DLAG-TA: Deep Learning-Based Adaptive Grid Builder for System-Level Thermal Analysis

This work develops an adaptive grid builder for system-level thermal simulators by embedding a well trained deep neural network (DNN) model, and we call it DLAG-TA. To automatically generate adaptive grids for various sizes/structures of handheld devices, it also contains a parameter scaling procedure and a virtual block to element grid mapping procedure to assist the trained DNN model. The results show that DLAG-TA can effectively build grids with high quality.