22.214.171.124 Dependence of CG on T-Gate Cross Section
To investigate the influence of the gate cross section, different gate
geometries were simulated by varying the two characteristic dimensions
dG and dT. All calculations
were performed for LG = 220 nm and LT
= 820 nm except for the case of a rectangular gate as depicted in
6.36. One limiting case for the gate geometry is dG
= 0 (as shown in Figure
6.33 and used for all simulations in Section 6.2
discussed up to here) whereas the other extreme is given by dG
= dT ("ideal" Tgate cross section).
6.37 presents simulations of CG and fT
as a function of dG for a completely passivated device
with dT as a parameter. The closed circle in Figure
6.37 designates measurement and simulation of the particular HEMT shown
6.33. Here, a distinct Tgate stem does not exist (dG
= 0). According to Figure
6.37, the fT of this device could be increased from
53 GHz to a value close to 60 GHz if the gate cross section could be improved
in a way that dG changes from zero to 200 nm (dG
= dT, ideal Tgate). Further increase of dG
and dT would further improve fT. Values
of about 65 GHz are expected for dG = dT
= 400 nm, without decreasing the gate length. In the case that dG
= 0 and only dT = 400 nm (gate stem sidewalls and device
surface enclose an angle of 53°), again fT
60 GHz is expected, comparable to the case dG = dT
= 200 nm.
Thus, it is of immense importance that the process technology is able to realize a gate cross section where the gate stem sidewalls are really perpendicular to the semiconductor surface (ideal Tgate). If the enclosed angle is significantly smaller than 90° (Vshaped gate stem), dT has to be substantially greater compared to the perpendicular case if the same fT has to be realized. The multi-resist level approach of most EBL processes facilitates the fabrication of such vertical Tgate stems as shown in Figure 6.3. Gate technologies that use optical lithography are usually based on narrowing the comparatively large resist openings by spacers. They often have the property not to result in really vertical sidewalls as illustrated in Figure 6.2. If such a technology is used for the fabrication of devices for high-frequency application, it is important that at least the part of the gate immediately adjacent to the semiconductor surface has perpendicular sidewalls.
A rectangular gate shape shown in Figure 6.36 represents the limit for arbitrarily increased dT and dG. The CG of a fully passivated device with a rectangular gate is indicated in Figure 6.37 by the upper bold line and the CG for a device without passivation by the lower bold line.
The upper limit in fT is marked by the case of a rectangular gate and no passivation which decreases the total CG to about 820 fF/mm and increases fT to 88 GHz. Both rectangular gate and no passivation are rather unpractical limits although they can be realized as reported in . The disadvantage of a rectangular gate is a significant increase in the gate resistance RG. This becomes especially important for devices with long gate fingers such as power devices and for extremely high frequencies of up to 100 GHz were the skin effect becomes significant.
Based on the results presented so far the following considerations on the optimization of the RF performance of the HEMT investigated here can be made. A reduction of the gate length from 220 nm to 120 nm would reduce CG by 320 fF/mm. The drawback would be a larger g0. As shown in Section 6.1.2 for the single heterojunction HEMT g0 increases by 100 % to 30 mS/mm which is still an acceptable value. In addition a single heterojunction HEMT is a worst case assumption because it is expected that short channel effects are more severe due to the lower barrier height under the channel compared to a double heterojunction HEMT.
A realistic optimization of the Tgate shape with dG = 150 nm and dT = 300 nm would yield a reduction of CG of 210 fF/mm. This reduces A2 of (62) from 85 fF/mm to about 55 fF/mm such that a reduction of the passivation thickness would have a reduced impact. For a realistic reduction from 700 nm to a residual passivation thickness of 200 nm a reduction in CG by about 70 fF/mm can be expected. These improvements together sum up to a reduction CG from 1.38 pF/mm to about 0.8 pF/mm which translates into fT = 92 GHz.
This shows that to practically increase fT to significantly
over 100 GHz a higher gm is necessary. This can be achieved
in first place by a smaller gate-to-channel separation. Unfortunately this
goes along with an increase in CG due to an increase
in A3 of (62). But even
if the quotient of gm/A3 would remain
constant this would still increase fT due to the parasitic
capacitance which are unchanged by a reduction of dGC.
A detailed investigation on the optimization of the RF performance of double
heterojunction HEMTs will be given in the following section.
Helmut Brech 1998-03-11