IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-27, NO. 9 , SEPTEMBER 1980 1809 Dual-Gate Bucket-Brigade Devices RADU M. BARSAN, MEMBER, IEEE Abstract-The low charge-t ransfer efficiency of the basic bucket- brigade circuit and the speed limitation of th e tetrod e approach have so far prevented bucket-brigadedevices (BBD’s), in spite of their greater fabrication simplicity, from competing with charge-coupled devices, excepting for audio applicati ons. A novel charge-t ransfer device is presented and experimentally eval- uated. The dual-gate BBD” is wo-step-transfer devicea CCD transfer followed by a BBD transfer), howing operational performances comparable t o surface-channel CCD’s and, at the same time, enjoying the fabrication simplicity proper to bucket-brigade circuits. The con- cept is implemented usi ng conventional p-channel al uminum-gate tech- nology combined with a n additionalshadowed-gap float-off process. The haracteristic eatures and theperformance apabilities of the device are discussed. D I. NTRODUCTION URING th e last decad e, charge-transfer devices (CTD’s) [ l ] have developed from the laboratory concept into a mature class of semiconductor devices with significant impact in the areas of sampled-data signal processing, maging, and memory.The CTD family ncludes harge-coupled devices (CCD’s) [2 ] an dbucket-brigade devices BBD’s ) [3 ]. Gener- ally, CCD’s have better operational performances than BBD’s. A tmoderate requencies,charge-transfer nefficiency nsur- face-channel CCD’ss o n h eo r d e ro f n do n h eo r d e r of lo-’ in buried-channel devices, whereas BBD’s are limited to values greater than Darkurrentndoisere also smaller in CCDs and the packing density is greater than that of BBD’s. In exchange, CCD’s are more difficult to fabricate. In order to achieve their full capabilities, CC Ds require lead- ingedge technologies, nvalidating heearlypredictions hat their processing would be simpler than onventional MOS circuit technology. In contrast, BBD’s can be fabricated using the simplest of the MOS processes, such as a p-channe l alumi- num gate. This paper describes a novel bucket-brigade tructure, termed dual-gate BBD, which exhibits performan ce capabilities com- parable to surface-channel CCD’s. The device is fabricated using ahigh-thresholdp-channel Al-gate technology with an additional “float-off” process yielding self-aligned gates with submicrometereparations.he device characteristicsre discussed in Section I1 and som e processing details are presented in Section 111. Experimental results on device performance are reported and interpreted in Section IV. 11 . CHARACTERISTICS F THE DUAL-GATE BD In he basicBBD shiftregister, shown n Fig. l(a), charge transfer takes place through the channel of a simple MOSFET whosegate-draincapacitancehas been largely increased to Manuscript received July 15,1979; revised April 15, 1980. The author is with the R and D Center for Semiconductors, Buchar est 72996, Rumania. BIAS LINE I 1 1 1 ! 1 2 j ( 4 Fig. 1. Schematic cross sections through a p-chan nel (a) simple BBD, (b) tetrode configuration, an d (c) novel dual-gate bucket brigade. form on e of the storage elements in the delay line. The most significant effects contributing to charge-transfer nefficiency at owand ntermediate requenciesare he initedynamic drain conductance, threshold-voltage modulation, storage- capacitancemodulation 4] -[8] , an dsubthreshold leakage [7], [9]. Athigherclock ates, the ransfer inefficiency n- creases due to the finite time allotted for charge to flow from one capacitor to the next [4] -[8] . This inherent limitation is determined by he gain characteristicsof the ransistors, as well as by the ag nitud e of the storage capacitances, and the corresponding intrinsic inefficiency term [7] is given by where 0 is the saturated MOSFET square-law factor, Vc t h e clock voltage level, V , the hreshold voltage, & , the initial source potential, C ,, the storage capacitance, and f, he clock frequency.Th e dominating mechanisms at low ndnter- mediaterequenciesre rimarilyetermined by channel (barrier) ength mo dulation, a ypical example being the dy- namic drain conductance contribution to incomplete transfer, given by where g is the MOSFET saturation output conductance and gm the transconductance. A general method to educe the modulation effects in CTD’s is to isolate the source from the charge sink by means of an 0018-9383/80/009-1809$00.75 0 1980 IEEE
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-27, NO. 9, SEPTEMBER 1980 1809
Dual-Gate Bucket-Brigade Devices
RADU M. BARSAN, MEMBER, IEEE
Abstract-The low charge-transfer efficiency of the basic bucket-brigade circuit and the speed limitation of th e tetrod e approach haveso far prevented bucket-brigadedevices (BBD’s), in spite of their greater
fabrication simplicity, from competing with charge-coupled devices,excepting for audio applications.
