ECE PhD Thesis Defense: Guowei Yang

Title: Domain-Specific Accelerators Using Optically-Addressed Phase Change Memory

Presenter: Guowei Yang

Advisor: Professor Ajay Joshi

Chair: TBA

Committee: Professor Ajay Joshi, Professor Ayse K. Coskun, Professor Sabrina M. Neuman, Professor Martin Herbordt, Professor Carlos A. Ríos Ocampo

Google Scholar Link: https://scholar.google.com/citations?user=BdQnapwAAAAJ

Abstract: In recent years, the exponential growth in data generation and the increasing complexity of computational tasks have highlighted the need for more efficient computing solutions. To meet this demand, researchers have developed domain-specific accelerators (DSAs) for various applications, including machine learning (ML), combinatorial optimization, and fully homomorphic encryption (FHE). However, traditional electronic accelerators face significant challenges in performance and energy efficiency. Optically-addressed phase change memory (OPCM) has emerged as a promising solution for accelerating computing as it offers high bandwidth, superior energy efficiency, and processing-in-memory (PIM) capabilities.

In this prospectus, we focus on designing OPCM-based DSAs, by applying device-, architecture-, and algorithm-level optimizations. We first present our work on an ML accelerator using OPCM. This work introduces a system-level design and proposes a thresholding and reordering technique to reduce the OPCM programming overhead, achieving up to a 65.2x throughput improvement over existing photonic accelerators for practical DNN workloads. Then, we present SOPHIE, a Scalable Optical PHase-change-memory based Ising Engine. SOPHIE integrates architectural, algorithmic, and device optimizations to address scalability challenges in Ising machines. SOPHIE is 3x faster than the state-of-the-art photonic Ising machines on small graphs and 125x faster than the FPGA-based designs on large problems. SOPHIE alleviates the hardware capacity constraints, offering a scalable and efficient alternative for solving Ising problems. Finally, we propose PHAT, a Photonic Accelerator for FHE over the torus (TFHE) that leverages OPCM. PHAT introduces a novel electro-photonic architecture featuring OPCM-based fast Fourier transform (FFT) units, a twiddle-stationary dataflow optimized for OPCM, and a scheduling mechanism to improve utilization of FFT units. PHAT achieves 1.39x-1.77x speedup over the best application-specific integrated circuit (ASIC) accelerator across four programmable bootstrapping configurations, and delivers 2.14x-5.10x speedup on real-world TFHE-based ML workloads. These results demonstrate that PHAT significantly improves the practicality and efficiency of TFHE, paving the way for scalable, privacy-preserving computation in cloud environments.