Topology optimization of heat conduction based on U-ResNet neural network

In order to break through the thermal dissipation bottleneck limitation of microelectronic integration technology. Topological optimization based on high thermal conductivity materials and structures to achieve efficient heat dissipation has become one of the key strategies to solve this problem. The conventional topology optimization methods rely heavily on extensive finite element iterations, leading to high computational costs and significant time consumption, which restrict their application in rapid design scenarios. To substantially improve computational efficiency while maintaining optimization accuracy, this study innovatively integrates the U-Net and ResNet architectures to construct a U-ResNet model, which replaces the core iterative process of the traditional SIMP method, enabling rapid optimization. Experiments conducted on two typical types of two-dimensional heat transfer topology optimization problems demonstrate that the proposed model reduces the computation time by an average of over 91 % while achieving accuracy comparable to the SIMP method. The results indicate that U-ResNet offers a new, efficient, and reliable approach to heat transfer topology optimization, providing a viable pathway toward the intelligent development of thermal design in microelectronics.