Impact Ionization in GaAs MESFET’s
KELVIN HUI, CHENMING HU, sestor wanes, 1, PETER GEORGE, avo PING K, KO
Absact—A method» mesore impact onan creat
[MESFETs Is prseted. The impact lonzaton current Is then sed
{ovctelate the maou lected i the channel and the impst
1 Demopven0s
[riz ansanig ofiptatonin Gans MES
FET is important for both device degradation and break-
down studies. Since the number of electron-hole pairs gen-
crated by impact ionization is proportional to the product of
the ionization coefficient and the carrier concentration, both
mobile carriers and a strong electric field are required for
impact ionization to occur. In MESFET's, impact ionization
wwll occur atthe field singularity at the drain edge of the gate
‘where the electric field is highest (1, 2] or inside the high-
Field domain along the conduction channel where the electric
field is high and current density highest. The site where im
pact ionization occurs first is bias and technology dependent.
"The effect of surface stats in GaAs in suppressing the electric
field at the field singularity poit has been studied [3]. In gen-
eral, the singularity atthe drain edge of the gate determines
the drain breakdown voltage when the MESFET channel is
pinched off, whereas impact ionization along the channel eur~
Fent path is responsible for the weak impact ionization current
with the channel open.
'A knowledge of the electric field distribution along the con-
“duction path is essential for the study of impact ionization.
However, de tothe negative differential mobility of electrons
in GaAs at high electric fields, an accurate and analytic elec~
tte field mode! is still unavailable today. As the impact ioniza
tion coefficient is exponentially proportional tothe negative
inverse ofthe electric field, a small etror inthe estimation of
the electric field will result in large error in the calelation
ofthe ionization rate. The limited experimental data on eb
{ron impact ionization rate, especially in the (110) direction
in which most MESFET’s ate fabricated, has rendered both
modeling and simulation of impact ionization difiutt
"Accurate models for the channel electric fekd and impact.
‘onization curent (substrate current) have been developed for
Mamscipereeined October 4, 1989; eed December 7, 1989. Thi
swt pone yt oa Serves eros Pepa ve Chaat
ape. 84 007 and an IBM ean fellows.
“he mtr re wh he Eletonts Rese Laboratory nd he De
pane! of lac Engineering and Comper Scenes, Unversity f
Eafe, Bere, CA 672.
THEE Log Suber 03420.
‘Si MOSFET's in recent years [4], [5]. The measurement of
‘mpact-onization-generated substrate current has become the
‘most convenient and effective means of characterizing hot
carrier phenomena in $i MOSFET"s and has yielded the low-
field impact ionization coefficient. In this paper we will ex-
amine the feasibility of extending these models to GaAs MES:
FETs.
1, Tuzony
“The holes generated by impact ionization along the channel
‘current path wil follow the eletrc field lines and be collected
at either the gate or source electrode. Two-dimensional (2-D)
104
104
ee a eeeeeeceaea ERB Ee ates
Vug
Fg. 2._ Gat caret ves dang vlogs Vay fr ere a
oscars bis tars fom 101002 ¥ in ep of 02 V. The
Ser Sl ip crea wh ere fig)
be approximately qual to Vzs. Ths is much higher than the
‘value in normal FET operation (Vg = 0). Using this method,
the sum of the gate-source and gate-drain diode leakage cur-
rents measured individually ean be much larger than the ac-
tual gate leakage under normal bias. A more accurate value
of gate-drain leakage current is obtained by biasing the MES
FET in the pinch-off region and subtracting the gate-source
leakage from the measured gate current. In pinch-off the drain
currea is very small and soi the contribution to gate current
by impact ionization, The electric field distribution, on the
‘other hand, will resemble that under normal FET bias. In our
measurements we set Vy t0 1.4 times the threshold voltage.
Fig. 2 shows the gate current of « depletionsmode 250-
um x 0.S-am ion-implanted recessed-gute GaAs MESFET
‘measured at differen gate and drain biases. The channel was,
formed by a 50-keV Si implant into a Cr-doped HB substrate,
with dose ¢ = 10! em~, The nominal Ly, and Lya valves
‘are 0,25 and 0.5 wm, respectively. The impact ionization cu-
rent Fy is obtained by subtracting Isseay fom Ty according
to (4). Tae increases rapidly with Var and is often many or-
ders of magnitude higher than say. Is also a function of
Vax since itis dependent on Fay (Se€ (2).
Fig. 3 shows the normal Jay versus Vy characteristics with
superimposed constant yells contours. According 2 (3),
Teller is unique fonction of Vas ~ Vue since the char-
‘scteristc length 1 i @ constant, Thus these contours also rep-
resent constant Vy ~ Vx contours. Indeed, the contours in
Fig. 3 re found to be parallel to cach other. When 2 con-
tour is parallel shifted toward the origi, it gives a reasonable
saturation voltage Vs contour. Experimental data shown in
Fig, 4 demonstrate this behavior for differen values of Vs
Using (3) and (5), Fig. 4 can be repoted to yield graph
of ionization rate versus 1/Emax a8 shown in Fig. 5, A valve
‘of 0.34 um was chosen for the effective length to translate
Vas ~ Vi ito Eye according 1 (3) and rake the a data
match the impact ionization coeicient measured by Pearsall
et al (7 a high electric fields. This value of is very close
to the value of 0.36 um obtained from Fig. 1.
Fig. Sextends the cy data range beyond what is available in
the literature by five orders of magnitude For electric fields