Reproduce “Predictive Performance of Photonic SRAM-based In-Memory Computing for Tensor Decomposition” 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
Sasindu Wijeratne, Sugeet Sunder, Md Abdullah-Al Kaiser, Akhilesh Jaiswal, Clynn Mathew, Ajey P. Jacob, Viktor Prasanna
Researchers from the University of Southern California and the University of Wisconsin-Madison developed a predictive performance model for photonic SRAM (pSRAM)-based in-memory computing to accelerate tensor decomposition. The model demonstrates that the pSRAM array can achieve a sustained performance of 17 PetaOps with 8-bit precision for the core Matricized Tensor Times Khatri-Rao Product (MTTKRP) operation.
Key insights
- Photonic computing provides inherent parallelism through wavelength multiplexing, enabling multiple data elements to be processed simultaneously within a single pSRAM cell.
- Performing computations directly within photonic SRAM (pSRAM) memory units can effectively address the 'memory wall' by eliminating data movement overheads.
- The pSRAM architecture, using cross-coupled microring resonators and photodiodes, supports both optical data storage and direct in-memory multiplication operations, compatible with standard CMOS fabrication.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.