1194 IEEE TRANSACTIONS O N MICROWAVE THEORY AND TECHNIQUES, VOL. 39, NO.

I , JULY 1991

GaAs Nonlinear Transmission Lines for

Picosecond Pulse Generation and
Millimeter-Wave Sampling
Mark J. W. Rodwell, Member, IEEE, Masayuki Kamegawa, Ruai Yu, Michael Case, Eric Carman,
and Kirk S. Giboney

Abstract -The GaAs nonlinear transmission line (NLTL) is a under test to a lower frequency before acquisition. The
monolithic millimeter-wave integrated circuit consisting of a instrument bandwidth is limited by the sampling circuit
high-impedance transmission line loaded by reverse-biased
Schottky contacts. Through generation of shock waves on the bandwidth, which in turn is limited by sampling diode
NLTL, we have generated electrical step functions with 5 V parasitics and by the duration of the strobe pulse used to
magnitude and less than 1.4 ps fall time. Diode sampling gate the sampling diodes into forward conduction. Since
bridges strobed by NLTL shock-wave generators have attained 1966 131, step-recovery diodes (SRD’s) [4] have been used
bandwidths approaching 300 GHz and have applications in for strobe pulse generation [5], [6], and the attainable
instruments for millimeter-wave waveform and network mea-
- 25-30 ps SRD pulse widths have limited sampling
surements. We discuss the circuit design and diode design
requirements for picosecond NLTL shock-wave generators and
NLTL-driven sampling circuits.
circuit bandwidth to - 15-30 GHz. The nonlinear trans-
mission line (NLTL) generates step functions more than
15 times faster than SRD’s, allowing instrument band-
widths approaching 300 GHz.
I. INTRODUCTION
While submicron lithography and advanced materials
ITH recent work in the GaAs/AlGaAs, InGaAs/ are variously required to attain 200-350 GHz transistor
AIGaAs/GaAs, and AlInAs/GaInAs/InP mate- f,,,,,, 2-3 THz Schottky diodes can be easily fabricated
rial systems [ll, transistor cutoff frequencies have ap- with 3-mask processes, large 2-3 Fm device dimensions,
proached 350 GHz, and 100 GHz [2] monolithic millime- and simple GaAs material technologies [71, [81. The cir-
ter-wave integrated circuits (MIMIC’S) have been cuits which can be implemented with such diodes will
reported. Both the transistors and the MIMICS have attain far larger bandwidths than circuits using 350 GHz
developed more rapidly than the instrumentation used to transistors. Using nonlinear wave propagation effects
characterize them. [9]-[14] on nonlinear transmission lines, the Schottky
In millimeter-wave transistor work, dc-65 GHz net- diode can be used for impulse or step-function compres-
work measurements are extrapolated, yielding estimated sion. The speed of such circuits is limited by the diode
100-350 GHz f,,. There is significant uncertainty in the cutoff frequency, and picosecond pulses can be generated.
extrapolation of power gain and of the circuit model, Because sampling bridges and NLTL’s require the same
impairing physical understanding. In MIMIC design above circuit elements, monolithic millimeter-wave sampling cir-
65 GHz, an extrapolated device model must be used, and cuits using NLTL strobe pulse generators can be fabri-
the circuit design will show poor correlation with the cated which attain bandwidths much greater than SRD-
measured MIMIC performance. Further, while device gated circuits [191-[211.
models can be extrapolated, circuit measurements cannot.
Sampling oscilloscopes and microwave network analyz- TRANSMISSION
11. THENONLINEAR LINE
ers use diode sampling bridges to down-convert the signal The GaAs nonlinear transmission line (Fig. l(a)) is an
MIMIC consisting of a high-impedance line of impedance
Manuscript received November 29, 1990; revised March 12, 1991. This
work was supported by a National Science Foundation Presidential Z , loaded at spacings d (spacings of T = d / uCpwin units
Young Investigator Award and by the Air Force Office of Scientific of time delay) by reverse-biased Schottky contacts serving
Research under Grant AFOSR-89-0394. Support was also received from as voltage-variable capacitors. Here uCpwis the propaga-
Tektronix, the Hewlett-Packard Corporation, and the Hughes Aircraft
Corporation. tion velocity of the high-impedance coplanar waveguide
M. J. W. Rodwell, R. Yu, M. Case, E. Carman, and K. S. Giboney are (CPW) line sections. Fig. l(b) shows the approximate
with the Department of Electrical and Computer Engineering, Univer- equivalent circuit, where
sity of California, Santa Barbara, CA 93106.
M. Kamegawa is with the Department of Electrical and Computer
Engineering, University of California, Santa Barbara, CA 93106, on L = Z 1 7 and C l = 7 / Z 1 (1)
leave from the Shimadzu Corporation, Kyoto, Japan.
IEEE Log Number 9100152. are the line section inductance and capacitance, and