A novel charge-transfer device is presented and experimentally eval-uated. The dual-gate BBD” is wo-step-transferdevicea CCD
transfer followed by a BBD transfer), howing operational performances
comparable to surface-channel CCD’s and, at the same time, e njoyingthe fabrication simplicity proper to bucket-brigade circuits. The con-
cept is implemented using conventional p-channel aluminum-gate tech-
nology combined with an additionalshadowed-gap float-off process.The haracteristic eatures and theperformanc e apabilities of thedevice are discussed.
DI . NTRODUCTION
URING th e last decad e, charge-tran sfer devices (CTD’s)
[ l ] have developed from the laboratory concept into a
mature class of semiconductor devices with significant impact
in th e areas of sampled-d ata signal processing, maging, an d
memory.The CTD family ncludes harge-cou pled devices
(CCD’s) [ 2 ] an dbucket-brigad e devices BBD’s) [3 ]. Gener-
ally, CCD’s have better operational performan ces than BBD’s.
cont.ributions to incom plete transfer are reduced from those of
the ;simple BBD with the same storage capacitance, by a factor
The main drawback of the tetrode configuration is its low
speed of operatio n, as comp ared with th e basic circuit. The
effective intrinsic transfer rate term of a two -step transfer cell
is given appro xima tely [7] by
of ( ( 7 p + Cg1 -t Cov)/Cst *
where the ubscripts1and 2 refer to he firstand econd
transfer step, respectively. el i is given by (11) in which V, is
replaced by V B , he voltage applied to he bias line. e , j is
given b y a similar expression, in which C is replaced by Cz
an d Kn by the initial island p otentia l (whiclh is a func tion of
Kn, ,, an d &). Inorder to improve th e low-flrequen’cy
performance, C has to bemade large compared to CI=
C pt Cgl t C whichdrasticallyaffects el i. Inaddition,
noneof he wo channels of the etrode cell canbemade
shorter than the channel ofa simple BBD ceU.. Thus the better
low-frequency behavior of the tetrode con figuratio n is accom-
panied by a poor high-frequency performance. This limits the
usefulness of the tetrode conce pt to audio a.pplications (typi-
cally below 50 kHz).
The novel dual-gate BBD structure is shown schematically in
Fig. l(c). This onfiguration achieves the screening of the
source with respect to the drain of each MOSFET by a self-
aligned bias-gate lectrode btainedwith single level of
aluminum and a submicrom eter gapprocess.’ Thus each cell
contains a dual-gate MOSFET [121 , which has th e bias gate
located near the source and the active gate near the drain. In
contrast to th e tetro de stru cture discussed above, in which the
charge transfer s by twosuccessive “bucket-brigade’’ processes,the charge transfer n he dual-gateapproachconsistsofa
“bucket-briga de” rocessollowed by acharge-coupled”
process, provided the dc voltage applied to the bias gate is low
enough with respect to th e clock voltage to ensure a complete
charge-transfer mod e of operation of the “CCD” half cell.
The ntermediatecapa citanc e of the dual-gate BBD cell is
‘The same concept could also be implemented using overlapping-gatetechnolog ies but wh ich, beside increasing the capacitive loading of theclock lines, are mo re complex and, consequently, less suitable fo r cost-effective CTD’s.
n m d e u p clrlly o f th e effective gate capacitance associated with
t.heias-gs-te electrode c bg , , Consequently, all modulation
ef fects : do~nh~at in~ ;he low- and intermediate-frequency trans-
fe:: rle!ffi.c~.e:nc:y areeduc ed .by a facto r of c b g / c , , compared
t o th.e simple BBID, and by a fac tor of c b g / ( c p t C t Co,)
compa.red to the tetrode configuration. The first ratio may lay
in the range of 1/50-~1/100.Considering a tetrode cell and a
dulal-,gaie st:ll of :qlud chann el width s, the second factor can
be expaessed alpprorrimately as
bg
Cd e pr == (4)
(I, ;I-2 lo, t 2:x )- g l + 2 lovcox
wherle l ; b g is the affective length of the bias gate of the dual-
gate c1d1,Lgl the channel length of the first MOSFET in th e
tetroldt:cell., 1, theminimum nterelectrodespacing, Zo v the
minimu m overlap betw een gate and diffusion: xj the junction
depth , ‘Cde:p the depletion capacitance per unit area, and Cox
th e oxide ‘capacitance per unit area. Taking L b g = 4 pm, Lgl =10 pm , 1, =- 8 pm.,o, = 6 pm , xj = 2 pm, dep = COx/3 , the
factor r given by (4 ) is about 0.13 . These examples show that
the d.u.al-gate BBDI has a low -frequ ency transfe r ineff icienc yof
7-8 times smaller tha n a tetrod e BBD and up to t wo orders ofmagnitude smaller than the basic circuit.
