Universal analog computation on programmable nanophotonic integrated circuits
Abstract/Contents
- Abstract
- Universal programmable nanophotonic meshes, or networks of Mach-Zehnder interferometers, are an exciting new avenue to arbitrarily shape light beams for energy-efficient chip-scale linear optical matrix multiplication. Applications include sensing, imaging, LIDAR, quantum computing, cryptography, and machine learning. Such devices can be scalably manufactured commercially in semiconductor foundries but can suffer from manufacturing and other systematic errors. In this thesis, I discuss both theoretical and experimental contributions for programmable optical meshes to address these concerns. First, I introduce a new class of binary tree meshes that are more error tolerant than currently proposed architectures. To prove this error tolerance formally, I explain my architecture-dependent error sensitivity theory relating individual component errors to overall system performance. I furthermore discuss several calibration and configuration algorithms aimed at reducing these errors, including self-configuration and "parallel nullification" that minimizes programming time of any feedforward photonic mesh network. Next, I present my design and implementation of a fully packaged 6 x 6 programmable "triangular mesh" outfitted with an experimental optical rig setup in our lab and explore its use as a matrix multiply accelerator for machine learning. As a key application, I demonstrate in situ backpropagation training (the most popular and widely used gradient-based training algorithm for machine learning) directly through optical measurement for the first time on our chip, which agrees well with digital simulations of the training process. Finally, I apply the same chip to study new photonic cryptocurrency and blockchain applications by exploring a paradigm shift: implementing discrete-valued matrix multiplication in a photonic mesh for robust and energy-efficient digitally verifiable computation. In this vein, I propose LightHash, a hash function that incorporates photonic computation into the Bitcoin protocol to secure cryptocurrency transactions. I analyze component error scaling to overall LightHash error rates with size and bit depth of our photonic matrix multiplier and demonstrate new error correction protocols to minimize such error. These experimental achievements coupled with my new theoretical framework could have significant and lasting implications on photonic integrated circuits, programmable optics, major tech sectors for artificial intelligence and cryptography, and beyond.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2022; ©2022 |
| Publication date | 2022; 2022 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Pai, Sunil Kochikar |
|---|---|
| Degree supervisor | Solgaard, Olav |
| Thesis advisor | Solgaard, Olav |
| Thesis advisor | Fan, Shanhui, 1972- |
| Thesis advisor | Miller, D. A. B |
| Degree committee member | Fan, Shanhui, 1972- |
| Degree committee member | Miller, D. A. B |
| Associated with | Stanford University, Department of Electrical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Sunil Pai. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2022. |
| Location | https://purl.stanford.edu/sr136rn5978 |
Access conditions
- Copyright
- © 2022 by Sunil Kochikar Pai
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...