Reproduce “Scaling Up Analog CIM: A Simulation Framework for Topology-Aware Mapping” 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

Yuxin Ren, Danella Zhao, Huanrui Yang

Isobar estimates communication contention for models mapped across analog compute-in-memory dies, deriving transfers from legal matrix decompositions and execution dependencies, then routing them over shared physical resources. For Gemma-7B batch-one, 512-token prefill on 1,232 dies, Isobar-guided mapping reduces modeled communication time by 23.3% and Transformer-backbone latency by 18.6% versus the strongest evaluated baseline; these are modeled, not silicon-measured results.

Summary by alphaXiv.

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