Simulation of AlGaAs/GaAs HEMT

This example is based on a 1.0 micron gate length HEMT. The device has an 200A AlGaAs layer on top of GaAs. There is a 5.7e11 pulse doping in the AlGaAs layer approximately 50A above the GaAs channel, as illustrated in Figure 1.

Figure 1. Device structure and doping of the AlGaAs/GaAs HEMT.

This type of simulation typically has a high number of grid points for two reasons, firstly to resolve the hetero interfaces and secondly to resolve the pulse doping. Consequently, this type of simulation can take some time, particularly if there are multiple heterojunctions to resolve. For this reason the use of a parallel version of atlas provides much faster solution times. Figure 2 shows the mesh used to simulate this device.

Figure 2. Finite element grid values in these simulations.

This type of simulation utilizes a higher order approximation of the Boltzmann Transport Equation than drift diffusion, known as the energy balance model. Energy balance includes carrier energy terms in the evaluation of semiconductor properties, this leads to a more accurate result in cases when, for example, high fields or high currents are present. For HEMTs and other heterostructure devices it is often necessary to use energy balance to more accurately simulate the structure, giving more realistic gate leakage currents and transconductance characteristics.

Figure 3 shows the band structure of this device and the pulse doping through a section of the gate. Figure 4 illustrates the predicted Id/Vg curve for this device - note the down turn of the drain current as the Schottky barrier becomes forward biased and the gate begins to leak heavily.

Figure 3. Conduction and Valance bands at Vg=OV alongside the doping profile.

Figure 4. Gate and drain subthreshold characteristics as Vd2 2-OV.

Figure 5 shows the current flowlines that are displayed by the device when the drain is at 2V and the gate is at 1V i.e. just as the gate Schottky contact is becoming forward biased. The flowlines clearly show the movement of current from the source contact to the gate contact leading to the increased gate terminal current that is exhibited.

Figure 5. Current flowlines as the gate begins to leak heavily.

BLAZE can be used to perform design studies on HEMT and general HFET structures. For example, ternary or quaternary compound semi-conductor mole fractions can be varied, layer thicknesses can be varied as well as device geometry issues such as gate recess angles, gate lengths and contact spacing. Also, experiments can be performed on areas such as Schottky gate barrier height which are difficult to vary significantly experimentally.

Figure 6 shows a more complex psuedomorphic AlGaAs / InGaAs / GaAs HEMT structure, with a recessed gate and multiple pulse doping layers.This device was created using DEVEDIT (Silvaco's device and mesh generation tool). This is typical of the type of structure it is possible to generate with DEVEDIT - note that it is possible to produce non-planar type structures.

Figure 6. Recessed gate Pseudomorphic HEMT created by DEVEDIT.

Often compound materials and hetero-interfaces are characterized by significant defect levels. ATLAS will allow the use to incorporate traps into semiconductor materials, specifying their characteristics in detail, so as to model their effects on the overall device behavior both in DC and AC modes.

Figure 7 demonstrates the increase in speed possible for this type of simulation by using the parallel version of ATLAS. With 4 CPUs, for example, the time is reduced from over one and a half hours to thirty minutes. This represents more than 3 times speed up, giving an efficiency of over 75%.

Figure 7. Execution time improvement with
number of processors for HEMT characterization.