Speaker: Iason Chalas, IBM Research Zurich
Date and time: Tuesday, July 7, 2026, at 2:00 p.m. CEST
Location: online via Zoom
Abstract:
Large language models are transforming modern computing, but their deployment at scale is increasingly limited by the cost and energy of inference. A key bottleneck lies in the separation between memory and computation, where data movement dominates overall system efficiency.
In this talk, I will introduce analog in-memory computing (AIMC) as an emerging paradigm that addresses this challenge by performing matrix-vector multiplications operations directly in the memory arrays. I will discuss recent advances in AIMC hardware and architectures, including opportunities enabled by 3D memories and model designs tailored for efficient execution (e.g MoEs).
Finally, I will outline how recent algorithmic developments make LLMs robust to the non-idealities of analog hardware, opening the door to practical and scalable deployment. Together, these advances point toward a new class of energy-efficient systems for large-scale AI.
Short Bio:
Iason Chalas is a PhD candidate at IBM Research-Zurich and ETH Zurich, where he works on large language models for next-generation AI hardware and energy-efficient computing. His research focuses on developing algorithms that enable large language models to run reliably on analog hardware despite noise and device-level imperfections. He holds a Master’s degree in Data Science from ETH Zurich and a Diploma in Electrical and Computer Engineering from the National Technical University of Athens.
For a list of past and upcoming NHR PerfLab seminar events, please see: https://hpc.fau.de/research/nhr-perflab-seminar-series/