1).Bias Dependence of GaAs and Ip MESFET Parameters

8/20/2019 1).Bias Dependence of GaAs and Ip MESFET Parameters

1/9

Bias Dependenceof GaAs and InP MESFET Parameters

REINHART

W.

H. ENGELMANN A N D CHARLES

A.

LIECHTI,

SENIOR

MEMBER,

IEEE

Abstract-A comparative analysis of the bias dependence

aE

critical RF parameters in GaAs and InP metal-semiconductor

field-effect transistors MESFET’s) led to the following

conclu-

sions.

1) The drain-gate feedback capacitance in GaAs MESFET’s

f i

lower than in InP MESFET’s, because of a stronger tendency

i n

GaAs to form stationary Gunn domains at the typical drain bias

levels employed.

2) The drain-sourceoutput resistance in InPMESFET’s is lower

than

in

GaAsMESFET’s mainly forhigh drain current units, fac’t

which is linked to substrate related softer pinch-off behavior

i o

InP.

3 ) The current-gain cutoff frequency f ~ n the current satu-

ration range of the GaAs MESFETdecreases strongly with draiu

bias as a result of the formation of the stationary Gunn domain.

n

the InP MESFET, his effect is weaker. At the optimum bias, r sa

only

10-20

percent higher in InP MESFET’s than in GaAs ones.

LISTOF

SYMBOLS

Drain-gate capacitance.

Fringing edge capacitance of gate stripe.

Gate-source capacitance.

Capacitance of high-field region.

Depletion-layer penetration.

= ( 2 d J e f f / q n ) 1 / 2 ,

effective epilayer thickness.

Velocity-peak field.

Velocity-valley field.

Unilateral-power-gain cutoff frequency.

Current-gain cutoff frequency.

Intrinsic dc transconductance.

Drain current.

Metallurgical gate length.

Parasitic source inductance.

= 1- a)&,

length of depletion-layer fringing

Drain portion of depletion-layer length.

Length of high-field region.

Source portionof depletion-layer length

(effective length of gradual channel).

Epilayer carrier density.

Electron charge.

Parasitic drain resistance.

Parasitic gate esistance.

Intrinsic channel esistance.

Parasitic source esistance.

= Vbexp)

+

VB, ffective pinch-off potential.

Built-in potential.

beyond gate stripe.

=

VDS- VGS,

rain-gate bias.

drain-source bias.

Experimental saturation voltage.

Gate-source bias.

Experimental pinch-off voltage.

Electron velocity.

Peak velocity.

Valley velocity.

Channel width.

= Gnoe-jWTO,ntrinsic transconductance.

Permittivity.

Radian frequency.

I

I. INTRODUCTION

NTEREST in metal-semiconductorield-effect

transistors (MESFET’s) for microwave applications

has grown steadily in recent years [l] . After

Si

devices [2],

GaAs realizations were studied extensively because f their

higher frequency capabilities [3].

Initial work by Barrera and Archer [4] on the InP

MESFET with a 1-pm gate lengthevealed a 50-percent

increase of the current-gain cutoff frequency f T , beyond

best values generally observed inaAs devices of the same

geometry and doping (Table I).This compared to a theo-

retically predicted 30-percent ncrease in f T for the satu-

rated InP MESFET based on the idealized model of

Turner andWilson [6], [7] as result of a 47 percent higher

electron peak velocity in InP. On the other hand, the

combined influence of a much larger drain-gate feedback

capacitance1c d g , a smaller drain-source output resistance

R d , , and a somewhat larger input resistance R i

+

Rg+ R,,

in the InP MESFET,ave rise o somewhat lower available

power gain and hence a lower unilateral power-gain cutoff

frequency f m a x [4] since

[5]

fmax

f T / [ 2 d ( R i R g

+

R s ) / R d s + 2 r f T R g C d g l .

It

was desirable to determine if the substantially higher

c d g

and the ower

R d ,

values in IriP MESFET’s resulted

from fundamental material propertiesr from remediable

flaws in the materials and MESFET fabrication technol-

ogy. In addition, theelatively large discrepancy between

experimental and theoretically estimated

f T

values (Table

I) needed some explanation. The idealized theoretical

model is based on the hypothesis tha t the electron dri ft

velocity saturates in the channel a t its peak value

up

[6],

171. The effect of the actual elocity drop at fields beyond

Manuscript received March

3,1977;

revised July 6,1977.This wcrk the velocity peak was not considered, In ordero arrive at

was sup orted

in

part

by

USAECOM, Fort Monmouth,NJ, under Cc

n-

tract D-+ AB07-75-C-1300.

94304.

