Reproduce “SpiDR: A Reconfigurable Digital Compute-in-Memory Spiking Neural Network Accelerator for Event-based Perception” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Deepika Sharma, Shubham Negi, Trishit Dutta, Amogh Agrawal, Kaushik Roy

SpiDR is a fabricated digital compute-in-memory accelerator for spiking neural networks. It stores weights and partial membrane potentials in SRAM, accumulating weights in memory and supporting multiple layer mappings and precisions. The fabricated chip was evaluated on gesture recognition and optical-flow estimation. On IBM DVS Gestures, 4-bit weights yielded 87.5% accuracy and 179.5 µJ reported energy, versus 88.44% and 361.9 µJ at 8-bit weights.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.