Benchmarking of Machine Learning Methods for Multiscale Thermal Simulation of Integrated Circuits

Abstract

Multiscale thermal analysis in integrated systems is required for capturing both device-level and circuit-level dynamics. Traditional analysis with finite element (FE) models can be accelerated by using machine learning (ML) methods. In this paper a performance benchmarking between three ML methods for thermal simulation is carried out: Artificial Neural Networks (ANNs), Proper Orthogonal Decomposition with Radial Basis Functions (POD-RBF) and finally POD-RBF-ANN is used as a hybrid ML method. The (dis)advantages of the different methods are demonstrated for the thermal simulation of a multiscale photonic chip. The ML models are trained with FE data for both linear and non-linear dynamics and are tested for inter-and extrapolation prediction accuracy. A computational speed increase with factor >7500 compared to FE is obtained. Furthermore, ANNs prove to be the best suited for the simulation of non-linear dynamics. POD-RBF is the best method for minimizing training time and combining the best of both methods in POD-RBF-ANN creates a ML model with short training phase and highly accurate predictions.