Since the ntermediatecapacitanceof hedual-g ate cell is
reduced by a factor r from that of the tetrode ell, the storage
capacitamcle can be made r times smaller for the two structures
to ex.hibit similar low-frequ ency performances. This results in
a signifjicant high-frequenc y (intrinsic ransfer-efficie ncy) m -
provennent. Moreover, the high-frequency imit of he dua l-
gate device s further increased through the reduct ion of e Z i ,
whi ch,, n his calse, corresponcls to a CCD transferprocess.
Inadditio n, he activechannel. canbemade muc h smaller
while still avoiding short-ch annel effects [12] . In conclusion
of the preceding discussion, th e dual-gate BBD offers signifi-
cantly better transfer efficiency over a wide frequency range.
Furthe rmore , he possibility to reduce the cell dimensions,
Another advantage of the dual-gatestructure over existing
BBD’s co ncerns th e dy namic range. First, the parasitic capaci-
tance between the pulsed gate and the source is reduced due t o
the shielding effec t of the self-aligned bias gate, so tha t larger
signal chargepackets can be ransferred 13] . Second, the
absence ofan interm ediate island reducesdarkcurrentand
noise. In ad dition, the in put and output stages can be designed
to reduce the effects of parasitic capacitances. The dual-gate
BBD technology is also co mpatible with improved peripheralamplifiers using dual-gate MOSFET’s [141 , whichcan be
attrac tive or signal-processing circuits ncorp oratin g nalog
delay functions.
The dlual-gate bucket-brigadeconcepthasbeenreduced to
practice using p-channel single-levelaluminum-gate technology.
The self-aligned metalpattern s were obtaine dwith a ub-
21f the tetr od e BBD is fabricated with self-aligned islands, Z = 0.This higher degree of process complex ity is no t required b y the dual-gate approach, so that comparison is made for the simplest of MOSprocesses.
Fig. 2. (a )Microphotograph f experim ental ud-gate BBD chip.Lower half of chip con tains an 80-stage shift register. Upper portio ncontains est transistors and capacitors. (b) Enlarged view of hetransfer struc ture. The horizontal narrow stripes form the bias lines(gates). They are separated from adjacent clock lines by submicrom-eter (0.5-firn) gaps.
microm eter-gap process similar to th e “shadow-etch” approach
described by Browne and Perkins [151, [16] ,yielding typically
0.5-pm gaps. A photograph of the complete chip is shown inFig. 2(a) and an enlarged view of some of the 80 cells of th e
BBD circuit is sho wn n Fig.2(b). Thepathof he signal
samples along th e delay line follows a serpentine, alternating
between the clock buses from cell to cell and turning around
every 10 stages. Beside theshift register, thechipcontains
several dual-g ate, as well as norm al, MOSFET’s and som e MOS
capacitors used for characterization and evaluation purposes.
A schematic diagram of th e dual-gate BBD shift register is
shown n Fig. 3 . The circuitconsists of an nput sampling
stage (sampling transi stor ST and capacitor C,), 80 identical
transferstages,aprechargeror eset ransistor ( R T ) , and a
..INPUT
+OUTPUT
Fig. 3. Dual-gate BBD circuit.
source follower (output transistor O T ) . Note that , in contrast
to he transfer ransistors, he nput dual-gate MOSFET has
the gate near the drain connected to the dc bias line (whose
voltage is VB) , he gate near the source being pulsedby one of
the clock phases (G1). The reason for this particular design is
as follows. When the npu t ransisto r is on , he voltage at
node A equals th e inp ut level qn. Upon switching ST off
(raising $1 an ddropping &), the voltage at no de A instan-
taneously changes by an amoun t
where VGis the amplitude of theclock pulses. If Cgus greater
than Cgs,he input signal can no longer vary from zero, but
only rom he level given by ( 5 ) . Consequ ently, he ideal
maxim um nput swing VB- V, is reducedby hisamount.