The authors arewith Hewlett Packard Laboratories, Palo Alto, (:A

8 1

1 For thedefinition of the equivalent circuit elements, see41.


2/9

ENGELMANN AND LXECHTI:

GaAs

AND InPMESFETPARAMETERS

1289

TABLE

I

Summary of Initial MESFET Comparison [

41

L

= 1

pm,

w

= 500 p m ,

n = 10

~ m - ~ ;

Ds =

0)

G a A s

n P

fl (GHz)

E x p e r i m e n t

130-24

l h e o r y

25 32

vp

(107

c m

s - 1 ) *.7 2.5

*Used in theoretical model for

f r

R / '

TABLE I1

Experimental MESFET Properties at DC

V D S

= 5v, VG , =

~

&:

l?_p

5162-6 1 . 1 x 1017 5200 0 .9 iO.l

5 8 . 5

0

SC126.7-14

1 . 0 x 10

3000

0.9

iO.l

43.0 0 . 92

SC126.7-18 1. 0

~ 1 0 ' ~

3000

1 . 0 0 , l 1 1 5 . 0 1 . 5

u

InP:

S A 6 2 - 6 1.15

44 4.0 0.97

2.1 (1.80 2 . 9

0.20

0.074

St l26.7-14

2.05

38

6 .5

1 . 80

1.9 0.50 2.4 0.18

0.060

SC126.7-18

3 .0

104

6.5

2.32

3.6 0.45 4.05 0.24

0.120

A P P R O X I M A T I O N S FOR M E S F E T I N S A T U R A T I O N :

R / '

( 1 1 G H z ) .

Ri + R, + Rg

LI Gm4

Cgs

C g s ( l l G H r ) .

Car

C& ( 3 G H z l

a

Cdg

Y ~ ( 1 l G H z l

Y =

Grn,r;wO

R& ( 3 G H z l a Rdr

Fig.

1.

MESFET

?r

circuit model with approximations for equivalent

circuit elements defined in4].

a deeper understanding f the striking differences found

between GaAs and InP MESFET's and their possible

relatilon to the

electron-drift-velocity/field

haracteristic,

we undertook acomprehensive study of important MES-

FET parameters as functions of drain-source and gate-

source bias.

In general, a determinationof the element values of the

equivalent circuit of a MESFET follows from a model

optimization procedure based on measured and calculated

s

parameters [4],8]. T o avoid this relatively lengthy pro-

cedure for each different bias condition, we estimated the

crucial equivalent circuit elements from a setof T (circuit)

parameters (Fig.

1)

ha t can be derived from the setof

s

parameters directly. The approximations of Fig.

1

are

based on a detailed investigationf the P parameters of a

saturated MESFET as function of frequency [9].

11.

PARAMETERS

F MESFET's STUDIED

The GaAs and InP MESFET's studied were fabricated

from the same mask set with a center-fed gate,ominally

1

pm long and 500 pm wide. Equivalent technologies were

used including a gate-trough etch to adjust the FET

channel hickness [4].Active n-type epilayerswitha

1

I

1

2

4 6

8 - 7 0

v,,/v

-

(a)

nP: S C 1 2 6 . 7 - 1 4

80

70

L ' 8 1

f

4 2

3

.

vv

-

(b)

Fig. 2. Drain current characteristics

ID

versus

VDS

t

VGS=

0 illus-

trating the definition

of

Vp&i.(a) GaAs MESFET. b) InP MES-

FET.

doping density close to

lOI7

cm-3 were grown n Cr-doped

semi-insulating substrates using liquid.-phase epitaxy in

both cases. The dc properties of one GaAs and two InP

MESFET's, which were investigated in detail, are sum-

marized in Table

11.

The second InP MESSFET is included

to show the influence of a large drain saturation current

on RF properties. The experimental drain saturation

voltage Vb , ,',s defined n a somewhat arbitrary but con-

venient way, by the drainbias at which1 the drain current

saturates towo-thirds the extrapolated ohmic value (Fig.

2). V@:ii

is this saturation voltage corrected for the esti-

mated voltage drop across the source and drain parasitic

series resistances R , + E d . A convenient experimental

pinch-off voltage

Vb""p)

follows from B rough linear ex-

trapolation of the transconductance versus gatebias


3/9

1290 IEEE TRANliACTIONS

O N

ELECTRON DEVICES,

VOL.

ED-24, NO. 11, NOVEMBER 1977

80

~ 8 0 1 1

V G, i

v

-

v,,i

v

-

Fig.

3.

Transconductance pinch-off behaviorat

VDS= 4

V illustrai ing

thedefinition of (intrinsic a circuit ransconductance

1

YL, at

11

GHz). (a) G aAskESFET . (b) InPMESFET.

TABLE I11

Experimental MESFET Parameters at RF, VDS

= 5

V, VGs :=0

G

:

I

SG6 2 -6 8 8

34 I 2 5 5

3

1

0 . 0 1 8 7 - 1 5 6 0 320 3.CC7 0 51

1

3

iCl25,1-14 4 . 3 18 1 6 s 5 0 3

1.3

0.110

1 . 0 2

3 3 5

0.1C6

0.45

I

SC I6,7-18

5.1

21

1 5 . 7 I

5 4 . 7

.

3 107 55

8

0 0 5 3 0 5 2 : 7

1 P '

curves2 (Fig.

