A Neural Network-based Framework for Accelerated Device-Circuit Electrothermal Co-Simulations in GAAFETs

In this work, an efficient electro-thermal coupling model is developed and validated for 3nm Gate-all-around field effect transistors (GAAFET), which combines the artificial neural network (ANN) model for characterizing electrical properties of devices and a multi-stage RC thermal model with geometry dependence. The temperature rise induced by the self-heating effect (SHE) of the device is evaluated by the RC thermal model and feedback into the ANN model for self-consistent electrothermal interactions, and the number of hidden layers and neural nodes in the ANN model has a significant influence on the simulation accuracy and efficiency. A remarkable acceleration with an acceptable accuracy of electrothermal co-simulation compared with that of the BSIM-CMG circuit simulation framework can be observed, which provides a desirable scheme for thermal-aware reliability assessment at the device-circuit level.