Gated Transmission Line Model Structure for Characterization of Field-Effect Transistors

In the design and manufacture of FET integrated circuits it is desirable to monitor the characteristics of the active channel of devices comprising the circuit. The present invention, useful in the integrated circuit art, is a gated transmission line model (GTLM) structure which provides a unique characterization device for accurately determining a variety of such parameters.

In the field effect transistor art, Transmission Line Model (TLM) measurements are a standard technique for the determination of the contact resistance and sheet resistance of the active layer.

In this invention the test structure has Schottky gates between the ohmic contacts of a conventional transmission line model (TLM) pattern, and the new test pattern is hereafter referred to as a gated-TLM or GTLM pattern. The gate lengths are different, one from the next, and the gate-to-ohmic separations are kept constant. With this structure the series resistance is nearly independent of the gate voltage and the channel resistance is modulatable by the gate voltage to provide an accurate determination of FET channel parameters.

At present there is no way to measure sheet resistance of the active layer of an enhancement-mode FET due to complete surface depletion. Since the GTLM structure can modulate the extent of this depletion, it offers a way to characterize enhancement-mode channels.

In addition to channel resistance, the parasitic source resistance is an important parameter to know, both for digital and (especially so) for low-noise analog FET applications.

A reference H. Fukui "Determination of the basis device parameters of a GaAs MESFET", Bell Sys. Tech. Journal, pp. 771-797, March 1979, shows a technique used to deduce source resistance Rs. The drain-to-source resistance Rds of a FET is measured at small Vds and plotted versus [(Vbi-Vgs)/Vpo]-0.5 where Vbi is built in junction voltage, Vgs is gate voltage and Vpo is pinch-off voltage. Vpo is arbitrarily adjusted to get a good linear fit and the Y-intercept of a least-squares regression to the data is used as 2Rs. The value of Rs determined in this way, however, varies significantly as Vpo is adjusted and therefore the method does not give accurate results. Also, the data is taken on a single FET so there is no way to average out any device peculiarities. Therefore this technique only provides an estimate for Rs.

The GTLM structure offers a means to deduce the source resistance with no parameter-fitting involved. In addition, the GTLM works equally well on planar, recessed-gate, and modulation-doped structures where other methods will fail due to undesirable current paths.

A reference K. Lee, et al, "Low field mobility in GaAs ion-implanted FETs," IEEE ED-31, No. 3, pp. 390-393, March 1984, shows a technique used to obtain mobility profiles. Measurements of low-field FET channel resistance and transconductance versus gate voltage are used together with a numerical solution of ##EQU1## to profile the mobility. This method requires a fitting parameter (Rs) to be adjusted in an iteration scheme with an assumed doping profile. Therefore the results will be an approximation. Also, the area of the device under test is generally too small to permit CV analysis. The GTLM structure deduces Rs automatically and therefore can use an iteration scheme to additionally deduce the doping profile.

FIG. 1 discloses a top view of the gated Transmission Line Model structure.

FIG. 2 is a side view or cross section of FIG. 1.

FIG. 3 is a graphical presentation of certain data analysis of the structure.

FIG. 4 is a simple resistance model of the gated TLM structure.

Referring now to FIGS. 1 and 2 there is shown a gated-TLM structure 10 on a semiconductor substrate 11. As is most clearly shown in FIG. 2 which presents a side view of the gated-TLM pattern, there is disclosed a plurality of alloyed ohmic contacts 20, 21, 22, 23 and 24. The ohmic contacts are deposited on the wafer atop an implant or grown layer 12 identical to that of neighboring circuit FET's. These five ohmic contacts have been spaced apart by differing distances, one from the other. In this example, the ohmic metal contacts 20 and 21 are spaced apart by 5 .mu.m, contacts 21 and 22 by 10 .mu.m, contacts 22 and 23 by 15 .mu.m and contacts 23 and 24 by 20 .mu.m. Each of the ohmic contacts includes pads for probe contact identified in FIG. 1 as pads 40 to 49. Between the spaced contacts a Schottky metal is deposited to form gates and in FIG. 2 these gates are identified at 25, 26, 27 and 28. Each of the Schottky gates is of a different length dimension, that is, gate 25 is 1 .mu.m long, gate 26 is 6 .mu.m long, gate 27 is 11 .mu.m long and gate 28 is 16 .mu.m long. These gate lengths match with the related ohmic contact spacings so that the gate-to-ohmic spacings are all kept constant (that is, 2 .mu.m in this example), and are identical to those of circuit FET's. The four gates 25, 26, 27 and 28 are connected together by conductive path 29 and to contact pads 32 and 31 as best illustrated in FIG. 1.

With the gate lengths varied and the gate-to-ohmic separations kept constant, the series resistance is nearly independent of the gate voltage and the channel resistance may be modulated by the gate voltage. This allows an accurate determination of FET channel parameters including source resistance, electron mobility profile, and doping profile.

To measure the source resistance, a four-point resistance measurement is performed at each ohmic spacing (see FIG. 2). The voltage between the ohmics must be kept much smaller than the saturation voltage of the active channel. These resistances R.sub.TOTAL,i are plotted versus their respective gate lengths and a least-squares regression is performed to find the Y-intercept (see FIG. 3). This Y-intercept is equal to 2(Rc+Rp) or precisely twice the source resistance. In the resistance model of the gated-TLM structure shown in FIG. 4, R.sub.c =resistance of the contact, R.sub.p =parasitic ohmic-to-channel resistance and R.sub.ch (Vgs)=channel resistance modulated by gate voltage. Since the Y-intercept should be constant with respect to gate voltage, we can take the four-point resistance measurements at several gate voltages and compare the results as a check of our method.

The gated-TLM structure can be used to measure the effective free surface potential by varying the gate voltage until the slope of the line in FIG. 3 is identical to that of a conventional TLM pattern fabricated over a similar active channel. The gate voltage Vgs necessary to cause identical slopes is then equal to the difference between effective free surface potential .phi..sub.s divided by electronic charge q and the built-in junction voltage Vbi, i.e. ##EQU2##

The electron mobility profile and doping profile may be obtained by measuring the sheet resistance R.sub.SH (Vgs) of the active channel (slope of line in FIG. 3) as a function of gate voltage. This data is combined with conventional CV (capacitance-voltage) analysis of the largest Schottky pad (which determines X(Vg) and N(X)) to calculate the mobility profile using a numerical solution of ##EQU3## where Vgs=gate voltage

N=electron impurity concentration

R.sub.SH =sheet resistance of active channel in ohms/square

A=thickness of channel

.mu.=electron mobility

x=distance into channel from surface

The sheet resistance versus gate voltage data from gated-TLM may also be combined with conventional geometric magnetoresistance (GMR) measurements to find the mobility profile directly, although this requires use of an electromagnet.

Referring once again to FIGS. 1 and 2, it will be appreciated that in the drawing the ohmic metal contacts 20, 21, 22, 23 and 24 (not to scale) have been shown in a foreshortened way for illustration. In FIG. 2 each of the contacts 20, 21, 22, 23 and 24 is shown as broken and an exemplary length for each of the ohmic contacts of 140 microns is shown on the drawing at contact 24. Thus the actual length of each of the ohmic metal contacts is much longer than appears in FIGS. 1 and 2 and said ohmic contact length is also much longer than the spaces.