Mimi Xie
University of Texas at San Antonio, TX·Computer Science·Early-career
Tiny sensors that run on harvested energy, from light, heat, or vibration instead of batteries, can be left in remote or hard-to-reach places for years without maintenance. The catch is that harvested power is unreliable: the device keeps losing power and restarting, which makes running machine learning on the device difficult.
- Fields
Computer Science
Engineering
Materials Science
- Topics
battery-less systems
energy harvesting
intermittent computing
- Grants
- 2 known grants, through May 2030
What they work on
- Redesign and optimize tiny machine learning models for battery-less, energy-harvesting IoT devices so that inference and decisions still happen on time under strict power, memory, and computation limits.
- Build a framework that adapts sensor sampling, decision-making, and communication strategies to fluctuating ambient energy and changing environmental conditions rather than fixed sampling rates.
- Design code update mechanisms that reliably keep programs on deployed battery-less devices current despite frequent power failures, so models do not become obsolete as inputs drift.
- Demonstrate a functional skyrmion racetrack memory by combining voltage-controlled skyrmion motion with an industrial 8-inch-wafer magnetic tunnel junction stack, and characterize write, shift, and read error rates, speed, and energy.
- Create a mutable lookup-table, compute, and memory unit from racetrack memory that can operate as multi-context FPGA-style logic, parallel processing-in-memory logic, and analog-like storage, connected by a mesh network-on-chip.
- Develop a multi-level compilation flow that translates high-level PyTorch and C/C++ descriptions into binaries for the R3AP platform, with design-space exploration and runtime reconfiguration for IoT and cyber-physical applications.
More about the research
Work in this lab spans the full computing stack, from new memory devices up through compilers and applications, to make such self-powered systems both intelligent and durable.
Why it matters
Battery-less sensors could monitor remote environments, infrastructure, and industrial systems for years without battery replacement or disposal, and on-device intelligence would let them act without constant network contact. The memory-centric accelerator work could reduce the energy and delay spent shuttling data in future IoT, cyber-physical, high-performance, and cloud systems, and the fabricated platform would give designers a way to prototype and measure emerging memory technologies rather than only simulate them.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Shaahin Angizi
New Jersey Institute of Technology, NJ·Engineering·Early-career
Design reconfigurable circuits and architectures for a multi-modal in-sensor accelerator that moves AI computation onto sensor nodes as efficiently as possible.
Phuc Nguyen
University of Massachusetts, MA·Computer Science·Early-career
Build computational models that translate jaw movement and faint facial vibrations, captured by an ear-worn sensor, into speech sounds and commands so users can control devices silently.
Nirupam Roy
University of Maryland, MD·Computer Science·Early-career
Enable sensing of large-scale gestures and movements near walls, floors and tabletops using coded signal projection through passive acoustic micro-structures and an adaptive intermittent computing architecture for battery-free operation.
The school
University of Texas at San Antonio: 10 labs on PI Pages, $172M a year on research, 129 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.