Multiterminal Pathfinding in Practical VLSI Systems with Deep Neural Networks

A multiterminal obstacle-avoiding pathfinding approach is proposed. The inspired by deep image learning. key idea based on training a conditional generative adversarial network (cGAN) to interpret task as graphical bitmap and consequently map onto solution represented another bitmap. To enable the proposed cGAN pathfinding, methodology for generating synthetic dataset also model implemented in Python/Keras, trained synthetically generated data, evaluated practical VLSI benchmarks, compared with state-of-the-art. Due effective parallelization GPU hardware, yields state-of-the-art-like wirelength better runtime throughput moderately complex tasks. However, remain constant an increasing complexity, promising orders of magnitude improvement over state-of-the-art pathfinder can be exploited numerous high applications, such as, navigation, tracking, routing systems. last particular interest this work.