Implementing embedded neural network processing at the edge requires efficient hardware acceleration that combines high computational throughput with low power consumption. Driven by the rapid evolution of network architectures ...