0018-9480/91/0700-1194$01.00 01991 IEEE

I
RODWELL et al.: GaAs NONLINEAR TRANSMISSION LINES 1195

1
d

(C)
Fig. 1. (a) Circuit diagram, (b) equivalent circuit, and (c) layout of
nonlinear transmission line.
(b)
Fig. 2. (a) NLTL wavefront compression with step-function input.
C,(V) and R , are the diode capacitance and parasitic (b) Wavefront evolution caused by differing propagation delays of shock
wave and remaining waveform.
series resistance. The diode (small-signal) cutoff fre-
quency is u d= l / C , ( V ) R , at bias voltage V. NLTL's are
most readily implemented using coplanar waveguide lines is the propagation delay and C,(V) = C,(V)+ C, is the
(Fig. l(c)). total capacitance per NLTL section. The variation in
The NLTL is a nonlinear, dispersive, and lossy net- propagation delay, T ( V ) , results in the compression of
wo,rk. Frequency-dependent losses arise both from R , negative-going wavefronts during propagation. As an in-
and skin-effect losses. The periodic structure is an unde- put signal yn(t), a falling step function with initial voltage
sired low-pass filter with a cutoff (Bragg) frequency of V,, final voltage V,, and (input) fall time Tf,in propagates
wper = 2/\/L(C, + C,( V ) ) and strong group-delay dis- on the line, the fall time Tf:,at the nth NLTL section will
persion for u = uper.The variable diode capacitance in- at first decrease linearly with distance (Fig. 2(a)):
troduces a desired reduction in propagation delay with
increasing reverse bias, resulting in the reduction of the
Tf,n = Tf,in - n A' = 'f, in - n( T ( V ,) - T ( V, ) (4) >
fall time of waves propagating on the NLTL. where AT is the fall time compression per NLTL section.
Our work, engineering of functional monolithic NLTL's, After a sufficient number of NLTL sections, (2) and (4)
draws heavily on extensive analyses by Landauer [9], predict that the fall time will become less than zero. At
[22]-[241 and others [10]-[121, [14], [25]. We will restate this point, the voltage K ( t ) becomes discontinuous, form-
(without proof) needed results from the literature in ing a shock wave, and (2) and (4) apply only to the
terms of GaAs monolithic structures. continuous portions of V,(t) outside of the shock wave-
front. With large u, and wper, the number of NLTL
A. Shock Wave Formation sections necessary to form a shock wave is n = Tf,in/ A T .
Wavefront evolution during the compression phase,
B. Shock Wave Propagation
necessary NLTL length for wavefront compression, and
the NLTL characteristic impedance can all be predicted Shock waves have propagation properties determined
neglecting w, and uper. by the average or large-signal diode capacitance Cis:
First consider an idealized nonlinear transmission line
where R , = 0 and uperis made very large by simultane-
ously reducing both the diode spacings, d , and the diode
capacitances, C,(V), such that the diode capacitance per After formation, the shock wave will have a propagation
unit length, C , ( V ) / d , remains constant. In this limit of a delay per NLTL section of [22]
nearly continuous line [221, the voltage at the nth diode is
*lS(V,7V,) = ~ ~ l ( c l s ( V , *~ ~ ~ (6)
~ + c l )
V,( t ) = Vn(t - nT( V ) ) (2)
Unless V J t ) has the same time variation of voltage as
where T ( V ) (eq. (311, the shock wave will not form simultane-
ously over the full waveform amplitude, but instead will
(3) first form as a partial-amplitude shock wave whose ampli-

-
1196 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES. VOL. 39, NO. 7, JULY 1991

tude is a fraction of total signal voltage (Fig. 2(b)). The

partial shock of Fig. 2(b) between VI and V, will have a
per-section propagation delay of Tls(Vl,V2), while the
more negative voltages in the vicinity of V3 have a smaller
propagation delay T(V,), and the point of the waveform
at voltage V3 will thus eventually join the shock wave-
front, The partial shock amplitude will grow with propa-
gation, and a shock wave of full amplitude is eventually
formed.
The variation in Cd(V) introduces a voltage variation in
the small-signal characteristic impedance Z o ( V )
= d L /( C, + Cd( V ) ). For step functions propagating on
the NLTL after shock formation, the voltage and current diode 40
I,, are related by a constant large-signal wave impedance 0 , , diode 20

Z,, [241, [NI, ZJt) = Zdc+ ~ < t > / Z , , < vV,),

, , where
I

Shock waves are propagated without reflection or distor-

tion from the NLTL to the load if Z,,(V,, Vh)= ZlOad. -12
50
I ,
100
' . . . I
150
' ' ' ' I
200
' " ' I
250
0
Time. ps
C. Fall Time Limitations (b)
With nonzero wper and wd the shock fall time will
asymptotically approach a minimum compressed fall time,
Tf,min, at which wavefront compression AT is balanced by
the wavefront spreading associated with wper and wd.
-
If the diode cutoff frequency dominates (wd < wper),
the minimum compressed fall time can be found analyti-
cally, assuming uniform Schottky diode doping [25], [18],
[IT]. In this case cd(V) = cj0/ Jm, and
-7.5volts 1 J
8.8 1
Tf,min(10% - 90%) = - (8) 0 PS Tim 150 ps
wd(o) d m - 1
(C)
assuming that V, = 0 V. Given a typical GaAs Schottky Fig. 3. SPICE simulations of fall time compression showing the rela-
diode with 4 = 0.8 V and wd(0)/2r = 1 THz, (8) predicts tive effects of wper and O J ~ , , ~(a): 2.9 ps fall time NLTL shock formation

a 0.73 ps fall time for a 6 V step-function input. For more with diode cutoff frequency dominant; uper/ 2 ~ r= U ~ , , ~ =/ 500 ~ T GHz.

general diode doping profiles, Tf,mina l / w d and circuit (b) 2.4 ps fall time NLTL shock formation with Bragg frequency domi-
r 2 THz. (c) 0.9 ps fall time NLTL
nant; uwr/ 2 ~ r= 100 GHz, o d , l/ s2 ~ =
simulations are used to predict Tf,min. Because of the shock formation with Bragg and diode cutoff frequencies chosen to have
voltage variation in cd(V), varies with voltage. We comparable effects on compressed fall time. ope? / 2 ~ r= 500 GHz,
J ~ , ~ =~ 2 THz. Partial shock formation is also evident. Voltages are
use a large-signal cutoff frequency, wd,/, = l/RdCls, rather Oshown / ~ T
at the input, at 15-diode intervals, and at the output.
than w d ( 0 ) as a standard of comparison between diode
designs because stronger correlation between Tf,minand
w ~ , ,is~ seen in circuit simulations. Fig. 3(a) shows a and the duration of the shock-wave ringing. With wper -=K
SPICE simulation of shock formation on a 147-section wd,ls and with either impulsive or sinusoidal inputs, trains
NLTL with zl,(o, - 6 v) = Zload = 50 0, and wper/ 2 =~ of solitons are generated [12], [26]. While impulse com-
wd,,,/2r=500 GHz. For a 0 to -6 V, 20 ps Tf,,,, pression on NLTL's has been demonstrated [27], 1281
(10%-90%) input, a 2.9 ps fall time output with zero using soliton propagation, the topic is beyond the scope of
overshoot is generated. this text.
If the Bragg frequency dominates (uper-=Kwd,lS), then Circuit simulations indicate that w d , / , and wper have
7",mina l / w p e r and shock wavefronts show strong ringing comparable effects on T',minif the diode cutoff frequency
at the Bragg frequency. Fig. 3(b) shows a SPICE simula- is four to six times the Bragg frequency. With and
tion of shock formation on a 50-section NLTL with wper in these proportions, shock waves are generated with
AT(0 V, - 6 V ) = 50 ps, zl,(o,- 6 v)=Zload 150 0, moderate ringing. Fig. 3(c) shows a SPICE simulation of
u p e r / 2 r = 100 GHz, and wd,,,/2r = 2 THz. The input NLTL shock formation on a 147-section line with
fall time is 50 ps and the output has a 2.4 ps fall time and w p e r / 2 r = 5 0 0 GHz, ~ ~ , ~ , / 2 rTHz, = 2 and a 0.9 ps
shows strong and sustained ringing. Increasing w d ,Is rela- output fall time. The formation and growth of a partial
tive to wper progressively increases both the magnitude shock wave are also evident.

