Matrix-Matrix Multiplication Through Hyperspectral Compute-in-Memory
Byoung Jun Park, Mostafa Honari Latifpour, Y. Yamamoto, Myoung‐Gyun Suh · 2024
We propose a hyperspectral compute-in-memory architecture using optical frequency combs and programmable optical memories. By fully utilizing frequency, space, and time dimensions, this approach demonstrates energy-efficient and highly-scalable op-tical Matrix-Matrix Multiplication, potentially exceeding PetaOPS-level performance.