3)

and the built-in potential

VB

from the

voltage dependence of the gate capacitance at V D S= 0. An

effective open channel thickness,ff is then derived from

the effective pinch-off potential U,ff = Vbexp) + VB.The

channel constriction

d,

at the drain end of the gradual

channel region isestimated from the Turner-Wilson model

[6],

[7].

Pertinent RF parameters a t the bias condition

V ~ I S

= 5

V, VGS

=

0

are listed in Table

111,

including the

T

~ a -

rameters of Fig. 1at the ppropriate frequencies.

111. FEEDBACK

APACITANCE

d g

The dependence of the circuit feedback capacitance

Czg

3 GHz) on the bias between the drain and gateV;IG

is shown in Fig.. Fig. 4(a) and b) compares the GaAs with

the equivalent InP MESFET or three bias conditions:

I.)

the gate shorted to the ource (

VGS=

0) and the drain bias

VDSaised into saturation circles);2) the channel current

kept a t zero

(VDS

0) and a negative gate bias (-Vs:g)

applied toward and beyond channel pinch-off (triangles);

3)

the drain bias

VDS

ixed in saturation and the gate bills

VGS aried (dots, dashed lines).Fig. 4(c) summarizes the

first case (VGS

=

0) including thehigh-current Inl.

MESFET.

Most striking is the strong decrease of C i g or the G d s

MESFET at VGS

=

0, extending beyond saturation and

covering almost an order

of

magnitude in the saturation

range (Fig. 4(a)). Thiss quite contrary to theehavior i n

the channel current-free state OS= 0. Here C;, also dt -

creases strongly as long as the depletion layer mows

There is a certain dependencef V(exp)on

DS[18];

e adopted VL

and theVDS ependence are inconsistent withhe Turner-Wilson model

= 4

V

asour standard for comparisonofdevices. The linear extrapolation

[6]

7 ] ,

but the simplified definition forV P s well justified for coa .

paratwe purposes.

Zero

Bias

V a lue

0.01

0.1

1 0 10

~ v D , - v G ~ l l v-

(a)

1

o

e

C

126 .

7-14

Zero

Bias

Value

I

Y

0

.

:s

0.1

v, = 4 v

I I

cvos-vG, l /v

(b)

0.1 1

o

1 0

1 o

v =

0

G S A 62-6

+ E:

C 1 2 6 . 7 - 1 4

'.\b

o

S C 1 2 6 . 7 - 1 8

\.

\

c

\\

\ \ \

0.011 I

0.1 1 0

v i v-

(C)

Fig. 4.

a

circuit drain-gate capacitance

C i

at 3 GHz versus drain-gate

bias, VDG VDS

-

VGS. (a) GaAs MES6ET. (b)

InP

MESFET. (c)

Comparison betweenGaAs and InP MESFET s at

GS

=

0.

toward channel pinch-off at

Vbexp),

confining the capaci-

tance to the ringing depletion-layer edge. However, more

gradual decrease follows beyond

Vbexp),

which is expected

from the lateral movement of this depletion-layer edge

toward the drain. Thus at given

VDG,

he GaAs MESFET

has amuch smaller

C g

n saturation than n the pinch-off

condition, particularly when

VDG

s large. This leads

t o

a

strong increase of

C i g

when a negative gate bias (-VGS)


4/9

AND LIECHTI

G U S

AND InP MESFETPARAMETERS

1291NGELMANN

a.)

b.)

S.I.

SUBSTRATE

Fig. 5. Depletion-layer fringing. (a) Before pinch-off. (b)After pinch-

off.

is

ap.plied to he aturated GaAs MESFET toward

pinch-off (dashed lines,Fig. 4(a)).

From purely geometrical electrostatic arguments for

drain-,side ringing of the gate depletion-layer capacitance,

no maljor difference in Cdg between the current saturation

and tlhe pinch-off state is expected. In fact, the bias de-

pendence of

C

for the InP MESFET’sFig. 4(b))seems

to agree at least qualitatively with such aimple geomet-

rical viewpoint. In the case of the GaAs MESFET, how-

ever, a fringing capacitance equivalence between the sat-

urated and the inch-off MESFET is retained only at the

limiting bias values, viz., at VDS V {l, VGS 0 , and at

VDS

= 0,

VGS -Vbexp),espectively. Thus theabove re-

sults strongly suggest that the

low

of charges in the lon-

gitudinal channel field between the drain and source is

respo:nsible for reducing the feedback capacitance

c d g

in

the saturated GaAs MESFET. Before discussing this

question further (Section 111-B), a more quantitative

analylsis of depletion-layer fringing is appropriate.

A . Fringing Capaci tance of Gate S t r ipe

The gate fringing capacitance can be estimated from

published calculations of the depletion-layer geometry of

a biased metal stripe n a plane semiconductor surfacef

infinite extension

[lo].