I
RODWELL et al.: GaAs NONLINEAR TRANSMISSION LINES 1197

Ohmic Ion Schottky

Contacts Implant Interconnect Y %2

e!
Q

P
75n interconnect impedance
F
N- Layer 0 Semiza:ating
o ~ , , , I , ,I
0 10
, , ,
20
I , , , , , , , , ,
30 40 50
qN,x;/e, Volts
N+ Layer Ohmic Metal Schottky Contact 6
interconnect Metal Fig. 5. Normalized NLTL compression K as a function of diode dop-
ing parameters. Z,, = 50 R, Z , = 90 R, V, = - 7 V.
Fig. 4. NLTL fabrication sequence. Ohmic contacts to the N i layer
are followed by a proton isolation implant and a Ti/Pt/Au lift-off
which creates both the Schottky contacts and metal interconnections. A where 4 = 0.8 V, E = 136, is the GaAs permittivity, and
is the process minimum feature size.
A is the diode junction area.
Combine (11, (41, and (7) to find the NLTL fall time
111. DIODEDESIGN compression AT normalized to the propagation delay of
For picosecond shock generation, the diodes must have the unloaded high-impedance C P W
several terahertz cutoff frequencies, strongly varying de- AT(KI,K) -
-ucpwAT(V,,V,)
pletion capacitance, and a breakdown voltage sufficient to K=
7 d
support the - 5 - 1 0 V shock wave amplitudes. These
parameters depend upon the diode doping profile and
geometry.
First consider the fabrication sequence (Fig. 4). Starting
with a semi-insulating GaAs substrate, a heavily doped where
N+ buried layer is grown, which serves as the diode I T7 \2
cathode connection. A surface N- diode active layer,
preferably with a hyperabrupt doping profile, is then
grown. Ohmic contacts to the N+ layer are formed by a
recess etch through the N- layer, a self-aligned AuGe/ and ucPw= c / f i = 1.13(108) m/s is the CPW propaga-
Ni/Au lift-off, and a subsequent anneal. Proton implan- tion velocity. Since the required NLTL length for shock
tation converts the N+ and N- layers to semi-insulating formation is
material [291, defining diode contact areas and eliminat- . I = Tf,ind/AT = Tf,inucpw/~ ( 12)
ing N+ and N- layer conductivities adjacent to transmis-
sion lines, which would otherwise cause substrate conduc- K is maximized to reduce NLTL die area and minimize
-
tivity losses. The 1.4 p m maximum proton penetration
depth [29] with typical 200 keV ion implanters limits the
CPW skin-effect losses. With V, = 0, normalized compres-
sion (Fig. 5) is a function of Z,, /Zl, V,, and the doping-
combined N+ and N- layer thicknesses. During implan- dependent parameter qZVox,'/E (if ~N,x,'/E< V, + 4, the
tation the diode regions are protected by a 1-2 p m Au diode depletion region extends into the N+ layer and (9)
on polyimide mask. Interconnections are formed with a does not hold). As shown in Fig. 5, diodes with more
Ti/Pt/Au lift-off; Schottky contacts result where this abrupt doping profiles yield larger compression and
metal layer overlays the unimplanted N- layer. Diode therefore reduce the required NLTL length. Reduced die
contact areas are independent of mask misalignments, area and reduced skin effect losses are thereby attained.
.and the only critical alignment is the lateral (in the Unfortunately, as the diode doping profile is made more
orientation of Fig 4) registration between the ohmic and abrupt (qiV,xi / E decreased) the diode cutoff frequency
Schottky masks. Diode minimum dimensions of 2-3 p m will also decrease.
are easily attained in processing and are sufficient for Given minimum diode dimensions (process resolution)
wd,ls/2r = 2-4 THZ. of A (Fig. 41, we can calculate wd,ls. Given A = wA, Cd(V)
Given an exponentially graded hyperabrupt doping pro- is calculated from (9). Series resistance includes ohmic
file N;(x) = N,exp(- x / x J [8], the diode capacitance contact resistance R, = p, /2w, where p, is the specific
is given implicitly by contact resistivity. The series resistance of the N+ layer,
including spreading resistance under the Schottky con-
tact, is R N + =7p+A/12wT+, where p + and T + are the
resistivity and thickness of the N + layer. T+ is limited by
the maximum proton implant depth available, while p + is
1198 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 39, NO. 7, JULY 1991

10 , , I , l I I I I I 8 , I I I , I > ,I > , / , 1 1 1 1 , 30
3 ~ " " " ' " " " " " " " ' ' ' " ' ~ ~ l " " L

N
r = l . l , 90R interconnections,
I
I-
x=0.75, 75R interconnections. -25 D
i
c l=lpm I
a 20
U
?!
c

5a 15
-
2
._
10
ol
U)

