Approximating the Steady-State Temperature of 3D Electronic Systems with Convolutional Neural Networks

Monika Stipsitz,Hèlios Sanchis-Alepuz

doi:10.3390/mca27010007

Abstract

Thermal simulations are an important part of the design process in many engineering disciplines. In simulation-based design approaches, a considerable amount of time is spent by repeated simulations. An alternative, fast simulation tool would be a welcome addition to any automatized and simulation-based optimisation workflow. In this work, we present a proof-of-concept study of the application of convolutional neural networks to accelerate thermal simulations. We focus on the thermal aspect of electronic systems. The goal of such a tool is to provide accurate approximations of a full solution, in order to quickly select promising designs for more detailed investigations. Based on a training set of randomly generated circuits with corresponding finite element solutions, the full 3D steady-state temperature field is estimated using a fully convolutional neural network. A custom network architecture is proposed which captures the long-range correlations present in heat conduction problems. We test the network on a separate dataset and find that the mean relative error is around 2% and the typical evaluation time is 35 ms per sample (2 ms for evaluation, 33 ms for data transfer). The benefit of this neural-network-based approach is that, once training is completed, the network can be applied to any system within the design space spanned by the randomized training dataset (which includes different components, material properties, different positioning of components on a PCB, etc.).

Highlights

Physics simulations are becoming an essential aspect in the design of electronic systems
The size of the dataset was increased by four without further computational effort via three 90 degree in-plane rotations of each system, so that the total dataset consisted of 1856 systems. 75% of the generated dataset was used as training set and the remaining 25% as test set
The mean relative L1 error per system is at the percent level, with values below 2% in most systems in the test set: This can be seen in Figure 4, where a histogram of all systems of the test set binned according to their corresponding L1 error is shown

Summary

Introduction

Physics simulations are becoming an essential aspect in the design of electronic systems. Electronic and thermal co-simulations for the design of an efficient power converter have been studied in [1]. Such coupled simulations lead, to large computational requirements. The long simulation time render automatic optimizations of designs impossible. The application of ML techniques to the design of electronic systems focuses mainly on two aspects [2]. A genetic algorithms required around 1000–10,000 FEM simulations to find the optimal placement of a chip in a system [5], which takes a considerable amount of computation time. ML is used to accelerate the evaluation of individual designs, i.e., replace the time consuming simulations by neural networks [6,7,8,9]

Methods

Results

Conclusion