As indicated in Fig. 5(a), deple-

tion-l#ayer ringing tha t is relevant for

c d g

extends under-

neath the metal stripey a fraction

a

of the undisturbed

depletion-layer penetration dd. Deviating from previous

estim,ates

[2],

[ll] he total ringing capacitance

C j

is as-

sumed to be composedof this internal egion I1 inaddition

to the externalegion I. I t follows, independently of

d d ,

cj

= S€W

1)

where

]permittivity of the semiconductor,

w

stripe length (corresponds

t o

“width” of gate),

6 =

61

+ 611. 2)

For region I, including the effect of the air-semiconductor

interface [lo],

61

=

0.71 +

0.86 to/€ =

0.78. (34

For region

11,

we estimate

611

=

2a 0.70

(3b)

by considering the depletioncharge ncrement both

TABLE IV

Gate F ringing Capacit ance a t Diffe rent 13ias Condit ions

~ _= === ___________i_j ;l

~ -.-:~

===

l -

B i a sa p a c i t a n c e

C o n d i t i o n *a l u e s

GaAs

( S A 6 2 - 6 )

n P

(SC126 .7 -14 )

0

C /pF.10.23

Cf /pF 0.086 0.091

‘ f ,excess /PF 0.01 0.15

0 C iS / p F 0.30

0.42

C ig / p F

0.32 0 . 6 3

‘g,exp/PF

0.62 1.05

‘g , theo r /OF

0.58 0 .68

‘ f ,excess /PF

0 .02.19

*

B i a s o n d i t i o n :

V D s

= Vi::

G S = 0

o r

- _-

V D S =

0, V G S

= - v ( e x p )

Cf,excess = ‘f

@

V S

=

0,

VGS =

0

‘ f ,excess,expg , theor)

= L(C .

downward and to theeft in Fig. 5(a). We conjecture tha t

c d g - Cf

for the two limiting cases VDS

=

V :t, VGS

= o

and VOS=

0 ,

VGS

-

V p )Table

IV,

bias condition

0

Experimental

(C

at

3

GHz) and theolretical ( C j )values

are in good agreement for the GaAs MESFET; for the InP

MESFET, however, experimental values are more than

two times larger. Factors contributing to an increase of the

observed

c d g

values are the presence of a gate trough,

possible surface charge effects (both increasing

SI),

and

incomplete saturation or channel pinch-off (increasing

611).

In the InPMESFET’s, pinch-off behavior is indeed much

softer and, hence, less complete at

Vbe.xp),

han in GaAs

devices as is evident from Figs.

3

and 4..

To test the importance of gate-trough and/or surface

charge effects, we measured the totalgate capacitanceat

zero bias, which can be estimated from lCg exp

= C i s

+ C

at

2

GHz (Table

IV,

bias condition

@a .

For the GaAs

MESFET, again, agreement with the theoretical value,

Cg,theor, is satisfactory3 when fringing at a planar surface

is included [lo] . The InP unitn the other hand exhibits

experimental values that are about50percent larger. Here,

the excess in fringing capacitance per side is 0.19pF, which,

within the experimental error, matches with the excess

capacitance deducedfrom C at bias clondition0.EM

studies of the cross-sectioned MESFElT chips revealed

that most of the excess fringing capacit,ance results rom

the gate trough. The GaAs unit had a gate-trough depth

of only 0.16 pm, in contrast toa depth (of0.43 pm for the

InP unit. In addition, thenP trough exhibited an asym-

metry to themetallurgical gate that exp1,ains he difference

in C i sand C of Table

IV.

We thus conclude that gate-

trough etching shouldbe kept toa minimum to avoid an

Note that the xperimental accuracy of L (Table 11)is not better than

10percent.


5/9

1292

IEEE TRANS1,CTIONSNLECTRONEVICES, VOL. ED-24 NO.

11,

NOVEMBER 1977

increase in the drain-gate fringing capacitance. Surfice

charge effects appear tohave little influence if present a t

all.

B. I n f l u e n c e of a S ta t ionary Gunn Domain Region

After having attained an understandingf the geom s t-

rical fringing capacitance, we willnow return o he

problem of the experimentally observed large c d g diff x -

ence, for the GaAs MESFET, between the bias cases

w

th

and without current flow. The minimum value measured

for the GaAs unit is about c d g = 0.014 pF in current SEL LU-

ration at VDS

=

8 V, VGS= 0, which compares to c d g =

0.057 pF in pinch-off a t V D S=

0, VGS

= -8

V

(Fig. 4(s)).

