Pareto based Multi-Objective Evolutionary Optimization of Multi-node Network for Thermal Modelling of Electronic Package

The study deals with the adoption of Differential Evolution (DE) algorithms to optimize a Compact Thermal Network Model (CTNM), an efficient multi-node model aimed to drastically reduce the numerical model size and computation time with respect to a detailed thermal model (DTM). The established model is a black-box object combined with a network of resistors that links a single temperature-sensitive node to major surfaces of heat extraction. The creation of the simplest multi-node thermal model involves several conflicting objectives between maximum internal node temperature and the nodal mean temperature of each black-box external surface and the heat flow rate leaving it. At first stage of the multi-objective optimization, the study of the optimal trade-offs between all objectives is done using Pareto dominance. The multi-objective optimization problem is then converted into a user-friendly single objective optimization problem using a weighting metrics method, emphasizing the best Pareto-optimal solution. A comparison of different variants of DE algorithm is presented through the application of miniaturized electronic component referred to as Quad Flat No-lead packages (QFN). The performances of the Classic-DE/Rand/1 algorithm, performed on several test cases of QFN's electronic component family, has shown to overperform the other variants. Whatever the package size under study, a deduced six-node matrix proves its ability for training data to yield high-accuracy resistance-network models and to perform well for training-independent validation scenarios of different unseen boundary conditions. The prediction of the component most sensitive temperature using a very simple black-box model, stays within 1% of expected DTM of the CFD in average.