Recovery of Intrinsic Heterojunction Bipolar Transistors Profiles by Neural Networks

Due to the shrinking size of devices, the calibration process for TCAD simulators is becoming increasingly complex with time mandatory to run simulations steadily increasing. In this work, we propose a Statistical Learning approach to predict Figures of Merit (FoMs) of a Silicon Germanium Heterojunction bipolar transistor (HBT) from its 1D doping and germanium profile. The surrogate model thus built allows to simulate electrical outputs with high precision much faster than traditional simulations. This model is then used to solve the inverse design problem and recover profiles that match target FoMs.