Optimizing Integrated Photonic Neural Networks under Imperfections
Integrated photonic neural networks (IPNNs) are emerging as promising successors to electronic AI accelerators as they offer substantial improvements in speed and energy efficiency. However, the underlying devices in IPNNs are susceptible to uncertainties stemming from lithographic variations and thermal crosstalk and can experience imprecisions due to non-uniform insertion loss and quantization errors due to low-precision encoding in the tuned phase angles. In this talk, we will first characterize the impact of such imprecisions and uncertainties in IPNNs in a bottom-up approach. Next, we will explore several novel photonic hardware-aware optimization approaches to improve the resilience of IPNNs under imperfections.
Sanmitra Banerjee received the B.Tech. degree from the Indian Institute of Technology, Kharagpur, Kharagpur, West Bengal, in 2018, and the M.S. and Ph.D. degrees from Duke University, Durham, NC, USA, in 2021 and 2022, respectively. He is currently a Senior DFX Methodology Engineer with NVIDIA Corporation, Santa Clara, CA, USA. His research interests include machine learning-based DFX techniques, and the fault modeling and optimization of emerging AI accelerators under process variations and manufacturing defects.
If you wish to modify any information or update your photo, please contact the Web Chair at the following address: