Reproduce “ReCIM-HD: A 40 nm On-Device Learning Accelerator with Hyperdimensional Computing via Importance-Based Hybrid Digital-ReRAM CIM” 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

Keming Fan, Chang Eun Song, Soumil Jain, Weihong Xu, Gopabandhu Hota, Haichao Yang, Leo Liu, Meng-Fan Chang and 4 more

ReCIM-HD combines hyperdimensional computing with ReRAM compute-in-memory inference and on-device learning without gradient-based training. It stores class vectors in nonvolatile ReRAM, comparing most dimensions there and a small offline-selected subset digitally. On UCIHAR, with dimension 1,024 and 10% protected dimensions, it reports 3.71% higher accuracy than CIM-only, with gains of 1.91–8.46% across three tasks. Separately, energy at dimension 1,024 is 9.32× lower than digital-only.

Summary by alphaXiv.

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