W
e 5
1

0
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 0 1 2 3 4 . 5 6
normalized compression, K Surface doping, N o ,x 10i7/cm3
Fig. 6. Large-signal cutoff frequency versus normalized NLTL com-
pression, with surface doping No as a parameter. h = 3 p m Z,, = 50 R, Fig. 7. Large-signal cutoff frequency and approximate breakdown volt-
Z1=90R,1/,=-7V. age versus surface doping No and process resolution A. Normalized
compression is constant ( q N , x ; / c = 14.1 V) at 1.1 for Z , / Z l s= 90
R /50 R, 0.75 for Z , /Zls = 75 R/50 R. V, = - 7 V.
constrained by doping limits in growth. The portion of the
N + layer between the bottom of the depletion layer and
the N--N+ interface contributes a resistance at zero bias
of
RN-=/'-p-(N;(x))dX =/
T-p-(Noe-"/"o)dx (13)
xJo XI0

where p - ( N ; (XI) is the doping-dependent N- resistivity

and x j o is the zero-bias depletion depth. R N - is rapidly
reduced as the diode is reverse-biased; since this cannot
be modeled in SPICE, circuit simulations pessimistically
assume the full zero-bias resistance. To minimize RN-,
the N- layer thickness, T-, is set at the diode depletion
depth at the maximum negative voltage 6. In our process, (C)
currently T++ T - = 1.4 pm, pc = 20 Cl.pm, and p + = 7.5
R ' p m (6 X 10"/cm3 N + doping).
For a given process resolution A, higher compression (b)
results in lower diode cutoff frequency and there is an Fig. 8. (a) NLTL cell layout showing interconnection parasitics. (b)
optimum surface doping dependent upon A and K (Fig. Parasitics will dominate if the CPW dimensions a and b are held
constant when the Bragg frequency is scaled. (c) To proportionally
6). For 2, / Z l s = 90 R /50 R normalized compressions the CPW dimensions must also be
reduce parasitics while scaling uper,
of 1-1.1 are attained without large sacrifices in w d , l s . scaled.
Fixing K = 1.1, Fig. 7 shows m d , l s / 2 ~versus process
resolution. To exploit smaller diode dimensions, surface
factors are interrelated. Microstrip circuits require
doping must be increased and (holding K constant) xo
through-wafer holes (vias) for ground connections having
and T - decreased. The avalanche breakdown voltage
then decreases (Fig. 7). For No above 5(10")/cm3 - - 10 pH inductance ( j l 8 R at 300 GHz) for a 100-pm-
thick substrate. Monolithic NLTL's are thus implemented
tunneling currents become substantial, and the reverse
using CPW lines. The cell layout is shown in Fig. 8. The
breakdown voltage is then smaller than shown in Fig. 7.
dimensional ratio (Fig. 8) a / b is set by the CPW charac-
Diodes with minimum dimensions of A=2-3 p m can
teristic impedance, which on GaAs is
attain 10-15 V breakdown voltages and 2-3 THz cutoff
frequencies, while maintaining reasonable normalized
compression. If the process resolution is improved to
A = l pm, diode cutoff frequencies can be increased to
approximately m d , l S / 2 r = 4-5 THz before the break-
down voltage becomes too small to support 5-7 V shock for 0 G k G 1/&, where k = a / ( a +2b). The CPW
generation. ground plane is extended toward the center conductor to
make contact with the diode ohmic contact. Both induc-
IV. NLTL LAYOUT tance in series with the diode and capacitance between
Circuit layout introduces parasitic series inductance the CPW center conductor and ground plane extension
and shunt capacitance at the diode locations. CPW skin (fin) are introduced. Neither parasitic is readily calcu-
loss is a major parasitic which must be minimized. These lated, but both increase as the CPW lateral dimensions
RODWELL et al.: GaAs NONLINEAR TRANSMISSION LINES 1199

are increased, and these parasitics will become compara- 2 " ' ~ " " " ' ~ " " " '
ble to the CPW inductance L and capacitance C, if the

(where uCpw= 1.13(108) m/s), while the diode junction

area must be chosen to give a large-signal capacitance of
120 160 200
1 I -72\ 0 40 80Time, ps

Small Tf,minrequires high wper and hence small d. The 2

CPW ground-signal separation b then becomes small,
as does (with fixed Z,, eq. (14)) the center conductor 0
width a .
The NLTL small-signal attenuation arises from diode
resistance and skin-effect losses, the latter increasing as -2
l / a . With n sections the attenuation is [8]
-4
Szl(w , V ) = exp [ - n w 2 C i ( V )R,Zo( V ) / 2

- ~ I / 2 Z O ( V l (18) -6
where rI is the center conductor skin resistance
-8
0 50 100 150
Time, ps
t is the metal thickness, and the total NLTL length, (b)
1 = nd, is given by (12). The high Bragg frequencies re- Fig. 9. SPICE simulations comparing step-tapered and exponentially
quired for small 7",,i, force narrow CPW dimensions and tapered NLTL's. (a) Shock formation on a step-tapered NLTL with 50
ps compression and having 30 sections with wper = 125 GHz and 40
high CPW losses. The losses not only decrease output sections with wper = 500 GHz. Ringing generated by the first line section
amplitude but also increase output fall time because the (diode 31) propagates to the line output. 25 ps input and 0.9 ps output
reduced signal swing reduces the fractional variation in fall time. (b) SPICE simulation of shock formation on an exponentially
tapered NLTL with 50 ps compression and with wper grading from 125
diode capacitance. GHz at the input to 900 GHz at the output. The 0.7 ps fall time output
The wavefront fall time decreases with distance (Fig. shows only small oscillation.
3); hence high wper is required only near the NLTL
output, where the transition times are short. To minimize
losses while attaining short fall time, we use low-loss nth NLTL section = k w p e r , n - l ( k> 1) increases
NLTL cells with low wper near the input and higher loss continuously with distance. The local Bragg frequency is
(high uper)cells with a small T',min near the line output. kept much larger than the inverse of the wavefront fall
An NLTL using two cascaded sections [17], [SI with low time at that point, and overshoot and oscillation are
and high wper attains lower losses than untapered struc- greatly suppressed (Fig. 9(b)).
tures. Because the compressed fall time approaches Tf,,in With fixed input fall time, Tf,in,increasing the intercon-
(Tf,mina l/wper), significant ringing (Fig. 9(a)) is gener- nect impedance 2, increases the normalized compression
ated in the first line section with low uper.The second (eq. (10)) and hence reduces the required NLTL length
line section has smaller Tf,min,thereby further compress- (eq. (1211, but the higher impedance results in narrower
ing the shock fall time, but the ringing generated from the CPW conductor widths a (eqs. (14) and (15)). Total NLTL
first line section is not suppressed. For picosecond instru- skin loss is proportional to the sum C ( d / a ) over the
mentation, minimum ringing is desired. Ringing can be NLTL length, and loss is minimized given the appropriate
reduced by varying the diode spacing, as suggested in choice of interconnect impedance 2, (Fig. 10). 2, of
private communications by J. Go11 (Tektronix). In an approximately 75 Cl maintains low skin losses and short
exponentially tapered NLTL the Bragg frequency at the NLTL length.