For a large enough drain-gate bias

VDG,

he depletim-

layer penetration Id toward the drain can be estimated

from c d g assuming aparallel-platecapacitance C p of

separation l p in series with a fringing capacitance C;

w

(SI

+

~ ) E WFig. 5(b)). With ; = 0.062 pF (SI

=

0.71,

CY =

(0.25)

and deff 0.20 pm (Table 11),one obtains Id = 0.22 pm :'or

the pinch-off case, and a four times larger value, viz.,

1,

=

0.84 pm, for he current saturation ase. On he other hand,

estimating

d

from the drain-gate potential,

U D G

=

v , ~

+ VB

according to electrostatic arguments, viz., l d / l j w

(UDG/Ueff)lI2 U D G 8.8 V, U,ff

=

2.9V (Table

II),

a i d

f

= 1

-

a)d,ff

=

0.13 pm], yields

l d

= 0.23 pm, whic:h is

consistently only with the above value for the pinched-',ff

transistor. Thisproves quantitatively that the propertes

of the carrier flow have to be taken into considerationbr

explaining the low c d g values observed in the saturated

GaAs MESFET.

It has been shown by two-dimensional model calcu la-

tions [12], [13] that velocity saturation in the constricting

channel establishes a charge dipole layer acrosshe velocity

saturated portion of the channel. The Gunn effect en-

hances such a dipole-layer formation and rapidlynduces

fields beyond valley velocity saturation ( E

>

E ,

= 1 2

kV/cm) [14]. The finite scattering timesof the electrcns

cause velocity overshoot effects [15]-[17] resulting in a

positional shift of the charge dipole layer toward the drain

[ E ] .

In fact, high fields may even extend as far as di redy

below the planar metallurgical drain contact andstablish

valley velocity aturation over a substantialportion of the

entire region betweengate and metallurgical drain contacts

(according to [15] most electrons are ifted to theirheevy

state in this region). We thus conjecture tha t the rapid

decrease in c d g with drainbias beyond MESFET satulna-

tion is a direct result f the growing charge dipole layer.

The situation is qualitatively illustrated in Fig. 6

[ lE l ] .

The "gradual" portion of the channel terminates t a po-

sition where the field E reaches a value E,

5

E p core-

sponding to the sustaining field appropriate for a 8 1 a -

tionary Gunn domain EPs the field at the eak veloc: y

up of an effective d y n a m i c u ( E )characteristic). Furtker

down stream, electron accumulation under the constricting

channel rapidly ncreases the field to

E

> E,, causing the

electron drift velocity to saturate at itsalley value,

w --+

vu These high fields can penetrate far toward theedid-

E <

Ep E-E, E > E v

Y < V P

5

Y

x -

GRADUALCHANNEL UNNOMAIN

Fig. 6. Location and equivalent circuit f stationary Gunndomain.

lurgical drain contact since, ecause of the reduced drift

velocity, electron depletion due to channel widening need

not be large. t is thus expected that most of the additional

drain-gate potential appliedbeyond the MESFET satu-

ration voltage does not dropbetween the drain dge of the

gatedepletion layer and he gatecontact as in the

pinched-off MESFET, but across the high-field region

between the metallurgical drain contact and the deple-

tion-layer edge [18]. Dynamically, this high-field region

acts as a capacitance,ch between the drain and gate in

series with he fringing capacitanceC j of the gate depletion

layer. The dynamic parallel resistance Rh of the high-field

portion is considered to be negligibly high because of ve-

locity saturation. WithCdg

=

0.014 pF (VDS 8 V, VGS

=

0)and C f = 0.086 pF (taken rom

VDS

V&&

VGS

= 0),

this capacitance is Ch

=

0.017 pF, which corresponds to a

length of the high-field region of lh = 0.66 pm. Assuming

a potential drop cross the fringing gate depletion layer of

about 2 V (pinch-off voltage,V p ) ) ,he average field along

lh becomes larger han 90 kV/cm. This value compareswell

with even the highest valley velocityields for GaAsquoted

in the literature( E ,

=

85 kV/cm [19], also, cf., [20]).Even

though these estimates areairly rough, they demonstrate

that the high-field domain model is consistent with the

experimental data.

The bias behaviorof the feedback capacitance C:g of the

InP MESFET's, on the other hand, uggests that the for-

mation of the high-field dipole domain is fairly weak in

current saturation (Fig. 4(b) and (c)) and,ence, the ca-

pacitance is mainlydetermined by the gate depletion layer

as in the pinch-off case.

A

qualitative explanation ollows

from the differences in the drift elocity versus field rela-

tionship [20] between GaAs and InP: In nP the egative

mobility is typically a factorof two smaller and the ields

for valley velocity aturation are substantially arger, viz.,

by a factor of five. Thus substantially higher drain bias

levels than currentlypossible appear to be necessary for

In P FET's to induce similar Gunn domain effects as ob-

served in GaAs MESFET's.

The length

lh

of the velocity saturated region (Fig.6) can

also be estimated from the phase shift in the intrinsic

transconductance Y, if the delay in the restof the device

is neglected (cf., [21]). In the GaAs MESFET, the delay


6/9

ENGELMANN AND LIECHTI:

GaAS AND InP

MESFET

PARAMETERS

time ;ri (see Fig. 1)was found to increase with drain bias

consistent with he simultaneous decreaseof cdg A t

VOS

=

9 V, we obtained ;

=

5.6 ps which leads to

lh

= 0.85 ym

assuming a delay analogous to thatn a collector depletion

layer ‘ofa bipolar transistor [ ZS ] , h = 2uu7i,with

v u =

0.75

X l o 7cm/s (estimated a t 15 percent below the room tem-

perature value due to elf heating, cf., [ 2 5 ] ) .This is quite

consistent with the

0.66

ym deduced from

c d g .

