Artificial Intelligence (AI) has revolutionized our daily lives in various domains with spectacular applications; however, it poses emerging challenges due to the ever-growing demands of performance and energy efficiency. Meanwhile, Moore’s law is coming to an end due to the physical limits of technology scaling, taking into consideration that AI models are being scaled by orders of magnitude higher than Moore’s Law. These challenges become more significant for edge AI applications (such as bio-electronic systems for health care or monitoring and surveillance for security & defense); because of the resource-constrained environment where only limited hardware budgets (Energy/Power/Area) are available.
In this project we propose building reconfigurable computing architectures that maximize not only the hardware specialization and computing parallelism for a higher energy efficiency but also the data utilization to mitigate the memory wall’s penalty. These reconfigurable architectures will be adaptable to the runtime AI workloads, by reconfiguring the hardware with the optimum architectural setup through different combinations of bit-precisions, data representations and micro-architectural organizations; thereby enabling the highest energy-efficient execution per each phase of the AI workload, unlike conventional fixed accelerators which are optimized holistically for a few workloads.
This project is expected to start in September 2027.
The University of Manchester, a Russell Group member, excels in diverse courses and research. Home to 25 Nobel laureates, we strive for academic excellence and aim to be a top …
(School of Engineering PhD Scholarships) Reconfigurable Computing Architectures for Energy-efficient Edge AI
Full-time
The University of Manchester
Manchester, GB
Application deadline: January 15, 2027
Artificial Intelligence (AI) has revolutionized our daily lives in various domains with spectacular applications; however, it poses emerging challenges due to the ever-growing demands of performance and energy efficiency. Meanwhile, Moore’s law is coming to an end due to the physical limits of technology scaling, taking into consideration that AI models are being scaled by orders of magnitude higher than Moore’s Law. These challenges become more significant for edge AI applications (such as bio-electronic systems for health care or monitoring and surveillance for security & defense); because of the resource-constrained environment where only limited hardware budgets (Energy/Power/Area) are available.
In this project we propose building reconfigurable computing architectures that maximize not only the hardware specialization and computing parallelism for a higher energy efficiency but also the data utilization to mitigate the memory wall’s penalty. These reconfigurable architectures will be adaptable to the runtime AI workloads, by reconfiguring the hardware with the optimum architectural setup through different combinations of bit-precisions, data representations and micro-architectural organizations; thereby enabling the highest energy-efficient execution per each phase of the AI workload, unlike conventional fixed accelerators which are optimized holistically for a few workloads.
This project is expected to start in September 2027.
The HiPEAC project has received funding from the European Union’s Horizon Europe research and innovation funding programme under grant agreement number 101296676. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them.
#J-18808-Ljbffr…