Datum und Uhrzeit:
Ort: Seminar room OMZ (U013, INF 350, floor -1)
Title: Metis AI Processing Unit – a 210 TOPS SoC Powered by Digital in-Memory Computing
Abstract: This talk explores the emerging field of In-Memory Computing (IMC), with a focus on the AxeleraAI Metis AI Processing Unit (AIPU). IMC integrates computation directly within memory, bypassing the traditional data transfer bottlenecks between the CPU and memory, and offering significant performance and energy efficiency benefits. The Metis AIPU’s core feature is a digital, SRAM-based IMC-array, designed to efficiently handle AI workloads. The talk will provide an overview of the system architecture of the Metis SoC, highlighting its key components and design considerations. Additionally, the talk will cover the importance of compilation strategies in optimising hardware utilisation, focusing on how computational tasks are mapped to maximise performance.
CV: Dr. Dennis Rieber is a Senior Engineer in System Architecture at AxeleraAI, where he specializes in compilers and architecture evaluation for AI Processing Units based on In-Memory Compute technology. From 2018 to 2023, he completed his PhD at Bosch Research, under the guidance of Prof. Dr. Holger Fröning from the Computing Systems Group at Heidelberg University. His research focused on optimizing the mapping of Deep Neural Networks to hardware accelerators. Prior to his current role, Dr. Rieber worked as a Research Engineer at Bosch, where he contributed to hardware-aware Neural Architecture Search and the deployment of Deep Neural Networks on embedded devices.