On the

other hand, no clear trend for the VDS ependence of

7;

evolv’edwith the InPMESFET’s

[9].

Another fact that might contribute to reduced space

charge effects in InP MESFET’s is their very soft pinch-off

behavior mentioned previously (Figs.3 and 4 ) . The softer

pinch-off implies less constriction of the flow of charge

carriers and hence less tendency for space charge forma-

tion. Additionally, the high gate leakage currents that flow

in InP MESFET’s [ 4 ]at the larger drain-gate bias (Fig.2 )

are expected to affect the field distribution in the region

between the depletion-layer edge and drain and may

prevent a substantial uildup of field.

In conclusion, the main reason for the larger feedback

capacitance

I

in current saturation in InP MESFET’s

as compared to GaAs ones has been linked to a reduced

field buildup and, hence, to aess complete velocity sa tu-

ration a t the “valley” velocity in InP at the bias values

employed.

Iv.

OUTPUT RESISTANCE

Rds

T:he bias dependence of the T circuit output resistance

R i s

3

GHz) is presented in Fig. 7. The drain bias plots

(VGS

0,

Fig. 7(a)) exhibit about an equal relative increase

of R i s up to theaturation voltage Vg”dfor both GaAsand

InP. However, beyond

V@$,

the resistance R i s increases

further for the GaAs MESFET by a much larger factor

than for the InP ersions. This difference appears to result

from the softer pinch-off behavior in InP devices already

noted in Figs.

3

and 4,which is clearly evidenced again by

the pinch-off behavior of R i s in Fig. 7(b). In thisonnec-

tion, i t is of interest tonote tha t in InP devices

R i s

tends

to increase for units with lower drain current

ID .

Fig. 8

shows data taken for a number of fabrication runs covering

a large range in epilayer dopingand gate-trough depth and

employing a variety of different substrates. An inverse

proportional relationship between Rds and

ID

is expected

when the RF current simply scales with the dc current.

Such a relationshipemerges for higher current InP units

(Fig.

8).

In GaAs MESFET’s, on the other hand, variations

in substrate, interface quality, and gate-trough depthlay

a major role in determining

Rds

and no correlation to

10

evolved. Here, Rds appears to be strongly influenced by he

varying properties

of

an epi-substrate interface depletion

layer resulting from electron trapping. Drain current re-

laxation effects and gating by substrate bias (“back gat-

ing”) as observed in GaAs MESFET’s have been related

to tlhese interface properties [ 2 2 ] , [ 2 3 ] .

o

far, we were not

able to produce “back gating” n InP units.

These facts taken together lead to the tentative con-

clusion that no electron trapping occurs at theepi-sub-

1000 I

I I

I

v

= 0

t

:B;

.

1293

v 1v

(b)

Fig. 7. P circuit drain-source resistanceR;,

at

3 GHz: (a)versus drain

bias VDS t

VGS=

0; (b) versus gate bias

VGs

at VDs

=

4 V.

101

10

20

40 60 8 100

200

I

1,ImA

-

m

Fig. 8. P circuit drain-source resistanceR i , at 3

GHz

versus drain sat-

uration current

ZD

a t VGS

= 0 for

different InP MESFET uns.

strate interface in InP.This also explains the softer

pinch-off behavior since carriers can be injected into the

substrate more easily if no enhanced interface barrier is

present. The absence of electron trapping might be related


7/9

1294

IEEE TRANS.l,CTIONSON ELECTRON DEVICES, VOL. ED-24, NO.

11,

NOVEMBER

1977

v =

0

-.

*

GaAs: SA62-6

-.

nP:

SC126.7-14

SC126.7-18

ob

2 4

6 8 10

11

V,JV

-

Fig. 9. Current-gain cutoff frequency

f T

versus drain bias VDS t

V G S = 0.

to a different nature of the deep evels in the lower resilr-

tivity, Cr-doped InP substrates used (103-104 a-cm, LIS

compared to lo6-los Qcm or GaAs).

v. CURRENT-GAIN UTOFF

FREQUENCYT

The current-gaincutoff frequency

f T

was directly de -

termined from current-gain versus frequency curves

2s

deduced from the measured s parameters [9].As a function

of drain bias (Fig. 9), we observe a rise toward a peak 2 t

about 1-2 V above the saturation voltage. The decrease c f

f~

beyond this peak is particularly strong in the GaAs

MESFET. Thissuggests that thedecrease is related to the

Gunn domain formation discussed in Section 111. For

testing this contention [18],we consider the following re-

lationships for the intrinsic small-signal transconductanc

2

G,

(cf., [21], [24]) and for the gate-source capacitanc,?

