ChewBaccaNN: A Flexible 223 TOPS/W BNN Accelerator.

Binary Neural Networks enable smart IoT devices, as they significantly reduce the required memory footprint and computational complexity while retaining a high network performance and flexibility. This paper presents ChewBaccaNN, a 0.7 mm$^2$ sized binary CNN accelerator designed in globalfoundries 22 nm technology. By exploiting efficient data re-use, data buffering, latch-based memories, and voltage scaling, a throughput of 233 GOPS is achieved while consuming just 1.2 mW at 0.4V/154MHz for the inference of binary CNNs with 7x7 kernels, leading to a core energy efficiency of 223 TOPS/W. This is up to 4.4x better than other specialized binary accelerators while supporting full flexibility in kernel configurations. With as little as 3.9 mJ, using an 8-fold ResNet-18, a Top-1 accuracy on ImageNet of 67.5% can be achieved, which is just 1.8% less than using the full-precision ResNet-18.