Design and Evaluation of Reliable Multi-Precision Systolic GEMM Acceleration for AI on RISC-V Edge Systems

Scientific-Disciplinary Group

09/IINF-01 - Electronics

Description

This Incarico di Ricerca position, aligned with the ARCHYTAS project, focuses on the design and evaluation of reliable matrix-acceleration solutions for AI workloads. Research will address: (1) extending the architecture and micro-architecture of a systolic GEMM accelerator for AI inference and fine-tuning on edge SoCs, with support for emerging AI-oriented data formats and multi-precision computation; (2) analyzing reliability and fault tolerance of AI algorithms and their execution on the accelerator, including numerical formats and alternative compute-unit micro-architectures; (3) integrating the accelerator into heterogeneous edge compute systems, including edge-space scenarios; and (4) validating functionality, characterizing performance, power, and area, and benchmarking representative AI kernels, also assessing resilience to radiation-induced faults. The goal is to advance reliable, energy-efficient acceleration in heterogeneous multi-tile systems.

Job posting website

https://bandi.unibo.it/ricerca/incarichi-di-ricerca

Funding body

ALMA MATER STUDIORUM - UNIVERSITA' DI BOLOGNA - - DIPARTIMENTO DI INGEGNERIA DELL'ENERGIA ELETTRICA E DELL'INFORMAZIONE "GUGLIELMO MARCONI"

How to apply

Other

Selection process