Automatic TCAD Model Parameter Calibration using Autoencoder

Modified autoencoders (AEs) have been used to capture the latent space physics of a given electrical characteristic curve (e.g. IV or CV). Therefore, it is expected that they can also be used to calibrate TCAD model parameters of novel materials such as Ga2O3 which is an emerging ultra-wide-bandgap (UWBG) material. In this paper, we demonstrate the use of an AE to perform automatic TCAD parameter calibration (Philips Unified Mobility model (PhuMob)) in Ga
2
O
3
with 6 parameters. We also discuss a noise technique to improve calibration accuracy and an efficient training data generation method using Latin Hypercube Sampling (LHS). The machine is validated with unseen noisy curves to mimic experimental data. The PhuMob parameters extracted from the unseen curves are used in TCAD simulation and can reproduce the original curves with high accuracy (thus the calibration is successful).