebner, “Stabiliza-
Output impedances of both the MOS and the bipolar tion of silicon surfaces by thermally grown oxides,” Bell Sys.
transistor are controlled essentially by the same volt- Tech. J., vol. 38, p.761, May 1959. ~

Lord RavleiKh, Theory - of- Sound, vol. 1. London:Macmillan,

age-dependent Early effect, leading to little advantage 1894.
- I

of one over the other at small active lengths. Problems Von Werner Koster, “Die Temperaturabhandigkeit des Elasti-
zitatsmoduls reiner Metalle,” Zeit. f u r Metallkunde, vol. 39, pp.
associated with the precise control of channel conduct- 1-8, 1948.
ance in the MOS transistor are admittedly quite real, N. Schwartz and R. W. Berry, Physics of Thin Films, vol. 2,
G. Hass and R. E. Thun, Eds. New York: Academic Press, 1964,
but with advancing surface technology this aspect may pp. 393-398.
become less important. See, for instance, W. P. Mason and R. N. Thurston, “A compact
electromechanical band-pass filter for frequencies below 20
In summary, below about 10 MHz (for typical bi- kilocycles,” IRE Trans. on Ultrasonics Engineering., vol. UE-7,
polar base lifetimes), significantly greater output signal pp. 59-70, June 1960.
is achieved using field-effect-type detection in RGT’s General References
than would be obtained utilizing bipolar-type detection [14] J. J. O’Connor, “A 400 cps tuning fork filter,” Proc. IRE, vol.
systems of equivalentgeometry.Thisresulthas irn- 48, pp. 1857i1865, November 1960.
portant implications for future design of resonant gate [15] F. Dostal, The increasing appllcations of tuning forks and
other vibrating metal resonators in frequency control systems,”
devices. Proc.19th Ann. Symp. o n FrequencyControl, pp. 59-77,1965.