7
1200 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 39, NO. 7, JULY 1991

0 Schottky
contact I
interconnect

50 60 70 80 90 100 110
Z,,CPW interconnect impedance, fl
Fig. 10. Minimizing NLTL loss: NLTL length and number of squares
of metal (length/width aspect ratio) of the CPW center conductor, as a
function of Z, for an exponentially tapered NLTL as in Fig. 9. The
NLTL has 50 ps compression, Z,, = 50 R, V, = 0 V and V, = - 7 V, and
the diode doping parameter is q N o x i / c = 14.1 V.

Signal input
Fig. 12. Sampling circuit layout.

In the circuit (Fig. 11) an NLTL (using the design of

Fig. 9(b)) compresses an input strobe signal, either a step
function or a microwave sine wave (10%-90% fall time =
0.3/fi,). The sampling diodes must be gated by a pair of
symmetric positive and negative impulses. These are gen-
Transmission erated from the NLTL output using a balun/differentia-
Line
tor network implemented using the coplanar strip (CPS)
mode of the input signal coplanar waveguide (CPW).
Coupled through the network R,, C,, and C,, the strobe
Fig. 11. Sampling circuit schematic diagram. step function is applied between the CPW ground planes,
and propagates on them in both directions as a CPS
mode. At a distance dshortfrom the sampling diodes, a
V. MILLIMETER-WAVE
SAMPLING
CIRCUITS
short circuit (on the left) and R , and R , (on the right)
Sampling circuits as used in microwave sampling oscil- reflect the CPS mode, generating an impulse at the sam-
loscopes [3], network analyzers, counters [5], and fre- pling diodes of duration equal to the 2dshort/ ucPw
quency synthesizers (as sampling phase detectors [6]) are round-trip delay.
both the primary application for the NLTL [19]-[21] and The sampling diodes ( D 2 and D 3 ) are held in reverse
the best tool for NLTL characterization. bias by two high-impedance supplies connected through
R , and Rio. The complementary strobe pulses are cou-
pled through the large capacitors C, and C, and applied
A. Design Considerations
across the sampling diodes, which are then driven into
A sampling circuit consists of a strobe pulse generator, forward conduction for a period comparable to the dura-
a diode/resistor bridge, and a balun/differentiator. In tion of the strobe impulse. During this period, the sam-
addition to the NLTL requirements of diodes and line pling aperture time, the input (RF) signal partially charges
sections, direct implementation of a sampling circuit re- the coupling capacitors. If the repetition frequency of the
quires process steps for coupling and hold capacitor fabri- RF input signal is a multiple nfo of the strobe frequency
cation and air bridges for interconnections. With appro- (local oscillator, or LO), at each successive strobe interval
priate circuit layout, these additional process steps can be the sampling diodes will further charge the coupling ca-
eliminated, allowing simple fabrication of monolithic sam- pacitors C, and C,, and the common-mode (average)
pling circuits. The circuit diagram is shown in Fig. 11 and voltage of the two sampled outputs (coupled through R F
its layout in Fig. 12. isolation resistors R , and Rio) will asymptotically ap-