Du e to the shielding effect of th e bias gate , the parasitic capac-
itance Cgss mu ch lower than the capacitance between a diffu-
sion and the gate overlapping it. According to the preceding
discussion, if the gate near the drain of ST were connected to
G 1 , th edyn amic range would be educed.The lternative
shown in Fig. 3 is self-compensating with respect to parasitic
capacitancemism atch, since bo th gates adjacent to C, are
formed by the same bias line.Th edynam ic range salso limited by th e parasiticcapaci-
tances at the output tage C,, an d cd, (junction capacitances).
When RT is switchedof fand the ast ransfer ransistor is
switched on, the voltage at node B should ideally change from
the reference level VB - V, to V, -t V, - V,. However, due
to parasitics, the voltage changes by only a raction of VG,
given by
which is the max imu m allowable inpu t swing. The use of a
dual-gate precharge transistor reduces Cgs nd henc e, increases
the dynam ic range. In con clusio n, assuming the effect of para-sitic capacitances at the input stage to be negligible, the maxi-
m um inp ut signal for a dual-gate BBD delay ine is given by
V, - V, or ( 6 ) ,which ever is smaller.
111. PROCESSING ONSIDERA TIONS
The dual-gate BBD structures have been abricated using
Fig. 4 . SEM photog raph and cross section of a dual-gate BBD cell. Lgis the effective length of the active channel and Lbg is the effectivelength of the channel under the ias gate.
evaporated and patterned to define he peripheral metalliza-
tions and the clock buses. By an additional photolithographic
step, the complement of the bias-gate pattern is printed. Using
thismask, thefirstmeta l is slightlyoveretched inorder to
produceshadowed edges [16] , Without removing the resist,
a second alumin um film s evaporatedwhich forms he bias
gates self-aligned with respect to th e active gates. Liftoff ofthe meta l from over the resist is subsequently made in acetone
ornitricacid,dependingon ts hickness. Since thesecond
layer metal pattern is surrounded b y shadowed spacings and
the bias line being accessed by an und erpas s, the only places
where the metal has to break over the resist step are narrow
termination s at the periphery of the BBD structure. The pro-
cessing s comp letedby hedeposition of an overlay oxid e,
or simply by a heat treatm ent (10 min in hydrogen and 10 min
in n ytrogen, b oth at 50°C).
A typical nterelectrode area is shown in Fig. 5 . Note he
steepness of the wet-etche d metal 1 wall, as compared to the
large slopeof themetal 2 pattern,produced by shadowed
evaporation naplanetary y stem.Th eupper edge of th eslope corresponds to the resist (shadow) edge and the distance
to th e first layer m etal wall shows the magn itude of overetch-
ing (the overhan g of the resist). Fig. 6(a) shows a schematic
crosssection hrough hestructureduring hedeposition of
the second aluminum ay er. On a geometrical basis, the size
of the gap g is determ ined by the thickness of the first metal
layer h, the degree of shadowing (overetching) e, and the most
obliqueevapo ration angle in theplanetarysystem amin. In
practic e, he resulted gap s smaller due to the migration of
Fig. 5. Scanning electron m icrograph (10 000 X ) of the region separat-ing the bias and clock metal buses (after final heat treatment). Metal1 is 0.8 pm thick and metal 2 is 0.7 pm thick. The gap size is abo ut6000 A and th e "shadow" (th e resist overhang) appears to have beenabout 1.5 pm. Lower right corner shows an area where aluminum hasbeen removed to prod uce a cross section throug h the metallization.
/'
2 3 L
O V E R E T C H I N G l y m )
(b)
Fig. 6 . (a) Cross section hrough the structure during the evaporationof the second m etal layer. (b) T he gap size as a function of the over-etching of the first metal pattern. Points represent measured valuesand solid lines are plotted according to (7).
aluminum atom s during evaporation (typically about 3000 A
[161) and during the final heat treatment or oxide dep osition(this was experimentally observed to be about 2000 A fo r
aluminum). Consequently, the gap size is given by
g = e - h cotan a m i n - m (7)
where m represents the gap reduction due to aluminum migra-
tion. For amin 60" an d m = 0.5 pm , good agreement exists
between (7 ) and the measured gaps, as shown in Fig. 6(b) for
tw o d ifferent thicknesses of th e first metal layer. As a result
of the investigation of gap formation, it was confirmed that
5000 A is a ypical ower imit for he size of reproducible
gaps obtained using aluminum [16] . Since elatively table
devices have been fabricate d with gaps as large as 1.5 pm, the
0.5-pm margin is not a serious limitation from the standpoint
of device performan ces, tability, nd eproducibility.The
transfer nefficiencymeasured on 0.5-pm-gap hift egisters
showedweakdependence on background harge, which
means tha t very good con trol (absence of spurious barriers)
ofuc hnterele ctrode spacings is possible.However, the0.5-pm limit could become a problem when very small chan-
nels (<3 pm) aredesired,because uch sho rt gates ail to
con trol gap regions greater than a few tenths of a micrometer.