cg,

:

and, hence,

5 )

Here,

V

is the average drift velocity in he gradual channe

region and I , is its effective length; is the average gate

depletion-layer penetration; and

Q

is the total charge irl

the gradual channel egion. It can be shown [18] ha t

k

hardly affected byhe drainbias. Thus the drainias de.

pendence of G,, and

Cg,

Figs. 10 and 11) s indicative

aj’

that of

V

and

l,,

respectively. Fig. 10 then clearly indicates

that

i7

falls off more strongly in GaAs than in I nP with

drain bias VDS beyond its maximum value.This reduction

of

5

with drain bias can be considered as evidence for the

formation of a Gunn domain at the drain end of the

channel: Because of current continuity, the lower carrier

velocity in the high domain fields requires a simultaneous

drop in carrier velocity further upstream in the gradual

channel region, as long as the educed velocity in the do-

main is not completely compensated by carrier accumu-.

Fig.

10.

6 0

1

50

b 40-

E

LE 30-

.

20

-

10-

r

S A 6 2 -6

I np:

SC

126.7-14

SC 126,7-18

I I

- 1 0 2

4

6

8 10

12

VDSIv

-

drain bias

VDS

t

VGS 0.

Intrinsic

a

circuit transconductance

IYL

at

11

GHz

v= 0 G SA 62-6

+ LP:

C 126. 7-14

versus

0.1

1

-1

0

2 4 6 10

v,v

-

Fig.

11. a

circuit gate-source capacitance C“ at

11GHz

versus drain bias

vDS at vGS

=

8.

lation and/or channel widening [MI. Fig. 10 is thus con-

sistent with the conclusion, reached in Section III-B, that

stationary Gunn domain formation is stronger in GaAs

than in InP. Fig. 11,on the other hand, suggestsan increase

of 1, beyond saturation that s similar for both GaAs and

In P devices.

It

has been argued [18] tha t thencrease of 1,

is linked to carrier relaxationeffects [15] within the short

channel (velocity “overshoot”). Hence, it appears that

these effects are at east as strong in InP as n GaAs (cf.,

[17]) even though the stationary Gunn domainhat actu-

ally has formed is weaker in InP. On the other hand, a

field-dependent surface conduction (a surfacenversion

layer spreading from the gate toward drain) might cause

a similar effect.

The stronger decrease of the current-gain cutoff fre-

quency

f T

for GaAs MESFET’s than or InP equivalents,

as demonstrated in Fig. 9, is thus mainly caused by a

stronger reductionof the average carrier velocity

V

in the

gradual channel from which a stronger stationary Gunn

domain growth is inferred.

Fig. 9 reveals that the riginally reported improvement

of

f~

for In P over GaAs MESFET’s obtained at VDS

= 5

V (Table I)

[4]

s partly due to the strong reductionf f~

already present in GaAs at this bias. In fact, or the units

shown, the f~ peak at the ptimum drainbias of the GaAs


8/9

ENGELMANN AND LIECHTI: GaAs AND InPMESFETPARAMETERS

1295

’20

v=s=O

8

I

‘I

2

:

=

-2 -

O-

?

-4

-6

-10-

-8

-1412 2 0 2 4 8 10

VDS/V

-

Fig.

12.

Maximum available gain (MAG) at

11

GHz versus dra in bias

VDS

or VGS

= 0.

MESFET happens to be even higher than those of the two

In P MESFET’s.

Typ ica l

peak values, however, are still

some,what higher in In P devices ranging from 18-24 GHz

as compared to 16-20 GHz in GaAs devices.

Folr completeness, Fig. 1 2 compares the maximum

available power gain of the two kinds of devices at 10 GHz.

As already pointed out in 4], he gain is lower in the InP

devices due to the ombined influence of

c d g , R d s

and of

R, + R, + R i (see Section I).

VI. DISCUSSIONND CONCLUSION

The investigation of the bias dependence of MESFET

parameters in GaAs and in InP led to the ollowing con-

clusions.

1)

The roughly five times higher feedback capacitance

c d g in current-saturated InP MESFET’s asompared to

GaAs, ones (VDS 5

V,

VGS=

0,

see Table I) is mostly a

consequence of a weaker Gunn domain formation in nP

between the gate and drain at the drain bias levels em-

ployed. This leads to a lesser degree of decoupling of the

depletion-layer fringing capacitance from the dra in, par-

ticularly far beyond the onsetof current saturation.

2)

The small drain-source resistance R d s , frequently

observed in InP MESFET’s, was found to be correlated to

high drain current.We suggest th at a softerpinch-off be-

havior in InP is responsible for this trend. The ower re-

sistivity of the InP ubstrates

lo3-lo4

Q cm, Cr-doped,

as colmpared to

lo6-los

f2

cm for GaAs) th at were avail-

able for this study and the apparent bsence of trapped

charges at the nterface are most ikely the cause for this

behavior.

