AASS Seminar
22 maj 2025 13:00 – 14:00 T101, Teknikhuset
The research centre AASS arranges a seminar with Marco Angiolo, PhD student at the Sapienza University of Rome, Italy.
Who: Marco Angioli, PhD student, Sapienza University of Roma, Italy
What: Seminar: "Hardware Acceleration for Sustainable AI in Embedded Devices"
When: May 22, 2025 at 13:00
Where: T101, Teknikhuset
Remote attendance: https://oru-se.zoom.us/j/65096623293
Abstract
The explosive growth of AI is driving an unprecedented demand for computational power and energy. With Moore’s Law slowing and Dennard scaling long dead, technology scaling alone is no longer enough to meet the demands of modern models. The GPUs that have driven much of AI’s progress are now reaching their limits — both due to their high energy consumption and their unsuitability for the fast-growing domain of real-time, low-power edge devices. Fundamental changes are needed in how we design and execute AI workloads.
This talk explores how domain-specific hardware acceleration and hardware-software co-design can overcome these limitations. We will examine the challenges of deploying AI at the edge, the shift toward heterogeneous architectures that combine general-purpose cores with application-specific accelerators, and the critical role of RISC-V as an open, extensible ISA in this context. We’ll then highlight how algorithmic optimization and hardware-aware co-design can yield substantial improvements in both performance and efficiency. Rather than adapting hardware to suit generic algorithms, co-design encourages rethinking the algorithm itself, making better use of available computational resources, simplifying dataflows, or even redesigning the model structure. Through concrete examples and ongoing research, the talk will illustrate how these strategies can dramatically reduce energy consumption and execution time, enabling deployment in time-critical, resource-constrained embedded systems.
Biography
Marco Angioli is a PhD student at the Sapienza University of Rome, Italy and he is currently doing a visiting research period at Örebro University, Sweden. He received his M.S. degree "cum laude" in electronics engineering in 2022. His current research activity focuses on the digital design of hardware accelerators for the implementation and execution of artificial intelligence algorithms on low-power embedded systems with stringent real-time requirements. Contact him at marco.angioli@uniroma1.it