A variety of metal thicknesses for bo th aluminum films have
been tes ted. It was foun d that a difference of maximum 1000
A between the two is sufficient for easy liftoff of unwanted
second-layer metal from over the resist. From threshold volt-
age,etchingcontrolability,an dbondingconsiderations, he
thicknessof thefirstaluminum layer or the devices to be
characterized was chosen of 0.8 pm.The second ilm was
0.7 pm thickand he resulting gaps were ypically 0.5 pm .
Th eoxide hickness was in he rangeof 1200A-1300 A
(slightly hinnerunder he bias gatesdue to anadditionalcleaning sequence prior to the second aluminum deposition).
The average surface mpurityconcentration of the substrate
((1 11) orientation) was 2 -4 X 1015cm-’. The boron diffu-
sion yielded junction depths of about 1.6 pm.
The average effective surface mobility of holes was fou nd to
be 140 cm2/V - s. The threshold voltage was typically - 4 V
for MOSFET’s form ed by the first aluminum , and about - 4.5
V for those whose gates were formed by the second deposition.
IV. EXPERIMENTALESULTS A N D DISCUSSION
OF DEVICE PERFORMANCE
The principal parameters (after processing) of the 80-stage
dual-gate BBD’s which were characterized are
active channel length,L , = 8.5 pm;
bias-gate length,Lbg = 6 pm;
channel width,W = 32 pm;
storage capacitance,Cst= 1.4 pF.
The devices were operated with zero substrate bias and clock
voltageamplitudesof 20 V , at requencies varying between
1 kHz and 1 MHz. The max imum in put swing for 20-V square
clocks and a dc gate bias of -2 0 V was about 11V. Typical
device responses are shown in Fig. 7.The registers were characterized with respect to transfer effi-
ciency by the step-response meth od, whereby the incompletetransfer parameter E is determined from
L T = ( N - l ) E (8)
where N is the numb er of delay elements and L T the total
normalized loss, defined as the sum of all trailing p ulses divide d
by the steady -state outpu t level. Plots of experimentally de-
termined transfer inefficiency,as a function of clock frequency
(&), are show n in Fig. 8 for two different voltages applied tothe bias line. Also shown is the measured transfer inefficiency
ofa128-stagesimple BBD delay ine,which was processed
(b)
Fig. 7 . Ou tpu ts of t he BO-stage dual-gate BBD. Verticalscaleis 1V/div.(a) Step function response a t 26 0 kHz. (b) Impulse response at 100kHz. The inputword is a sequence of the form 10001000 . . .
together with the dual-gate devices. The observed dependence
of E on the bias gate voltage Vj is illustrated in Fig. 9.Let us first discuss the behaviorof hedual-gate BBD for
large bias gate voltages. It is clear in Fig. 8 that for this case
(VB = - 16 V) the dominant mechanisms responsible for trans-
fer inefficiency below abo ut 1 MHz are subthresho ld conduc-
tion (diffusion over the barrier) and the dynamic drain con-
ductance effect, since E is weakly dependent on clock frequency
and shows th e same dependence as the simple BBD (the two
curvesareparallel)which is surelyaffectedbymodulation
effects only (Lg = 5 pm), within this frequency range. With
regard to he relativemagnitude of he wo ransfer-ineffi-
ciency values, the first-orde r analysis in Section I1 predicts an
improvement of cb,/cst for the dual-gate device, compared to
the simplecircuit. By taking intoaccou nt he difference inchannel lengths, the above factor has to be multiplied by the
ratio of the simple BBD channel length to tha t of the dual-
gate device. The result is 0.023 , which is close to th e observed
improvem ent. Also no te (Fig. 9) that or V, in he range
VG- VT < V ,< V,, E increases with increasing V (absolute
value) for V, fixed. This is due to the reduction of the fring-
ing ieldbetween the bias and heactivegatewhen VB is
increased towards V, (when V, is higher tha n VG the channel
und ernea th the bias gate is no longer pinched off and acts as
a simple series resistance). In term s of saturation conductance,
1814 IEEE TRRXSACTIONS ON ELECTRON DEVICES,VOL.ED-27, NO. 9, SEPTEMBER 1980
t i
tVG i- 0
lo'
106 lo5 106
C L O C K FREQUENCY ( H z )
Fig. 8. Dual-gate bucket-brigade ransfer nefficiency as a functio:n ofclock freque ncy. The broken line corre spon ds to a simple BBD clhiftregister having 5-pm channels (40 pm wide) and a storage capacitanceof 0.4 F.
