6.1.2.3 Reduction of the Recess Length

As demonstrated in Section 6.1.1, g_{m  ext} is increased when L_R is reduced. The reason is a decrease of the ohmic resistance in the current path. Therefore, g_m is expected to depend roughly linearly on L_R. It is found that the dependence of C_G on L_R can be modeled by
 

(63)

where a and b are constants. The simulated dependence of f_T on L_R is plotted in  Figure 6.17. In the hypothetical situation e_r = 0, there is almost no dependence of C_G on L_R, and f_T increases monotonously with decreasing L_R due to the improvement of g_{m ext}. For the case that the gate and the adjacent device surfaces are encapsulated by a dielectric with e_r = 7, the decrease of f_T on the left hand side of the maximum at L_R  90 nm is caused by the rapidly increasing C_GD. If a device passivation with a dielectric constant e_r < 7 were available, it is evident that f_T would be higher for any e_r < 7 and the optimum of L_R would be shifted towards smaller values.
 

The rapid increase in C_GD has even a more significant impact on f_max. To determine f_max R_S is calculated by the assumption that the intrinsic g_mi remains constant and the reduction of the extrinsic g_m is purely attributed to an increase of R_S with L_R. The gate resistance is assumed to be independent of L_R and is taken to be zero in the simulations. To calculate f_max the measured value of R_G of HEMT A is taken. In  Figure 6.18 f_T and f_max are shown as a function of L_R for fully passivated (e_r = 7) and unpassivated devices (e_r = 1). For e_r = 7 and small L_R C_GD is very large due to the small distance between the gate and drain contact metals which leads to rather low f_T and f_max. Both f_T and f_max increases with increasing L_R and reach a local maximum. But the maximum of f_max at L_R = 150 nm is obtained for significant larger L_R than the maximum for f_T. The reason is that f_max depends much stronger on C_GD than f_T. Therefore, a decrease of C_GD is overcompensated by the linear increase in R_S for larger values of L_R.
 

The investigations on low noise HEMTs show very good agreement between the calculated and measured characteristics of the devices. It is possible to trace differences in manufactured devices, for instance, the recess depth with an accuracy in the order of 1 nm. Therefore, very accurate information for improvements of device characteristics or the technology of these devices can be given.

All devices were based on a homogeneously doped single heterojunction structure, i. e. a GaAs/InGaAs/AlGaAs layer sequence. As already discussed the disadvantage of this sequence is that the energy barrier on the backside of the channel is rather low which limits the power capability. In the following sections HEMTs based on delta doped double heterojunction structures similar to the structure of HEMT_ref will be investigated.

• 6.2 Power HEMT for 0.9/1.9 GHz and 40 GHz Applications
• 6.2.1 DC Characteristics
• 6.2.2 Dependence of Recess Geometry on the E-Field Distribution
• 6.2.3 RF Characteristics
• 6.2.3.1 Reduction of the Gate Length
• 6.2.3.2 Contributions to the Gate Capacitance
• 6.2.3.3 Dependence of C_G on Passivation Thickness
• 6.2.3.4 Dependence of C_G on T-gate Cross Section
• 6.3 Millimeter Wave HEMTs
• 6.3.1 DC Characteristics
• 6.3.1.1 Dependence on the Gate-to-Channel Separation
• 6.3.1.2 Dependence on the Gate Length
• 6.3.2 RF Performance
• 6.3.2.1 Dependence on the Gate Length
• 6.3.2.2 Contributions to the Gate Capacitance
• 6.3.2.3 Dependence of f_T on d_GC
• 6.3.2.4 Dependence of f_T and f_max on L_R
• 6.3.2.5 Dependence on the Passivation Thickness
• 6.3.3 Optimization of Millimeter Wave HEMTs
• 6.4 Comparison between Low Noise, Power, and Millimeter Wave HEMTs
• 6.5 Influence of Backside Doping on the C_G(V_GS) Characteristics