1
RODWELL et al.: GaAs NONLINEAR TRANSMISSION LINES 1201

proach the R F input voltage. The repetition frequency of fall time. Larger round-trip times broaden the strobe
the RF input signal is then offset in frequency by A f from impulse while shorter round-trip times reduce the im-
a multiple nfo of the strobe frequency fo, and the sam- pulse amplitude without significant reduction in the im-
pling signal is then mapped out in equivalent time at a pulse duration. In the absence of sampling diode para-
repetition frequency of A f. sitics, the strobe (full width at half maximum) impulse
To simplify processing, the sampling coupling and com- duration is then the NLTL output fall time. The diode
pensation capacitors C,-C, can be replaced with capacitances and series resistances further broaden the
reverse-biased diodes. Air bridge wiring crossovers can be strobe impulse at the sampling diodes. With given strobe
eliminated by using diode N+ buried layers. Where impulse duration, increasing the diode reverse bias de-
crossed by the strobe line, the buried Nt layers of com- creases the duration of forward conduction, and with
pensation capacitors C, and C, provide continuity for the reverse bias approaching the strobe peak amplitude, the
input signal CPW ground planes. The N+ layer of cou- aperture time can be reduced to a fraction of the strobe
pling capacitor C , connects the input signal CPW ground impulse duration (hence strobe NLTL fall time). Mono-
planes where they are crossed by the first sampled signal
(IF) output, and the second IF output crosses beneath the
lithic fabrication permits - 3-10 f F sampling diode ca-
pacitances, yielding subpicosecond signal line RC time
CPW ground plane using the N+ layer of C,. The maxi- constants, while a monolithic NLTL strobe generator
mum linear input voltage is small; hence a 20: 1 input permits picosecond aperture times.
attenuator ( R , , R,, and R 3 ) is used to allow testing
-
NLTL output signals of 5 V. R , and R , of the sampler
input attenuator short the CPW ground planes, eliminat- B. NLTL and Sampling Circuit Results
ing the need for an air bridge at that position. Picosecond pulse generators require picosecond instru-
Since RF input voltages greater than the sampling ments for their characterization. To evaluate the rise time
diode reverse bias will inadvertently forward bias the of our circuits, the output of an NLTL pulse generator is
sampling diodes, Vbias sets the maximum input voltage. To connected to an on-wafer NLTL-gated sampling circuit.
ensure diode conduction during the sampling interval, In this manner, we measure the combined (convolved)
Vbiasmust also be less than the strobe impulse magnitude responses of sampling circuit and NLTL shock-wave gen-
developed at the sampling diodes. Hence, the sampling erator. The experimental results of NLTL and sampling
circuit dynamic range is set by the NLTL strobe pulse circuit are inseparable.
amplitude and the differentiator efficiency. Attained NLTL‘s were fabricated using A = 3 p m design rules on
-
NLTL output voltages are 2: 1 smaller than the strobe semi-insulating GaAs material with a 1-pm-thick
amplitudes used in SRD-gated sampling circuits, and the 6(1018)/cm3 doping N+ layer, and an N- layer with
differentiator/balun must therefore have high efficiency. No = 2(10”)/cm3, x o = 225 nm, and T- = 425 nm. These
Without the matching and compensation capacitors C , parameters are not optimum for cutoff frequency
and C,, the peak stobe impulse magnitude is a fraction of ( ~ ~ , ~ ~=/ 1.7 r Fig. 7), but were chosen to maintain
2 7THz,
the NLTL output voltage, through voltage division be- moderate breakdown voltage (15 V, predicted, Fig. 7,
tween the R , = 50 fl NLTL termination resistor and the versus 12.5 V measured) required for other (soliton com-
parallel impedances Z,,, / 2 = 50 R of the counterpropa- pression [271) circuits on the wafer. The exponentially
gating CPS modes. The strobe impulse is further attenu- tapered NLTL design of Fig. 9(b) was used (125 GHz
ated through the parasitics of the sampling diodes. Adding input and 900 GHz output Bragg frequencies, AT = 50 ps,
capacitors C, and C,, a C - L - C .rr-section matching 0.7 ps simulated Tf,min,5.5 mm length, Z , = 90 0).
network is formed from the two capacitors, the shorted Small-signal microwave scattering parameter measure-
CPS line inductance, and the termination resistor R,. ments of the NLTL are useful predictive tools of perfor-
Nearly the full NLTL output voltage can be developed mance of the NLTL in wavefront compression. Circuit
across the sampling diodes while maintaining a low reflec- element values are also determined by fitting the mea-
tion termination to the NLTL. sured parameters to the calculated S parameters of the
The sampling circuit rise time (hence bandwidth) is NLTL circuit model. The NLTL group delay (Fig. 13) is
determined by the signal line RC time constant and by nearly independent of frequency, and shows the expected
the aperture time. The two sampling diode capacitances strong variation with voltage. The change in group delay
load the R F line, thus introducing a 10%-90% rise time over the 0-6 V range indicates 47 ps compression, hence
contribution of 2.2(Z0/2)2C&,de(I/bias), where Z, = 50 Cl a measured normalized compression of K = 1.0, 10% lower
and Cdiode(Vbias) is the sampling diode capacitance. Aper- than the design value. The measured small-signal inser-
ture time is determined by the NLTL-generated strobe tion loss is shown in Fig. 14; at the designed 10 GHz input
fall time, by the round-trip delay of the shorted-line and - 3 V bias the attenuation is a significant 4 dB. From
differentiating network, by broadening of the differenti- the variation of the attenuation with frequency and bias
ated strobe pulse by the sampling diode capacitance, and voltage the transmission-line skin-effect losses and the
by the diode reverse bias V,,ias relative to the impulse diode series resistance can be determined (eq. (18)). From
peak voltage. The round-trip time of the CPS shorted line the measured NLTL input match, SI,(Fig. 1.51, the volt-
differentiator should be approximately equal to the NLTL age-dependent characteristic impedance is determined:
 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 39, NO. 7, JULY 1991

Frequency, GHz
Fig. 13. Small-signal NLTL group delay versus bias and frequency.
determined by 45 MHz-40 GHz network analysis.
6 5 IQ 15 20 25 30 35 40
Time, ps
0 " " ' " ' " " " " " " ' " " ' " ' ' ' " ' ~
(b)
Fig. 16. NLTL pulse output measured by sampling circuits. (a) Results
for a wafer with uniform doping. (b) Results for a wafer with hyper-
abrupt doping. Substantial ringing occurs with the hyperabrupt wafer
because of repeated reflections on a misterminated transmission line
section.

Sampling circuits were fabricated on the same wafer

using an NLTL strobe pulse generator. A second NLTL
provides a test signal input so that the combined fall time
0 5 10 15 20 25 30 35 40 (rise time) of the NLTL and sampling circuit can be
Frequency, GHz
measured. The 3 p m by 3 p m sampling diodes have a
Fig. 14. Small-signal NLTL attenuation versus bias and frequency. 7.5 fF capacitance, introducing a 0.8 ps signal line RC
rise time. With the diodes biased to maintain 1 mA peak
impulse current, SPICE simulations predict an aperture
time of 0.85 ps with a 1 V reverse sampling diode bias.
The predicted 10%-90% fall time (or rise time) of the
sampling circuit is then TSampler = = 1.17 ps,
corresponding to 300 GHz -3 dB calculated bandwidth.
In testing, synchronized microwave synthesizers drive the
strobe NLTL and the test signal NLTL offset in fre-
quency to yield a 60 Hz I F output frequency. The mea-
sured response is the NLTL output waveform convolved
with the sampling circuit impulse response.
Owing to MBE growth difficulties, the N- doping on
the first wafer processed was 2(10")/cm3 with negligible
grading [20]. At 9 GHz LO drive frequency, the measured
(convolved) response of the NLTL and sampling circuit
shows 1.9 ps 10%-90% fall time (Fig. 16(a)) [20]. As a
result of the exponentially tapered NLTL design, the
observed combined step response of the NLTL and sam-
pling circuit is very clean, showing only 24% peak-peak
Fig. 15. 45 MHz-40 GHz NLTL input match ( S , l ) at 0 and 6 V ringing at 167 GHz. Fig. 16(b) shows measured results for
reverse bias.
a more recent NLTL and sampling circuit, fabricated on
MBE material having the designed hyperabrupt profile.
= 35 R at 0 V bias, and = 55 R at - 6 V. The large-signal The circuit is operated at 7 GHz LO drive frequency and
characteristic impedance lies between these two extreme 23 dBm NLTL drive power. The 10%-90% measured fall
values, and is approximately 40-45 R, lower than the time of the convolved response is 1.4 ps, approximately
intended 50 R. The NLTL input reflection coefficient 35% shorter than that observed for the uniformly doped
(Fig. 15) is much smaller than that of step-recovery diode wafer. The measured step response shows substantial 167
pulse generators. GHz ringing (6 ps period). In this process run, termina-