10-1 I I I I I I I I I
vo.-20v
$ lo-*-2 .
f c = 5 0 0 K H z
w .i!u -U
103:-E -wL C .v)
5 166:E .I-
1 0 - 5 ~ ~ ' ~ ~ ~ ' " .-10 -12 -l L -16 -10 -20
B I A S G A T EO L T A GE I V
Fig. 9. Measured transfer inefficiency versus bias gate voltage for th.reedifferent clock frequencies.
g& increaseswith V,, as will be illustrated ater on in this
section.The regime in which V, is not low enoughwith
respect to VG to pinch off he channel underneath he bias
gate corresponds to an incomplete (bias) charge-transfer mode
of oper ation of the "CCD" half cells of the dual-gate B13D.
When V, is reduced below approximately VG- V,, another
mechanism comes int o play at higher clock rates. TMIS s the
intrinsic transfer rate lim itation associated with the fir:;t trans-
fer in the dual-gate BBD cell, resulting in th e term el i in (3).
Indeed, the low bias voltage plot in Fig. 8 (V, = - 12V) showsa more rapid increase with clock frequency above abo ut 100
kHz. Also, the 5 00-kHz plot in Fig. 9' shows a decrease of e
with ncreasing VB for ow V values,which onfirms the
conclusion that he contribu tion of e l l is significantat low
bias gate voltages and high clock rates.
From the poin t of view of dynamic range, V, should be as
high as possible. Howe ver, as evidenc ed in Fig. 9, the choice ofV , only within an optim um range yields a minimum transfer
inefficiency. The upp er limit of this optimum range is on the
order of VG- V,. The choice of V, is critical only at higher
clock requencies, bu t sincenormally V, is derived on-chip
- 1 ' I
O 10 20 30 LO
G A T EV O L T A G E V I
Fig. 10. Dual-gate M OSFET outp ut cond uctan ce in saturation versusgate voltage for diffe rent bias gate voltages. L g = 8.5 pm, L b = 15pm, and W = 64 pm. The dashed line corresponds to a norm d t ran-sistor having the gatedefined by he secondaiuminumdeposition.The channel width s the same and the channelength is 11pm .
(from the VDD upply) it has to be set regardless of the oper-
ating rate. Concerning the clock frequency figures mention ed
in the preceding discussion, it shou ld be kept i n mind tha t the
dual-gate BBD under consideration has rather ong channels,
large storagecapacitances,anda owcarriermobility.Thus
other :structureswithmproved lectrical nd eom etrical
parameters mayexhibit ow-frequencycharacteristics up o
much higher clock rates.
In order t o get m ore insight into the behavior of the dual-
gatecell,measurementswereconducted on individualdual-
gate MOST'S on thesame chip. Emphasis was directed towa rds
the outp ut cond uctanc e in saturatio n, which is directly deter-
Fig. 11. Dependence of the saturat ion output conductance gds on biasgate voltage VB , for hreediffere nt dual-gate transistors. All threehav e th e s q e channel width W = 64 elm. Transistors A an d B onlydiffer throu gh bias gate length, whereas B an d C have different activechannel lengths. The broken l ine shows the output conductance of anorma l MO SFET whose chann el length equals the sum of L , and Lb,of the dual-gate transistor B .
pincho ff point, he dual-gate MOSFET current is controlled
by he first ransistorwhoseeffective“drain”voltage is in-
creased by increasing VG (this being the operat ing regime of
the dual-gate BBD).As a onsequence, gds decreaseswith
increasing VG,normally going t o z ero orvery large gate
voltages,and his is shown to be the case formoderate VB
values. For bo th VGand VB arge (high current levels), a nega-
tiveconductanceeffect shows up n Fig. 10 . Thiscould be
explained b y the decrease of mobility with increasing tempera-
ture in the ocally heated channel at high power levels [171 .In o rder t o assess more clearly the effect of the bias gate,
experimental results for gds measured on different dual-gate
MOST’S are show n in Fig. 11 as functions of V B . The output
conductance increases linearly with VBup to t he point where
the first transistor of the dual-gate evice is no longer saturated.
A direct observation is that a longer bias gate allows bett er
cont rol of the dual-gate ransistorparameters, as exp ecte d.
However, even a 2.5-pm bias gate s able to significantly reduce
gds of an 8.5-pmdual-gate device from hat of anormal
MOSFET,provided VB is lower than VG. When VB equals
VG, he output conductance is abou t the same as that of the
normal transistor whose channel length is equal to L , t Lbg .
