Deep learning for both broadband prediction of the radiated emission from heatsinks and heatsink optimization

Abstract

Heatsinks have quasi-antenna behavior in many cases and cause interference both at the system level and at the PCB design level. Therefore, determination or prediction of both resonance frequencies and maximum radiated emission are crucial at any design step. In this paper, as a novelty, in 2–8 GHz band, a model based on deep learning is developed to predict resonance frequencies in parallel plate-fin type heatsinks. Parameters taken into account are the number of fins, the width, length, and height of the heatsinks. 3888 heatsinks with different sizes are modeled to prepare data set and the Grey Wolf Optimizer algorithm (GWO) is utilized to optimize the heatsink parameters. Consequently, while this model obtains outputs for certain inputs, the optimization algorithm procures certain inputs for these outputs. Furthermore, the predicted and optimized results are compared with the simulation and measurement results. The proposed model successfully works according to the measurement and the proposed model results since R2 values are 0.96, 0.98, 0.97, and 0.99 for f1,f2,REmax1, and REmax2, respectively. The results are good agreement and R-squared values of resonances (f1,f2) and the maximum radiated emissions (REmax1,REmax2) are quite acceptable considering the sophisticate of the proposed model.