Research Areas
The Center's research spans the whole computer-system stack — the devices and chips that sense and compute, the architectures and software that run models, the signal processing and control that close the loop with a patient, and the power infrastructure all of it depends on. The areas below are not separate programs. The Center's projects are chosen precisely because they cut across several of them at once, in service of a clinical problem that no single layer can solve.
Device & Chip Design
Healthcare puts constraints on hardware that general-purpose silicon was never asked to meet: a device may have to run for years on a harvested microwatt, sit inside the body, or survive sterilisation. Work in this area builds the sensing and computing substrate to those requirements — nanomaterial and two-dimensional-material biosensors, fiber-optic sensing that can be threaded through a catheter or embedded in an instrument, photonic and in-memory computing primitives that perform inference without moving data, and ultra-low-power mixed-signal front ends that digitise physiological signals at the point they are measured.
Computer Architecture
Clinical AI workloads look nothing like the benchmarks processors are tuned for. They mix volumetric imaging, continuous waveforms, genomic data and free text; they run under hard latency limits in an operating room and under hard energy limits on an implant; and they carry data that must stay confidential even from the machine's own operator. This area designs accelerators and memory systems for that profile — near-data and memory-centric organisations for multimodal inputs, architectural support for confidentiality and integrity, and designs that keep working when power is intermittent rather than assuming it never is.
Hardware/Software Interfaces and Co-design
The largest gains in a medical device usually come from moving the line between hardware and software, not from optimising either side of it. Research here co-designs across that boundary: mapping model structure onto the analog front end so that most data never needs to be digitised, building always-on wake-up hierarchies that spend energy only when something clinically interesting happens, compensating for sensor drift and calibration in the loop rather than in the lab, and prototyping whole mixed-signal systems in simulation so that a clinical requirement can be traced all the way down to a circuit before anything is fabricated.
Signal Processing & Control
Physiological signals are noisy, non-stationary and different in every patient, and a clinician has to be able to say why a system reached its conclusion. This area works on both halves of that problem: statistical signal processing for neural and physiological recordings, including brain-computer interfaces and the network analysis of neuroimaging data; closed-loop control for assistive, prosthetic and rehabilitation devices that have to stay stable while a person moves through the world; and mechanistic, causal modelling that yields explanations a clinician can interrogate rather than scores they must take on trust.
Computing Framework & Security
A healthcare AI system is rarely one computer. It is a wearable talking to a phone, talking to an edge server in a clinic, talking to a hospital data center — across an unreliable network, under privacy rules, and often across institutions that cannot pool their data. This area builds the frameworks that hold that together: running foundation models on devices whose memory and energy budgets rule out the usual approach, federated and privacy-preserving training across sites, resilience and anomaly detection for connected clinical devices, and interfaces that fit the way clinicians actually work rather than demanding they adapt.
Power Infrastructure & Support
AI is making hospitals far more energy-hungry at exactly the point where an outage is least acceptable, and the same problem appears in miniature on every battery a patient carries. Work in this area covers both scales: the power delivery, conversion efficiency and thermal limits of AI computing inside a care facility; resilient architectures and microgrids that keep clinical systems running through a grid disturbance; and the energy harvesting and wireless power transfer that decide how long a wearable or implanted device can go between interventions.