l
RODWELL et al.: GaAs NONLINEAR TRANSMISSION LINES 1203

tion resistors R , and R , (Fig. 11) were much higher than [3] W. M. Grove, “Sampling for oscilloscopes and other RF systems,
DC Through X-band,” ZEEE Trans. Microwave Theory Tech., vol.
the intended 100 Q (i.e., 50 Q for R , and R , in parallel). MlT-14, pp. 629-635, Dec. 1966.
Round-trip reflections then result in the 170-pm-long (3 [4] J. L. Moll and S. A. Hamilton, “Physical modeling of the step
ps round-trip time) CPW connection between resistors recovery diode for pulse and harmonic generation circuits,” Proc.
ZEEE, vol. 57, pp. 1250-1259, July 1969.
( R 2 ,R 3 ) and ( R 7 ,R,) within the sampling circuit. The
[5] J. Merkelo and R. D. Hall, “Broad-band thin-film signal sampler,”
numerical Fourier transform of Fig. 16(b) yields a - 3 dB ZEEE J. of Solid-State Circuits, vol. SC-7, pp. 50-54, Feb. 1972.
bandwidth of 215 GHz for the combined response. To [6] S. E. Moore, B. E. Gilchrist, and J. G. Galli, “Microwave sampling
verify that the sampling circuit is operating within its effective for ultra-broadband frequency conversion,” Microwave
Syst. News, pp. 113-120, Feb. 1986.
linear dynamic range during the rise time measurements, [7] K. Lundien, R. J. Mattauch, J. Archer, and R. Malik, “Hyperabrupt
identical sampling circuits with bond-pad inputs were also junction varactor diodes for millimeter-wavelength harmonic gener-
tested under the same bias conditions with a 7 GHz ators,” IEEE Trans. Microwave Theory Tech., vol. MTT-31, pp.
393-397, 1983.
microwave sinusoidal input of larger amplitude than the [8] C. J. Madden, R. A. Marsland, M. J. W. Rodwell, D. M. Bloom,
NLTL output. No distortion was observed. and Y. C. Pao, “Hyperabrupt-doped GaAs nonlinear transmission
The bandwidth and rise time of the sampling circuit line for picosecond shock-wave generation,” Appl. Phys. Lett., vol.
54, no. 11, pp. 1019-1021, Mar. 13, 1989.
and NLTL cannot be uniquely determined from these
[9] R. Landauer, “Parametric amplification along nonlinear transmis-
data. The 1.4 ps 10%-90% measured fall time is consis- sion lines,” J . Appl. Phys., vol. 31, no. 3, pp. 479-484, 1960.
tent with a rudimentary sum-of-squares convolution of [lo] R. V. Khokhlov, “On the theory of shock radio waves in non-linear
the predicted 0.7 ps NLTL and 1.17 ps sampling circuit lines,” Radwtekh. Elektron., vol. 6, tw.6, pp. 917-925, 1961.
fall times, which leads to a predicted -
300 GHz sam-
pling circuit bandwidth. We estimate a sampling circuit
[ l l ] A. Scott, Active and Nonlinear Wave Propagation in Electronics.
New York: Wiley-Interscience, 1970.
[12] R. Hirota and K. Suzuki, “Theoretical and experimental studies of
bandwidth between 250 and 300 GHz. lattice solitons in nonlinear lumped networks,” Proc. ZEEE, vol. 61,
pp. 1483-1491, Oct. 1973.
[13] M. Birk and Q. A. Kerns, “Varactor transmission lines,” Engineer-
ing Note EE-922, Lawrence Radiation Laboratory, University of
VI. CONCLUSIONS California, May 22, 1963.
Owing to limitations in pulse generation technology, [14] R. H. Freeman and A. E. Karbowiak, “An investigation of nonlin-
ear transmission lines and shock waves,” J. Phys. D : Appl. Phys.,
the bandwidth of microwave instruments has not kept vol. 10, Pp. 633-643, 1977.
pace with advances in millimeter-wave transistors and 151 M. J. W. Rodwell, D. M. Bloom, and B. A. Auld, “Nonlinear
integrated circuits. Based on terahertz diode technology, transmission-line for picosecond pulse compression and broadband
phase modulation,” Electron. Lett., vol. 23, p. 109, Jan. 29, 1987.
the NLTL has generated Q 1.4 ps step functions and has
allowed the development of sampling circuits with
GHz bandwidths. Compared with previous technologies,
300 - 161 M. J. W. Rodwell et al., “Generation of 7.8 ps electrical transients
on a monolithic nonlinear transmission line,” Electron. Lett., vol.
24, no. 2, p. 100, Jan. 21, 1988.
the NLTL and NLTL-based sampling circuits have un- 171 C. J. Madden, M. J. W. Rodwell, R. A. Marsland, Y. C. Pao, and
D. M. Bloom, “Generation of 3.5 ps fall time shock-waves on a
precedented bandwidth, require a die area (6 mmX monolithic GaAs nonlinear transmission line,” ZEEE Electron De-
1 mm) typical of other GaAs microwave integrated cir- vice Lett., vol. 9, pp. 303-305, June 1988.
cuits, and require only a simple fabrication process. Com- [18] M. Rodwell, “Picosecond electrical wavefront generation and pi-
cosecond optoelectronic instrumentation,” Ph.D. dissertation, Stan-
mercial sampling oscilloscopes using NLTL strobe pulse ford University, Dec. 1987.
generators have recently become available [301. Using the [19] R. A. Marsland, V. Valdivia, C. J. Madden, M. J. W. Rodwell, and
NLTL, 300 GHz instrumentation for millimeter-wave sys- D. M. Bloom, “130 GHz GaAs monolithic integrated circuit sam-
pling head,” Appl. Phys. Lett., vol. 55, no. 6, Aug. 7, 1989.
tems can be developed. [20] R. Y. Yu, et al., “275 GHz 3-mask integrated GaAs sampling
circuit,” Electron. Lett., vol. 26, no. 13, pp. 949-951, June 21, 1990.
[21] R. A. Marsland, C. J. Madden, D. W. Van Der Weide, M. S.
ACKNOWLEDGMENT Shakouri, and D. M. Bloom, “Monolithic integrated circuits for
millimeter-wave instrumentation,” in Proc. 1990 GaAs IC Symp.
The authors acknowledge the contributions of their (New Orleans), Oct. 7-10, pp. 19-22.
past collaborators in the Stanford group, D. Bloom, C. [22] R. Landauer, “Shock waves in nonlinear transmission lines and
Madden, and R. Marsland [81, [161,, [171, [191, [21]. Thanks their effect on parametric amplification,” ZBM J. Res. Develop., vol.
4, no. 4, pp. 391-401, Oct. 1960.
are due to Y. C. Pao and J. Franklin (Varian) for the [23] R. Landauer and S. T. Peng, “Velocity modulation of propagating
growth of the hyperabrupt MBE material and to J. Go11 waves,” J. Appl. Phys., vol. 44, no. 3, pp. 1156-1161, Mar. 1973.
(Tektronix) for discussions concerning NLTL exponential [24] R. Landauer, “Phase transition waves: Solitons versus shock waves,”
tapering. J. Appl. Phys., vol. 51, no. 11, pp. 5594-5600, Nov. 1980.
[25] S. T. Peng and R. Landauer, “Effects of dispersion on steady state
electromagnetic shock profiles,” ZBM J . Res. Deuelop., vol. 17, no.
4, pp. 391-401, July 1973.
REFERENCES [26] A. C. Scott, F. Y. F. Chu, and D. W. McLaughlin, “The soliton: A
new concept in applied science,” Proc. ZEEE, vol. 61, pp.
[l] U. K. Mishra, A. S. Brown, and S. E. Rosenbaum: “DC and R F 1443-1482, Oct. 1973.
performance of 0.1 p m gate length AIInAs-GaInAs pseudomorphic [27] M. Case, M. Kamegawa, R. Y. Yu, M. J. W. Rodwell, and J.
HEMTs,” in Tech. Dig., I988 Znt. Electron Device Meeting (San Franklin, “Impulse compression using soliton effects in a mono-
Francisco), Dec. 11-14. lithic GaAs circuit,” Appl. Phys. Lett., vol. 68, no. 2, pp. 173-175,
[2] R. Majidi-Ahy et al., “100 GHz high-gain InP MMIC cascode Jan. 14, 1991.
amdifier.” in Proc. 1990 G d s IC Svmo.
~ (New
- Orleans).
. Oct. 7-10. [28] M. Tan, C. Y. Su, and W. J. Anklam, “7 x electrical pulse compres-
173-176.’ sion on an inhomogeneous nonlinear transmission line,” Electron.
1204 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 39, NO. 7, JULY 1991

