Introduction
The global burden of health issues is dominated by chronic, non-communicable diseases, which cause the majority of illness, disability, and death worldwide, with significant economic and social impacts(M. et al., 2025).
Therapeutic interventions and disease treatments are most effective when they are diagnosed early; however, there is a lack of access to timely detection and monitoring of multiple diseases, including dementia, cancer, respiratory and cardiovascular diseases, and as a result, many cases are misdiagnosed or missed and left untreated.
In this context, the significance of continuous health monitoring is on the rise; however, hospital visits for such monitoring are time-consuming and place an additional burden on already stretched national healthcare systems, which are short on resources, staff, and finances. To avoid such challenges, remote health monitoring via wearable medical devices (WMDs) offers a viable solution for timely disease detection and prevention (Khan & da Silva, 2024).
WMDs measure several physiological signals, including tissue oxygenation, heart rate, blood pressure, and pulse, among others, but these data are typically collected by different devices and processed independently. The advent of Artificial Intelligence (AI) has enabled the integration of multi-modal information to detect complex patterns and causal interactions indicative of abnormalities, which have revolutionised the diagnosis of diseases(D’Adderio & Bates, 2025).
Currently, AI-based processing pipelines are implemented on power-hungry General-Purpose Processors (GPPs), which are not suitable for integration into wearable devices. An alternative solution is cloud processing, which addresses scalability issues; however, cloud services rely on a stable internet connection, and the risks of latency, disconnection, and data loss are inherent. Furthermore, data centres that support cloud processing consume ~3% of the world’s electricity, a figure expected to double by 2030 (Chen, 2025), in opposition to the NHS's goal of net-zero carbon emissions. In addition, sharing of personal data and sensitive information over networks poses huge risks of data breaches and corruption.
To address power consumption and privacy concerns, edge intelligence has been introduced (Covi et al., 2021; Shumba et al., 2023; Su et al., 2023). It refers to a computational framework that processes and analyses data directly where it is generated. This approach drastically reduces power demands, saves bandwidth and improves data privacy.
Energy-efficient solutions include Application-Specific Integrated Circuits (ASICs) and Microcontrollers (MCUs); however, the former require long, costly development, and the latter lack the high-level functions required for complex signal processing. Field Programmable Gate Arrays (FPGAs) offer an energy-efficient solution whilst providing the necessary computational power (Xia et al., 2025). FPGAs are renowned for their real-time and parallel capabilities, enabling faster processing while maintaining high precision at high data density (Xia et al., 2025).
Research Aim
The aim of this project is to exploit FPGA software and hardware capabilities to design energy-conscious AI-based processing algorithms.
Crucially, this approach will open the opportunity to scale from single-point measurements to multi-channel systems, such as wearable electroencephalogram and near-infrared spectroscopy, for robust, precise diagnostics. The project will deliver AI FPGA-based accelerators to transform how health is monitored, and how diseases are diagnosed and managed, bridging the gap between AI innovation and practical implementation.
Although the case study focuses on healthcare, the algorithms can be translated to other domains that require AI processing capabilities while minimising energy consumption, such as transportation (electric vehicles) and industries. The strategies will include, but are not limited to, simplification of arithmetic operations, parameter quantisation, feature reduction, signal format conversion, bit width customisation, and parallelism.
References:
Chen, S. (2025). Data centres will use twice as much energy by 2030 — driven by AI. Nature. https://doi.org/10.1038/d41586-025-01113-z
Covi, E., Donati, E., Liang, X., Kappel, D., Heidari, H., Payvand, M., & Wang, W. (2021). Adaptive Extreme Edge Computing for Wearable Devices. In Frontiers in Neuroscience (Vol. 15). https://doi.org/10.3389/fnins.2021.611300
D’Adderio, L., & Bates, D. W. (2025). Transforming diagnosis through artificial intelligence. In npj Digital Medicine (Vol. 8, Number 1). https://doi.org/10.1038/s41746-025-01460-1
Khan, M. I., & da Silva, B. (2024). Harnessing FPGA Technology for Energy-Efficient Wearable Medical Devices. Electronics, 13(20). https://doi.org/10.3390/electronics13204094
M., N., H.H., K., A., B., M.A., A., H., A., H.S., A., B.J., A., U.O., A., C., A., M., A., F., A., H., A., A.H.A., A. A. M., S., A. E., A.N., A., M.A., A., E.M., A., B.A., A., N.M.I., A. R., … C.J.L., M. (2025). Global burden of 292 causes of death in 204 countries and territories and 660 subnational locations, 1990-2023: a systematic analysis for the Global Burden of Disease Study 2023. The Lancet, 406(10513).
Shumba, A. T., Montanaro, T., Sergi, I., Bramanti, A., Ciccarelli, M., Rispoli, A., Carrizzo, A., De Vittorio, M., & Patrono, L. (2023). Wearable Technologies and AI at the Far Edge for Chronic Heart Failure Prevention and Management: A Systematic Review and Prospects. In Sensors (Vol. 23, Number 15). https://doi.org/10.3390/s23156896
Su, X., An, L., Cheng, Z., & Weng, Y. (2023). Cloud–edge collaboration-based bi-level optimal scheduling for intelligent healthcare systems. Future Generation Computer Systems, 141. https://doi.org/10.1016/j.future.2022.11.005
Xia, Y., Chen, J., Li, J., Gong, T., Vidal-Rosas, E. E., Loureiro, R., Cooper, R. J., & Zhao, H. (2025). A Deep-Learning Empowered, Real-Time Processing Platform of fNIRS/DOT for Brain Computer Interfaces and Neurofeedback. IEEE Transactions on Neural Systems and Rehabilitation Engineering : A Publication of the IEEE Engineering in Medicine and Biology Society, 33, 1220–1230. https://doi.org/10.1109/TNSRE.2025.3553794