III-V Material Parameter Tuning using VWF Automation Tools


A recurring issue for those wishing to simulate III-V or other compound semiconductor devices is a way in which to efficiently calibrate and tune the material parameters used within the simulation environment. This is of particular concern for those involved in compound materials, as the volume and accuracy of material data is much less than for silicon.

The application of the Virtual Wafer Fab Automation Tools to experiments using process simulation has been well described in previous issues of the Simulation Standard, and has become an accepted methodology for process simulator calibration and experimentation.

This article deals with the application of the VWF framework to device simulation experiments to used for calibration, tuning and general device physics experiments, without necessarily using process simulation.


One of the lessor known applications of the VWF Automation tools is in creating experiments for device simulation alone using either S-Pisces or Blaze. The use of the Automation tools for device simulation experiments is a little less obvious than a process simulation based experiment. This is because the analog of VWF runs to split lots means the majority of VWF experiments have variations defined in the process simulator.

In the case of a device simulation, however, the experimental variation must be defined before the device simulator is executed as Atlas is a non re-entereant simulator. This has been made possible with the VWF Automation tools by utilizing an 'internal' simulator before the device simulation is executed. This internal simulator can be used to establish variables to be used later in the experiment.

MESFET Experiment

A basic experiment was devised using a 1.5 micron channel length Gallium Arsenic (GaAs) MESFET, as illustrated in Figure 1. The initial experiment consisted of studying the effects of gate contact Schottky barrier height, electron mobility and electron velocity saturation. The purpose of this experiment was two fold; firstly to create calibration curves for this type of material and device, secondly, to establish relationships between key design parameters of the device and these material parameters.

The experiment was based on the standard Latin Hypercube DOE (design of experiments rule), with 25 separate runs. Table 1 shows the allowed variation of each of the material parameters. Note that, in this experiment, the values of material parameters were allowed to vary significantly to establish device performance limitations as well as trend information.


Workfunction Difference eV 1.0 0.43 - 1.13
Mobility cm-2/Vs 4000 3000 - 5000
Saturation Velocity cm/s 1x107 0.1x107 -5x107

Table 1. Input parameters for tuning the MESFET characteristics.

The methodology of a VWF experiment is to extract certain design parameters from electrical characterization of the device. In this experiment, an Id / Vd sweep was performed at Vg=0V, and an Id / Vg sweep was performed at Vd=2.0V. The parameters extracted from these curves are listed in Table 2.


Id leak A Off state drain leakage current
Id sat A Drain saturation current
Vt V Threshold Voltage
Ig leak A Forward gate leakage current
Id max A Maximum drain current

Table 2. Results from BLAZE used as targets in the VWF.

Analysis of the Results

Two types of results are produced as a result of the experiment: firstly, the curves from each run can be viewed and compared, secondly the extracted design parameters can be used to generate RSMs (response surface models) linking the material and device input variables to the output design parameters with an equation. The first, gives us a qualitative view of how changes in the material parameters effect, while the second gives a quantitative relationship that can be used to tune the simulation.

Figure 2 shows 10 different Id / Vg and 10 different Id / Vd curves overlaid onto the same plot. It is clear from this that as the number of different curves increases, it becomes increasingly difficult to interpret the results, and use them for tuning without some type of analysis tool. In contrast, Figure 3 shows a RSM generated for forward gate leakage as a function of gate barrier height, and also the drain saturation current as a function of electron mobility. These two plots demonstrate the relative ease of calibration using the VWF production tools to analyze the results of a large number of simulations. Note that the RSM plots actually represent a 3 - dimensional result space, in this case, where the user can interactively change any of the input variables and see their effect of the response surface.

Figure 2. Id/Vd and Id/Vg for 10 different results from one experiment.

Figure 3. RSM of Idsat as a function of electron mobility (left) and Ig leak as a function of gate workfunction (right).

A Calibration Example

One of the key parameters in a MESFET performance characteristic can be the forward gate leakage current, as this can provide a large source of power loss. However, this value is dependent on the barrier height of the gate Schottky barrier. Surface states lead to Fermi level pinning at this interface, resulting in an Schottky barrier height which can only be determined experimentally. Additionally other effects such as image force lowering, may cause further experimental variations in the barrier height with field. Using the response surface model shown in Figure 3, it is possible to tune the forward gate leakage current from the simulations to the value found experimentally.


By running a designed experiment in the VWF using material parameters as experimental variables it is possible to tune III-V device characteristics efficiently. Further experiments and analysis on the device physics and model behavior are possible, without the user having to create a huge set of different files and results, and then exporting the information to an external data analysis system.