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.
Before you apply: We strongly recommend that you contact the supervisors for this project before you apply.
Main Supervisor: Dr. Shady Agwa,
Email: [email protected]
For Further Information please visit the full post:Here
