Heat removal is one of the major challenges faced in developing the new generation of high density integrated circuits. Future design technologies strongly depend on the availability of efficient means for thermal modeling and analysis. These thermal models must be also accurate and provide the most efficient level of abstraction enabling fast execution. We propose an innovative thermal simulation method based on Neural Networks that is able to solve the scalability problem of transient heat flow simulation in large 2D/3D multi-processor ICs by exploiting the computational power of massively parallel graphics processing units (GPUs).
展开▼
