Sudeep Pasricha

Colorado State University, USA

Carbon-Efficient AI Accelerators for Sequence Learning Applications with Silicon Photonics

Abstract

State-space models (SSMs) have emerged as a compelling alternative to attention-based architectures for complex sequence learning applications, offering linear complexity while maintaining strong accuracy. However, efficient hardware support for SSMs remains limited. Unlike transformers, SSM workloads are dominated by recurrent state updates, element-wise operations, small matrix-vector computations, and dynamic state transitions, resulting in low arithmetic intensity, dataflow irregularity, high data movement, and poor utilization on conventional GPUs and TPUs. In addition to these performance challenges, the carbon footprint of modern computing systems, spanning both operational energy and embodied manufacturing footprint, has become a critical concern. While optical accelerators offer promising energy efficiency, achieving sustainable gains in carbon costs requires careful co-design across circuit and architectural levels. In this talk, I will present STORM, the first systolic array-based optical accelerator tailored for SSM workloads. STORM combines reconfigurable optical systolic arrays with hybrid optical–electronic execution to support both time-invariant SSMs (e.g., S4/S5) and selective time-varying SSMs (e.g., Mamba) within a unified architecture. By enabling in-place state storage and minimizing opto-electronic conversions, STORM reduces data movement and supports high-throughput streaming execution. Additionally, STORM integrates design optimizations to significantly reduce both operational and embodied carbon.

Biography

Sudeep Pasricha is the Aram and Helga Budak Professor in the Department of Electrical and Computer Engineering at Colorado State University. His research focuses on the design of AI/ML algorithms, hardware architectures, and hardware-software co-design techniques for energy-efficient, fault-tolerant, real-time, and secure embedded computing. Dr. Pasricha has co-authored 7 books, multiple patents, and 350+ papers that have received 20 best paper awards and nominations at IEEE and ACM conferences. He currently serves as Editor-in-Chief of the ACM Transactions on Embedded Computing Systems (TECS). Dr. Pasricha is the recipient of the AFOSR Young Investigator Award, the IEEE Computer Society TCVLSI and TCSC Mid-Career Research Achievement Awards, and the ACM SIGDA Distinguished Service Award. He is a Fellow of the IEEE and Distinguished Member of the ACM.