From in-memory computing to secure machine-learning hardware, Saion is exploring how integrated circuits can be designed for both efficiency and security.
At the intersection of integrated circuits, machine learning, and hardware security, Saion K. Roy is investigating a question that is becoming increasingly important as intelligent computing moves closer to the edge: How can machine-learning hardware be made not only energy efficient, but also secure?
Roy is a Postdoctoral Research Associate in the Department of Electrical and Computer Engineering at Northeastern University. His research focuses on secure, energy-efficient machine-learning accelerators, with an emphasis on understanding how decisions at the algorithm, architecture, and circuit levels influence both computing efficiency and hardware security.
Watch: Saion explains his work researching secure and energy-efficient machine learning accelerators
Roy’s path to this research began with his work on emerging non-volatile memory technologies and in-memory computing. He received his B.Tech. and M.Tech. degrees in Electronics and Electrical Communication Engineering from the Indian Institute of Technology Kharagpur in 2018, followed by a Ph.D. in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign in 2024. His doctoral research, Energy-efficient Resistive In-memory Computing Architectures, examined how emerging memory technologies perform computation directly where data is stored, reducing the energy costs of moving data between memory and computing blocks.
Today, Roy is extending that work toward security.
His research examines how algorithm mapping, architectural decisions, and silicon implementation affect the efficiency, robustness, and attack surface of machine-learning systems. Rather than treating security as an additional layer applied after a system has been designed, his work aims to make security a fundamental consideration from the beginning.
“Saion’s research develops secure and energy-efficient machine-learning systems through algorithm architecture circuit co-design.”
This approach considers security and privacy as first-class constraints alongside traditional hardware-design objectives such as accuracy, energy consumption, and reliability.
In his 2026 work, SLAC, accepted at the ACM Conference on Computer and Communications Security (CCS) 2026, Roy and collaborators investigated access-driven CPU-to-GPU side-channel attacks through system-level cache behavior on Apple Silicon. They then used this side-channel to recover private information from graph neural networks and large language models. A second paper, ROBBIN, accepted at the IEEE/ACM International Conference on Computer-Aided Design (ICCAD) 2026, shows how Rowhammer bit flips in a specific DRAM chip can plant a backdoor in a deployed neural network. Together with his prior work, these projects highlight a broader research direction: understanding how the physical and architectural characteristics of computing hardware can create new security vulnerabilities.
Roy’s research has appeared in leading conferences and journals, including AsiaCCS, ICCAD, IEDM, ESSCIRC, CICC, ISCAS, TCAD, JSSC, JxCDC, and OJ-SSCS. He has also contributed to the research community through service on Technical Program Committees for conferences including ISCA, MICRO, and ICCAD.
As machine learning becomes increasingly embedded in everyday devices and specialized hardware, the boundary between computing efficiency and cybersecurity is becoming harder to separate. Roy’s work addresses that convergence directly, seeking to develop machine-learning systems that are secure by design rather than secured as an afterthought.
For Roy, the challenge is ultimately one of co-design: building the algorithms, architectures, and circuits together so that future intelligent systems can achieve high performance without compromising security or efficiency.
Selected Publications
- Tianhong Xu, Saion K. Roy, Ruyi Ding, A. Adam Ding, and Yunsi Fei. SLAC: Access-Driven CPU-to-GPU Side-channel Attacks via System-Level Cache on Apple Silicon. ACM SIGSAC Conference on Computer and Communications Security (CCS), 2026, to appear.
- Saion K. Roy*, Yufei Wang*, A. Adam Ding, and Yunsi Fei. ROBBIN: Rowhammer-Based Backdoor Injection during Inference. IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2026, to appear.
- Saion K. Roy, Ziyue Zhang, A. Adam Ding, and Yunsi Fei. Attack from Shadows: Unsupervised Side-channel Transfer Learning across Devices and Modalities. 21st ACM Asia Conference on Computer and Communications Security (AsiaCCS), 2026.