Predictive Performance of Photonic SRAM-based In-Memory Computing for Tensor Decomposition

2025-03-23

AI Summary

To address bandwidth and energy-efficiency bottlenecks of the MTTKRP (Matricized Tensor Times Khatri–Rao Product) kernel in in-memory computing, this work proposes a novel photonic-in-memory architecture based on photonic static random-access memory (pSRAM). Methodologically, we first design a scalable pSRAM array integrated with wavelength-division multiplexing (WDM) to enable hyperspectral parallel computation, and develop a hardware-aware MTTKRP mapping strategy alongside the first sustained-performance prediction model for pSRAM. Our contributions are threefold: (1) a WDM-enhanced hyperspectral operation mechanism; (2) a comprehensive pSRAM performance modeling framework tailored to MTTKRP; and (3) achieving 17 PetaOps of sustained computational throughput under realistic hardware configurations—significantly outperforming state-of-the-art electronic in-memory computing systems.

Technology Category

• Machine Learning: Hardware-aware ML
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Photonics-based in-memory computing systems have demonstrated a significant speedup over traditional transistor-based systems because of their ultra-fast operating frequencies and high data bandwidths. Photonic static random access memory (pSRAM) is a crucial component for achieving the objective of ultra-fast photonic in-memory computing systems. In this work, we model and evaluate the performance of a novel photonic SRAM array architecture in development. Additionally, we examine hyperspectral operation through wavelength division multiplexing (WDM) to enhance the throughput of the pSRAM array. We map Matricized Tensor Times Khatri-Rao Product (MTTKRP), a computational kernel commonly used in tensor decomposition, to the proposed pSRAM array architecture. We also develop a predictive performance model to estimate the sustained performance of different configurations of the pSRAM array. Using the predictive performance model, we demonstrate that the pSRAM array achieves 17 PetaOps while performing MTTKRP in a practical hardware configuration.

Problem

• Modeling photonic SRAM array for in-memory computing performance
• Enhancing throughput via wavelength division multiplexing in pSRAM
• Mapping tensor decomposition kernel MTTKRP to pSRAM architecture

Innovation

• Photonic SRAM array for ultra-fast computing
• Wavelength division multiplexing enhances throughput
• Predictive model estimates 17 PetaOps performance