Journal of Low Power Electronics and Applications | Interview with the Editorial Board Member—Prof. Dr. Sanghamitra Roy
We are pleased to share the following interview with Prof. Dr. Sanghamitra Roy (Professor, Electrical and Computer Engineering, Utah State University; Director, Bridge Lab), Editorial Board Member of the Journal of Low Power Electronics and Applications (JLPEA, ISSN 2079-9268).
Sanghamitra Roy received the MS degree in computer engineering from Northwestern University, Evanston, IL, USA, in December 2003, and a PhD in electrical and computer engineering from the University of Wisconsin–Madison, Madison, WI, USA, in 2008. Her PhD research was sponsored by Intel Strategic CAD Labs and the National Science Foundation.
She is currently a Professor in the Electrical and Computer Engineering Department, Utah State University, Logan, UT, USA. She has authored more than 50 peer-reviewed publications in top-tier journals and conferences, as well as a book chapter in VLSI Design Automation. Her current research interests include VLSI circuit design and optimization and exploring reliability-aware novel circuit styles and architectures.
The following is an interview with Prof. Dr Sanghamitra Roy:
1. Could you briefly introduce your research area and the focus of your recent work?
My research sits at the intersection of energy-efficient AI hardware and hardware security. I direct the Bridge Lab at Utah State University, where we design processors and accelerators that deliver more inference per joule without giving up trustworthiness. Recent work has focused on architectural techniques for large language model inference, exploiting the statistical structure of the workload so that precision, voltage, and dataflow are not all sized for the worst case. In parallel, we study timing-error resilience and the security of AI hardware, including side-channel leakage and fault-injection attacks. We also release open-source simulation tools so that others can reproduce and build on the results.
2. As a journal Editorial Board Member, what do you believe is the most critical step in maintaining the journal’s academic standards — evaluation of innovation during the pre-check stage, the precise selection of reviewers, or the final decision on controversial manuscripts?
All three matter, but if I must choose one, it is the precise selection of reviewers. Pre-check catches what is plainly out of scope, and editorial judgment resolves contested cases—but both of those steps depend on the quality of the reviews in front of us. A well-matched reviewer sees in an hour what a mismatched one will not see in a week, and mismatched reviews are the root cause of most disputed decisions. I look for reviewers whose own recent work touches the paper’s specific contribution, not merely its keywords, and I like to pair a domain expert with someone from an adjacent area who will test whether the claims travel.
3. How do you see this research field evolving in the next few years?
Low-power design is being reorganized around AI inference. For two decades, the pressure came from mobile and embedded devices; now it comes from large language models and, increasingly, agentic workloads that invoke a model many times for a single user request. Energy per token is becoming the metric that matters, and I expect the gains to come from cross-layer co-design rather than any single trick—arithmetic, dataflow, memory hierarchy, and runtime scheduling tuned together. I also expect security to stop being a separate track. Aggressive voltage and precision scaling widens the attack surface for side channels and fault injection, so power, accuracy, and trustworthiness will have to be optimized jointly.
4. Are there any emerging technologies or trends that you believe will significantly influence this field?
Several. Mixture-of-experts and sparse models change the hardware problem: the bottleneck moves from raw compute to routing, memory bandwidth, and the idle power of experts that are not selected. Co-packaged optics and advanced packaging are arriving at the same moment, which makes thermal behavior a first-class design constraint rather than a back-end check. On-device inference keeps pushing tiny transformers into milliwatt budgets. On the security side, a growing class of hardware-level attacks targets the model itself rather than the software around it, showing that an adversary can degrade a network’s accuracy without ever touching its code. Finally, I would watch the design and simulation tools themselves, since they are what make any of this tractable at the system level.
5. What aspects do you value most when choosing journals for publication?
First, the quality and fairness of the review. A journal earns my submissions when the reviews are substantive, and the process is transparent about why a decision was reached. Second, the readership: I want a paper to reach the people who will build on it, which means the journal’s scope must genuinely match the work rather than merely overlap with it. Third, turnaround: in a field moving this quickly, a year-long cycle costs the community more than it protects it. Open access matters to me as well, along with support for artifacts and reproducible data, since our students learn as much from reusing other groups’ setups as from reading the text.
6. Do you have any suggestions for the development of this journal?
Looking at the highly-cited papers from 2024 and 2025, JLPEA’s strongest pull is in neuromorphic and spintronic computing, wireless-sensor energy efficiency, and ultra-low-power analog and edge machine learning. I would build deliberately on that. A special issue on energy-efficient LLM and agentic-AI inference—from tiny transformers at the edge up through accelerator and package-level co-design—would carry the journal’s existing edge-ML readership into where the field’s attention is now. I would also encourage a companion track on power–security co-design, which is underrepresented relative to its importance. More concretely: invite artifact and dataset submissions alongside papers and keep publicizing the highly-cited list; it shows authors where the journal’s real center of gravity lies.