Ferroelectric Neuromorphic Devices and Systems
January 23 (Thursday), 2025
11:30 am to 12:30 pm (EST)
Virtual via Zoom
Abstract: The advent of novel large neural networks such as large language models e.g. ChatGPT increases the computing demand dramatically. Conventional hardware limited by the von-Neumann bottleneck is not able to follow suit. Novel pathways to meet the ever-increasing system requirements are highly needed. In this talk, Dr. Kämpfe will discuss the advantages of digital accelerators employing computing-in-memory macros, which benefit from the extreme parallelization inside memory arrays for the acceleration of costly multiply-accumulate operations. He will discuss various non-volatile, particularly ferroelectric memory concepts to demonstrate this concept and show first realizations of ferroelectric field effect transistor (FeFET) based memory crossbars for the acceleration of algorithms ranging from convolutional neural networks over decision trees to few-shot learning.
Biography: Dr. Thomas Kämpfe is group manager for Neuromorphic Systems and department manager Components & Systems at the Center Nanoelectronic Technologies at Fraunhofer IPMS. He earned his habilitation in electrical engineering in 2022 and his Ph.D. in Physics in 2016, both from TU Dresden. Following research scholar positions at UC-Boulder and Stanford University, he joined the Fraunhofer Society in 2017. He has authored over 200 peer-reviewed articles and helped organize several international conferences. His research focuses on brain-inspired computing, approximation computing, and computing-in-memory. In recognition of his contributions, Dr. Kämpfe was awarded the George E. Smith Award in 2023, the Dresden Excellence Award in 2023, the Excellent Paper Award at RFIT 2022.