Intelligent Design and Tuning Method for Embedded Thermoelectric Cooler (TEC) in 3-D Integrated Microsystems
Peng Zhang, Da‐Wei Wang, Wen‐Sheng Zhao, Bin You, Jun Liu, Cheng Qian, Hong-Bo Xu
Abstract
In this article, a genetic algorithm (GA)-based optimal design of thermoelectric cooler (TEC) and a hybrid Bayesian optimization (BO)-artificial neural network (ANN)-based tuning method are presented to implement efficient thermal management in 3-D integrated microsystems. The design approach is implemented based on the MATLAB-COMSOL co-simulation platform, where parametric modeling, simulation setup, and intelligent optimization modules are programmed in MATLAB and the Multiphysics simulations are conducted in COMSOL. The numerical simulation model is validated through comparison with the measured data from a previous work first. After that, embedded TECs are designed for the thermal management of a 3-D integrated microsystem, including two power chips with dimensions of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3\times 3\times0.15$ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.5\times 2.5\times 0.1\,\,{\mathrm {mm}}^{3}$ </tex-math></inline-formula> , utilizing the proposed design approach. The maximum of 14.4 and 7.3 K drops in temperature are achieved in power chips with the heat flux of 166 and 128 W/cm2, respectively. Finally, the mechanism and implementation of the hybrid BO-ANN-based dynamic thermal management (DTM) method are presented in detail. The numerical experiment results indicate that the power dissipation of the cooler is significantly improved through utilizing the proposed management approach.