Thermal Management by Using Small-Area Chips and AI-Based Design Optimization in SiC Modules

The aim of this study was to reduce the cooling system volume of resin-based SiC modules, leading to an increase in the power density of power conversion systems. Although resin-based modules exhibit high power cycling capability, resin-insulated substrates have poor thermal conductivity, which may lead to deterioration of the thermal resistance of the module. Therefore, we arranged a large number of small-area chips in a distributed manner in order to reduce thermal resistance by increasing heat dissipation area. However, increasing the number of chips made the module structure more complex and the optimal design difficult. Therefore, the design parameters of the module were optimized using Bayesian optimization, an AI-based optimization method. The optimized module was fabricated and demonstrated significant improvements in thermal resistance, switching loss, and parasitic resistance over conventional ceramic-based modules. This module allowed the cooling system size to be reduced by 61%.