This Grant is a continuation of N000141410505 - Optical and Optoelectonic Computing
Abstract
Approach:A new computing paradigm, sometimes called ?inference on graphical models?, is becoming ubiquitous. Graphical Models are broadly important in ?big data?, for social network analysis, for the ?Netflix challenge?, in trellis codes, turbo codes, and low-density parity check (LDPC) codes, as well as in image de-noising, de-blurring, de-convolution and segmentation. This proposal will pursue the optical implementation of a graphical inference co-processor that offers enhanced performance beyond any electronic solution. Optical communication, within the Graphical Model paradigm, offers huge Bisectional Bandwidth to allow an unprecedented number of nodes and edges. The confluence of two important factors encourages such an investigation: (1) Increasing problem complexity/dimensionality and the resulting ?big data? loads and (2) new nonlinear optical nano-devices.Objective:Optical computing was an active field of research 30 years ago. Due to the lack of significant progress, activities in optical computing were reduced considerably in recent years. However, the field is being re-energized due to several factors: (a) availability of enabling technologies resulting from advances in anophotonics and metamaterials and devices, (b) recent advances in integration of optics with CMOS electronics, and (c) emergence of novel computational paradigms that can be realized using chip-scale optoelectronics, such as inference on graphical models. The research vision of this proposal is to focus on fundamental aspects of an optical computing co-processor that includes CMOScompatible optics and employs new advances in optical metamaterial and device technologies. Our proposed co-processor will physically implement highly connected graphical models, which have emerged asan important paradigm in computer science. The optical co-processor will include logic at the nodes. The lines (called ?edges?) would be formed by 1D and 2D arrays of light emitter/detectors, together with optical projection operations to define the connectivity, such as holographicelements. A 3D distribution of complex refractive index (metamaterial 3D communication fabric) will be developed for managing connectivity wherein we exploit sub-resolution material degrees of freedom. Optical communication, within the Graphical Model paradigm, offers huge Bisectional Bandwidth to allow an unprecedented number of nodes and edges. The optical computing co-processor will be integrated with CMOS electronics to introduce input signals andcontrol parameters for a specific algorithm, and to read out the results.Naval Relevance:Navy faces a number of challenging problems in signal/image processing and computing that are stressing capabilities of current and projected silicon micro electronics. This MURI will investigate application of emerging photonic technologies to address these problems within the resource constraints.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 21, 2018
- Source ID
- N000141612237
Entities
People
- Mark Neifeld
Organizations
- Office of Naval Research
- United States Navy
- University of Arizona