Energy use, timing variation, hardware faults, and changing workloads all shape the practical deployment of machine learning. Research into deep learning accelerators offers design approaches relevant to industrial edge devices, embedded systems, and cloud-based AI services.