An Optimization Design Method of Multi-chip Parallel Power Module Based on Machine Learning

As an important component in power electronic energy conversion system, power semiconductor module layout and its multi-objective optimization are considered to be the key steps to achieve excellent performance of silicon carbide. Low parasitic parameters, low junction temperature, high power density and high reliability are the key design elements of multi-chip parallel silicon carbide power module. The existing traditional design methods largely rely on the experience of trial and error, the design cycle is long and the cost is high. This paper proposes a multi-objective optimization design method of power module based on artificial neural network ANN and deep reinforcement learning (DRL). Firstly, ANN and FEM are used to solve the problem that the self -inductance and mutual inductance of power module and the thermal coupling between multi-chips are difficult to be represented by mathematical formulas. At the same time, deep reinforcement learning (DRL) and the training results of artificial neural network (ANN) are used for multi-objective optimization of three optimization objectives: parasitic inductance, junction temperature and power density. Based on this method, a 1200 V/300 A half bridge power module with three parallel chips is optimally designed. This power module has good parasitic inductance, junction temperature and power density. This design method has certain guiding significance for the packaging design of silicon carbide multi chip parallel power modules.