3)

The different behavior between GaAs and nP

MESFET’s in the bias dependence of the current-gain

cutoff frequency

f T

was linked to the ame Gunn domain

formation phenomenon that causes the c d g variation.

When a Gunn domain forms at the drain side of the

chanmel, the average carrier velocity in thechannel is re-

duced.

A t

the drain bias levels generally employed (VDS

= 4-5 V), this effect is stronger in GaAs than in InP and

hence, InP MESFET’s appear to ave a relatively higher

f~ value over GaAsones than theoretically estimated [4].

However, at theoptimum bias level for

f ~

he improve-

ment for InP against GaAs in only 10-20 percent ascom-

pared to the0 percent theoretical estimate.

4) The lower absolute value of the experimental cur-

rent-gain cutoff frequency f~ as compared to the theo-

retical estimates n both materials 4] tollows artly from

the fact that the verage carrier velocity in the channel is

significantly below the velocity peak value.A t higher drain

bias levels, in addition, an effective llengthening of the

channel beyond the metallurgical gate lengthL contributes

to the eduction in f ~ .or a GaAs MESFET, e.g., at a drain

bias of

VDS

= 4 V (VGS= 0), the average carrier velocity

U

in the gradual channel was estimated at about 30 percent

below the peak value up and theeffectlive channel length

1 was estimated at about 35 percent above L [MI. In ad-

dition, current saturation measurements n ungated de-

vices led to up values that are lower than expected from

theoretical estimates [4]. We determilned for GaAs at a

doping level near 1017 a value of u p

w

1.4

X

lo7

cm/s

[18] (theoretical value 1.7

X

107 cm/s [ E]) and for InP at

a doping evel near 0.5 X 1017 cm-3 a value ranging from

1.3-1.7 X l o7cm/s [9b]. These lower up values, in combi-

nation with the furtherreduction for the average channel

velocity, explain why the theoretical estimates for f~ in

Table I are too favorable.

The difference in Gunn domain formation between

GaAs and InP s related to the fundamentalifference in

the carrier velocity/field characteristic o f the two materials.

Similar Gunn domain effects as in GaAs would be expecte

in InP at drain bias levels that are at; least three times

higher than in GaAs, i.e., considerably beyond the cur-

rently achievable Schottkygate unction breakdown

values.

We have also speculated that the different epilayer-

substrate interface properties in aAs and InP which are

believed to cause the differences in Rtls might affect the

process of Gunn domain formation. This contention would

predict stronger Gunn domain formation in InP FET’s,

with its obvious effects on c d g and fT by providing an

enhanced interface barrier. This, of course, should also lead

to largerR d s values. Higher resistivity Fe-doped substrates

[26] or heteroepitaxial interfaces [27] should be investi-

gated in this context. On the other hanld, better control of

the GaAs epilayer-substrate interface may lead to less

electron trapping and, ence, a reductionof the interface

barrier causing a weakening of the Gunn in domain in

current saturation. This would be particularly desirable

for power FET’s which require ahigh drain bias capabili-

ty.

The overall conclusion of this investigation is then that

an

f~

advantage over GaAs is present in InP but s mar-

ginal. However, the generally higher fields of the InPve-

locity/field curve and its reduced tendencyor Gunn do-

main formation canbe of potential significance for a high

power device. n order to sustain the high drain bias levels


9/9

1296

IEEE

TRANSA(YI0NS O N ELECTRON DEVICES, VOL.

ED-24,

NO. 11, NOVEMBER

1977

necessary for such an application, a eakage-free gate is

of

paramount importance,

of

course. Replacement of the

Schottky gate by a p-n junction or an insulated gate all-

pears to be necessary.

An

additional advantage for

tile

high-power application would be he higher thermal con-

ductivity of InP.

ACKNOWLEDGMENT

The authorsexpress their gratitude to

.

Lizenby for his

measurement and computation assistance, to

.

Barrera

for supplying some

of

the InP MESFET’s and for

h..s

stimulating interest in the study, and to.

J.

Archer and

R. Van Tuyl for many critical comments to the manu-

script.

REFERENCES

[l] C. A. Liechti, “Microwave field-effect transistors-1976,’’

ZEE

E

Trans. Microwave Theory Tech., ol. MTT-24, pp.279-300, Jure

1976.

[2]

P.

Wolf, “Microwave properties of Schottky-barrier field-effect

transistors,” ZBM J Res. Develop.,vol.

14,

p. 125-121,1970.

[3]

ZEEE

Trans. Microwave Theory Tech. (Special Issueon Microware

Field Effect Transistors), vol. MTT-24, June1976.

[4]

J.

S. Barrera and R. J. Archer, “InP Schottky-gate field-effect

transistor,”

ZEEE

Trans.ElectronDevices, vol. ED-22, PI).

[5] S. Ohkawa, K. Suyama, and H. Ishikawa, “Low noiseGaAs fie]

Documents

1).Bias Dependence of GaAs and Ip MESFET Parameters