Adapting AI into Low Power Testing

11This work is partially supported by the research project sponsored by the Synopsys Inc., USAAdvancements in data science have enabled various fields to achieve unparalleled performance and efficiency enhancement. Different domains of Artificial Intelligence (AI) held the center of interest for researchers, academic and industrial practitioners throughout the last decade. When it comes to the low power testing, it still has a wide range of possibilities where existing methods and techniques can be made more efficient as well as faster with the integration of AI. In this proposal, first, we have come up with a new test vector reordering technique that attempts to reduce both shift and capture power during testing. Secondly, we claim that the use of AI models can significantly speed up such techniques where repeated simulation-based value estimation remains an essential bottleneck. We verify our claim by engaging a deep neural network (DNN) based predictive model to replace the repetitive simulation-based method on this new reordering technique. Experimental results show that the AI-based framework can speed up the simulation-based framework by 162 times, on average.