If the bias gate is muc h shorter than the active channel (case
C), gds tends to be less dependent on V B . It follows that the
relative length of the bias gate with respect to th e active gatealone is of consequence. Although he onger he bias gate,
the smaller theoutputconductance,our results how tha t
significant improvement over equivalent normal MOSFET’s
is possible even w ith bias gates 3 to 4 times shorter than the
active channel.
V. CONCLUSIONS
Due to a elativelyhighdegree of processingcomplexity,
CCD’s are less accessible CTD’s than BBD’s. On theother
h a n 4 BBD’s can be easily fabricated, but they offer poor per-
formances, mainly from the point ofiew of transfer efficiency.
A novel two-step transfer CTD concept (combining the BBD
and CCD principles)hasbeendescribedandexperimentally
verified. Th e dual-gate BBD achieves charge-transferperfor-
mancescomparable to surface-channel CCD’s. T heprincipal
advantage s of his new str uct ure over existing BBD’s can be
summarized as follows:
lower transfer inefficiency (more than an order of magni-
tude smaller than hat of simple BBD’s and up o an
order of magnitude smaller ha n that of tetrod e BBD’s);
larger high-frequenc y imit and higherpackingdensity
than tetro de BBD’s;
larger dynamic range (higher allowable inpu t swing and
lower noise).
conceptha sbeen mpleme nted using p-channel Al-gate
separated rom the clockbuses by reproducible ubmicron
gaps (typically 0.5 pm). The high yields proper to p-channel
aluminum-gate technology were not affected by the additional
process.ransfernefficiencieselowt 100 kHznd
below at MHz have beenurrentlybserved,espite
the low carrier mobility (140 cm 2 /V s).
The opera tion of the dual-gate BBD has been discussed andthe experimental results have been interpreted. Measurements
on individual dual-gate M OSFET’s having the active gate near
the drain thebuildingblocksof the novel CTD structure)
have revealed a dependence of the saturatio n output conduc-
tance on gate voltage diffe rent from that of single-gate tran -
sistors. The behavior of gdshas been explained and correlated
with the observed operational performances of he dual-gate
bucket brigade.
AC KNOWLEDGMENT
Theauthor wishes to han k A.Delibaltov who provided
invaluablehelpduring the developmentalprocessing tage.Th eassistanceof S . Prisecaru in the SEM evaluationsand
of G. Voicu in the measurements is appreciated. Continuous
supp ort and encouragement from C. Bulucea is also gratefully
acknowledged.
REFERENCES
[11 C. H. Skquin and M. F. Tompsett , Charge Transfer Devices, suppl.8 to Advances in Electronics and E lectron Physics. New York:Academic Press, 1975 .
[2] W. S. Boyle and G. E. Smith, “Charge coupled semiconductordevices,” Bell Syst. Tech. J . , vol. 49, pp. 587-593, Apr. 1970 .
[ 3 ] F. L. J. Sangster and K. Teer, “Bucket-brigade electronics-Newpossibilities for delay, timeaxis conversion, and imaging,” IEEEJ. Solid-state Circuits,vol. SC-4, pp. 131-136, Jun e 196 9.
[ 4 ] K. K. Thornber, “Incomplete charge ransfer in IGFET bucket-brigade sh ift registers,” IEEE Trans. Electron Devices,vol. ED-1 8,
[5] C. N. Berglund and H. J. Boll, “Performance limitations of theIGFET bucket-brigade shift register,” IEEE Trans. ElectronDevices, vol. ED -19, pp. 852-860, July 1972 .
[ 6 ] C . N. Berglund and K. K. Thorn ber, Incom plete transfer incharge-transfer devices,” IEEE J. Solid-state Circuits, vol. SC-8,
[ 7 ] -, “A fundamental comparisonof incom plete charge transferin charge transfer devices,” Bell Syst. Tech. J., vol. 5 2, pp. 147-182, Feb. 1973.
[ 8 ] W. J. Butler, M. B. Barron, and C.McD. Puckette, IV , “Practicalconsiderations for analog operation of bucket-brigade circuits,”IEEE J. Solid-state Circuits,vol. SC-8, pp. 157-168, Apr. 1973.
[91 D. D. Buss and W. M. Gosney, “The effect of subthreshold leakageon bucket-brigade device operation ,” presented at he Device
Research Conf., Edmonton, Alta., Canada, June 1972.
