Archived Seminars

Abstract: Silicon-photonic AI accelerators (SPAAs) are being explored as promising successors to CMOS-based accelerators owing to their ultra-high speed and low energy consumption. However, their accuracy and energy efficiency can be catastrophically degraded in the presence of inevitable imperfections such as fabrication process variations, optical losses, thermal crosstalk, and quantization errors due to low-precision encoding. In this talk, we will present a comprehensive analysis of these imperfections using a bottom-up approach. We will explore how these imperfections interact with one another and how their impact can vary widely based on the SPAA tuned parameters, physical location of the affected optical components, and the nature and distribution of the imperfections. We will also introduce a suite of novel photonic-aware low-cost design automation techniques that can significantly improve the resilience of SPAAs in the presence of these imperfections. These techniques can be easily combined with existing bias control and mitigation techniques in SPAAs.

Biography: Sanmitra Banerjee is a Senior Design-for-X (DFX) Methodology Engineer at NVIDIA Corporation, Santa Clara, CA, and an Adjunct Faculty at Arizona State University. He received the B.Tech. degree from Indian Institute of Technology, Kharagpur, in 2018, and the M.S. and Ph.D. degrees from Duke University, Durham, NC, in 2021 and 2022, respectively. His research interests include machine learning based DFX techniques, and fault modeling and optimization of emerging AI accelerators under process variations and manufacturing defects.  

Date and Time: March 1st, 2024, 8:00 PM - 9:00 PM (U.S. Eastern Time)

Abstract: Artificial intelligence (AI) powered by neural networks has enabled applications in many fields (medicine, finance, autonomous vehicles). Digital implementations of neural networks are limited in speed and energy efficiency. Neuromorphic photonics aims to build processors that use light and photonic device physics to mimic neurons and synapses in the brain for distributed and parallel processing while offering sub-nanosecond latencies and extending the domain of AI and neuromorphic computing applications. We will discuss photonic neural networks enabled by CMOS-compatible silicon photonics. We will highlight applications that require low latency and high bandwidth, including wideband radio-frequency signal processing, fiber-optic communications, and nonlinear programming (solving optimization problems). We will briefly introduce a quantum photonic neural network that can learn to act as near-perfect components of quantum technologies and discuss the role of weak nonlinearities.

Biography: Bhavin J. Shastri is an Assistant Professor of Engineering Physics at Queen’s University and a Faculty Affiliate at Vector Institute. He received a Ph.D. degree in electrical engineering from McGill University in 2012 and was a Banting Postdoctoral Fellow at Princeton University. Dr. Shastri is the recipient of the 2022 SPIE Early Career Achievement Award and the 2020 IUPAP Young Scientist Prize in Optics "for his pioneering contributions to neuromorphic photonics.” He is a co-author of the book Neuromorphic Photonics, a term he coined with Prof. Prucnal. He is a Senior Member of Optica and IEEE.

Abstract: In this talk, I will overview our work on analog neural networks based on photonics and other controllable physical systems. In particular, I will discuss why neural networks may serve as an ideal computational model, with the potential to harness the computational power of analog stochastic physical systems in a robust and scalable fashion. I will utilize photonic neural networks as a practical example to demonstrate their robust operation in low-energy regimes, which are typically constrained by quantum noise. Our experimental results indicate that photonic hardware offers a better energy scaling law than electronic for large-scale linear operations. This advantage is particularly significant for the scalability of modern foundational AI models, such as Transformers. Finally, I will show how nonlinear photonic neural networks may also help to enhance computational sensing for a diversity of applications, ranging from autonomous system control to high-throughput biomedical assays.

Biography: Tianyu Wang is an Assistant Professor to the Department of Electrical and Computer Engineering at Boston University. He is interested in developing novel methods for imaging, sensing, and computing by leveraging emerging technologies from photonics and artificial intelligence.

Date and Time: January 5th, 2024, 8:00 pm - 9:00 pm (U.S. Eastern Time) 

Abstract: Abstract: Advanced machine learning models are currently impossible to run on edge devices such as smart sensors and unmanned aerial vehicles owing to constraints on power, processing, and memory. We introduce an approach to machine learning inference based on delocalized analog processing across networks. In this talk, I'll detail Netcast, which uses cloud-based “smart transceivers” to stream weight data to edge devices, enabling ultraefficient photonic inference. We demonstrate image recognition at ultralow optical energy of 40 attojoules per multiply (<1 photon per multiply) at 98.8% (93%) classification accuracy. We reproduce this performance in a Boston-area field trial over 86 kilometers of deployed optical fiber, wavelength multiplexed over 3 terahertz of optical bandwidth. Netcast allows milliwatt-class edge devices with minimal memory and processing to compute at teraFLOPS rates reserved for high-power (>100 watts) cloud computers.

Biography: Alex Sludds received his B.S, M.Eng and Ph.D in Electrical Engineering and Computer Science from MIT in 2018, 2019 and 2023 respectively. Alex was an NSF graduate research fellow and has published in leading journals and conferences including Science, Nature Photonics, Science Advances and Physical Review X. His research interests focus on how the dense integration of silicon electronics and photonics enable orders of magnitude advances in computation and communication. Alex works as a photonic architect at Lightmatter.

Date and Time: December 1st, 2023, 10:30 am - 11:30 am (U.S. Eastern Time)