Novel Electrical and Optical Technologies for Efficient Neural Network Inference and Training

Bert Jan Offrein, IBM Research Europe - Zurich

Über den Vortrag:

Artificial intelligence has become part of our daily life. The size of neural networks providing this functionality is increasing fast with a strong impact on the energy footprint. New technologies are required to improve the performance and power-efficiency of neural network training and inference. In this presentation, electrical and optical silicon in-memory computing technologies will be discussed, mitigating critical performance limitations of today’s computing systems.

Curriculum Vitae

Bert Offrein received his Ph.D. degree in nonlinear integrated optics from the University of Twente in The Netherlands in 1994. He then joined IBM Research Europe - Zurich and contributed to establishing and commercializing adaptive integrated optical technology for DWDM networks. From 2004 to 2016, Bert Offrein was managing the photonics group, addressing optical interconnects for computing systems. Since 2016, he is leading the neuromorphic devices and systems group, focussing on novel hardware for neural networks. Bert Offrein is co-author of over 150 publications and co-inventor of more than 40 patents. He currently is a Principal Research Scientist at IBM Research Europe and is Professor by special appointment on ‘analog signal processing’ at the University of Twente.