Reproduce “TeraNoC: A Multi-Channel 32-bit Fine-Grained, Hybrid Mesh-Crossbar NoC for Efficient Scale-up of 1000+ Core Shared-L1-Memory Clusters” 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

Yichao Zhang, Zexin Fu, Tim Fischer, Yinrong Li, Marco Bertuletti, Luca Benini

The authors present TeraNoC, a hybrid interconnect for a 1,024-core cluster sharing 4,096 banks. It uses crossbars for nearby core-to-memory traffic and a narrow-channel mesh for longer trips. In a MatMul(f32) comparison, remapping raises reported global L1 bandwidth 2.7×. Its 12-nm physical-design implementation has 37.8% less die area than a same-scale hierarchical-crossbar cluster; it is not a fabricated-chip result.

Summary by alphaXiv.

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