Bhavin J. Shastri · PaperMetrix

Reconfigurable analog photonic networks

2017

Spurred by advances in scalable photonic system integration, ideas in photonic information processing have recently experienced a resurgence. We review the latest results in reconfigurable analog photonic networks, neuromorphic photonics, and multivariate RF photonics. These …

Demonstration of Multivariate Photonics: Blind Dimensionality Reduction With Integrated Photonics

2019 · Journal of Lightwave Technology

Multi-antenna radio front-ends generate a multi-dimensional flood of information, most of which is partially redundant. Redundancy is eliminated by dimensionality reduction. This class of operations presents harsh fundamental tradeoffs to contemporary digital processing techniques. These …

Integrated Photonic Tensor Processing Unit for a Matrix Multiply: a Review

2022 · arXiv (Cornell University)

The explosion of artificial intelligence and machine-learning algorithms, connected to the exponential growth of the exchanged data, is driving a search for novel application-specific hardware accelerators. Among the many, the photonics field appears to be …

All-function Integrated Silicon Photonic Tensor Core (PTC) AI Accelerator

2023

Here we present our architecture for Silicon Photonic Tensor Core (PTC) capable of accelerating computational needs of neural networks and augmented/virtual reality applications. We present a novel fully-integrated PTC including chip-based lasers, modulators, and photodetectors.

Online training and pruning of photonic neural networks

2023

Photonic neural networks have unique weight-actuating mechanisms and manufacturing variations, resulting in a suboptimal performance by conventional offline training. By incorporating a power-pruning regularization term in the loss function, we demonstrate an online training method …

Silicon Photonics for Training Deep Neural Networks

2022

Analog photonic networks as deep learning hardware accelerators are trained on standard digital electronics. We propose an on-chip training of neural networks enabled by a silicon photonic architecture for parallel, efficient, and fast data operations. …

Microring Modulation-and-Weight Banks

2024

For photonic neural networks, we propose a novel microring bank with carrier-effect and thermal dual-tunability, which can 1) combine modulating and weighting for saved space, 2) improve tuning efficiency, and 3) inherit WDM-enabled scalability.