Lett., uol. 24, no. 4, p p . 213-215, Feb. 18, 1988. Ruai Yu received the B.S. degree in electrical
D. C. D’Avanso, “Proton isolation for GaAs integrated circuits,” engineering from San Francisco State University
IEEE Trans. Microwace Theory Tech., vol. MTT-30, pp. 955-963, in 1988. From 1988 to 1989, he was with the
July 1982. AMPEX Corporation developing advanced
HP54124T 50 GHz Digitizing Oscilloscope, Hewlett-Packard Co., equalization techniques for high-speed record-
5301 Stevens Creek Blvd., Santa Clara, CA. ing channels. He is currently a graduate student
in the Electrical and Computer Engineering De-
partment at the University of California at Santa
Barbara. His main research interest is the devel-
opment of integrated circuits for milimeter-wave
instrumentation.

Michael Case received the B.S. degree in elec-

trical engineering in June 1989 and is currently
‘. ,
Mark J. W. Rodwell (M’89) received the B.S. pursuing the Ph.D. degree at the University of
degree in electrical engineering from the Uni- California at Santa Barbara. Since earning his
versity of Tennessee, Knoxville, in 1980 and the B.S., he has worked with Dr. Rodwell as a
M.S. degree from Stanford University in 1982. research assistant developing nonlinear trans-
From 1982 through 1984 he worked at AT&T mission line technologies and applications for
Bell Laboratories developing optical transmis- millimeter-wave and picosecond pulse genera-
sion systems. He received the Ph.D. degree in tion.
electrical engineering from Stanford in January
1988 and remained there as a research associate
until Seotember 1988.
Currently, he is an Assistant Professor in the
Department of Electrical and Computer Engineering, University of Eric Carman received the B.S. degree in elec-
California, Santa Barbara. His current research involves picosecond tronic engineering from Cal Poly, San Luis
electrical shock-wave and soliton devices, millimeter-wave generation Obispo, CA, in 1989. He is currently pursuing
and instrumentation, picosecond optoelectronic devices, optical probing the Ph.D. degree at the University of California
of electronics, and millimeter-wave HBT and HEMT circuit design. at Santa Barbara. For the past year he has
worked on microwave harmonic generation us-
ing nonlinear transmission lines.

Kirk S. Gibonev was born on July 19. 1960. in

Bakersfield, CA. H e received the B.S. degree in
Masayuki Kamegawa was born on September physics from the University of California at
27, 1958, in Shiga, Japan. He received the B.S. Davis in 1984. From 1985 to 1988 he developed
and M.S. degrees in applied physics from Osaka microwave modeling techniques and was in-
University, Osaka, Japan. in 1982 and 1984, volved in radiation effects research on GaAs
respectively. devices and integrated circuits at the McDon-
H e joined the Shimadzu Corporation, Central ne11 Douglas Astronautics Company. He re-
Research Laboratories, Kyoto, Japan, in 1984, ceived the M.S. degree in electrical engineering
where he developed semiconductor X-ray detec- from the University of California at Santa Bar-
tors for medical imaging systems. He is currently bara in 1990. He is currently pursuing the Ph.D.
a visiting scientist at the University of California degree there in electrical engineering. His research focuses on electro-
at Santa Barbara engaged in research on appli- .. optic measurement systems and nonlinear microwave and far-infrared
cations of nonlinear transmission lines and high-speed samplers. devices.