Superconductor Computing for Neural Networks

The superconductor single-flux-quantum (SFQ) logic family has been recognized as a promising solution for the post-Moore era, thanks to ultrafast and low-power switching characteristics of devices. Researchers have made tremendous efforts in various aspects, especially device circuit design. However, there little progress designing convincing SFQ-based architectural unit due lack understanding about its potentials limitations at level. This article provides design principles units with an extremely high-performance neural processing (NPU). To achieve our goal, we developed validated simulation framework identify critical bottlenecks performance-effective NPU. We propose SuperNPU, which outperforms conventional state-of-the-art NPU by 23 times terms computing performance 1.23 power efficiency even cooling cost 4K environment.