1816 IEEE TRANSACTIONSN ELECTRONEVICES, VOL.D-27, NO. 9, SEPTEMBER 1980
[ l o ] F. L. J . Sanster, “Integrated bucket-briga de delay line using MOStetrodes,”PhiZips Tech. Rev., vol. 31, p. 266, Aug. 1970.
[ l l ] L. Boonstra and F. L. J. Sangster, “Progress in bucket-brigadecharge-transfer devices,” in IEEE Solid-state Conf., Dig. Tech.Papers, pp. 140-141, Feb. 1972.
[ 1 2 ] W. P. Noble, Jr., “Sho rt-channe l effects in dual-gate field effecttransistors,” in IEEE Int . Electron D evice Meet., T ech. Dig.,pp. 483-486, Dec. 1978.
[ 1 3 ] L. G. Heller and H. S. Lee, “Chargetransfer device modeling,”Semicond. Silicon, pp. 814-828, 1973.
[14 1 N. Weste and J. Mavor, “MOST amplifiers for performing periph-
eral integrated-circuit function s,” IEE J. Electron. Circuits Syst.,vol. 1, pp. 165-172, Sept. 1977.
[ I S ] V. A. Browne and K. D. Perkins, “Buried-channel CCD’s withsubmicron electrode spacings,” in Proc. Int. Charge-CoupZedDevices Conf., pp . 100-105, Oct . 1974.
[ 1 6 ] -, “A nonoverlapping ate harge-couplingechnology forserial memory and signal processing applications,” IEEE J . Solid-State Circuits, vol. SC-11, pp. 203-207, Feb. 197 6.
[ 17 1 D.Sharma, J . Gau tier, and G. M erckel, “Negative dynam ic resis-tance in MOS devices,” ZEEE J . Solid -state Circuits, vol. SC-13,pp. 378-380, June 1978.
Charge-Transfer Noise Theory for Surface-ChannelCharge-Coupled Devices
YASUHISA DMURA A N D KUNIKI OHWADA
Abstra ct-The detailed expression of charge-transfer noise or surface-
channel CCD’s is derived from the surfacegeneration-recombination
rate equation. Not on ly the charge fluctuation due to occupied inter-
face states but also th at due to unoccupied interface states are taken
into accou nt. The derived expression can explicitly predict he spec-
trum intensity dependences on the signal charge concentration and theclock frequency for various amounts of the fat-zero charge. The com -
parison betwee n th eory and experim ental esults shows good agreementover a wide range of signal charge concentration and clock frequency.
It is shown tha t charge-transfer noise depen ds heavily on the ratio
of the signal charge concentration to the clock frequencyand he
cap ture cross section in the absence of fat-zero charge. Fo r the small
ratio, the charge fluctuation is uled by the ransfer-process noise which
is determined from the amount of uno ccupied interface states. For the
very large ratio, the storage-processnoise which isdetermined rom
the interface s tate distribution over the b and gap ru les the charge fluc-
tuation. Fat-zero charge hardly reduce s he transfer-process noise forthe small ratio. For the large ratio, he fat-zero charge enhances the
total transfer noise.
I. I N T R O D U C T I O N
N OISE ORIGINS have been analyzed so far norder to
apply charge-coupled devices t o low-noise analog devices
[11 , [2 ] . Thornber and Tompsett [3] have propos ed that the
charge-transfer noise is attributedmain ly o transfer neffi-
ciency , and that there is a correlation between charge fluctua-
tion s in each pair of adjacent charge pack ets. They classified
charge fluctuat ions into the transfer-p rocess oise which is due
to correlation and the storage-process noise w hich is indepen-
dent of correlation. They derived the noise spectrum formula
Manuscript received November 1 , 197 9; revised April 20, 1980.The authors a re with Musashino Electrical Communication Labora-
tory, Nippon Telegraph and Telephone Public Corporation , Musashino-shi, Tokyo, 180 Japan.
which consists of the transfer-process noise spectrum and the
storag e-proc ess noise spectru m. Noise spectrum characteristics
were experimentally demonstrated for a high clock frequency
(-1 MHz) [4] , [SI. On the other hand,Carnes and K osonocky
[6] have estimated the storage-process noise and the transfer-
process noise on thebasis of the surface generation-recombina-
tion theory [7] , [8 ]. They have considered that the transfer-
process noise is onlydue to interfacestatesoccupied by a
fat-zero charge . However, it can be considered, n fact, hat
not only occupiednterface states but lso unoccupied interface
states contribute t o the charge fluctuation during a transfer-
in period and a transfer-out period.
In his paper, an advanced heory on charge-transfer noise
spectrum is presented as an explicit fun ction of signal charge