Reconfigurable and designable structurally-nonlinear materials for photonic neuromorphic applications

Abstract

This project aims to develop a reconfigurable and designable structurally-nonlinear material platform based on liquid crystal-polymer composites (LC-PC) and metasurfaces, which can enable new classes of compact, low-power, photonic neuromorphic processors. The proposed platform exploits multiple scatterings of light, between a LC-PC scattering medium (and-or a metasurface) and a spatial light modulator (or a digital mirror device) used to encode the input data, to realize structural nonlinearity and nonlinear optical processing. The scattering process can be controlled by varying the external field applied to the LC-PC, which offers dynamic tuning of the scattering potential on a global scale; the metasurface, on the other hand, provides a fine spatial phase profile with subwavelength resolution. The metasurface can further promote non-Lambertian scattering, directing scattered light at any desired angles including exceptionally wide angles - achievements typically beyond the capabilities of conventional scattering media. The proposed LC-PC and metasurface, with complementary properties, will provide unprecedented capabilities and flexibilities to dynamically control the scattering strength, angular distribution, and anisotropy, enabling the control and reconfiguration of structural nonlinearity. This system will be used in several learning tasks, such as image classification, to demonstrate and benchmark its learning performance. The reconfigurability provides a balance between the discrimination and the generalization capabilities of the system for optimizing computing performances. The reconfigurability further enables photonic ensemble learning, a novel approach combining the results of a diverse ensemble of photonic models with distinct structural nonlinearities to achieve higher inference accuracies. Our study will provide fundamental insights into tunable structurally-nonlinear material systems and their applications in neuromorphic computing, promising a new learning approach with ultra-low power consumption.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 05, 2025
Source ID
FA23862414054

Entities

People

  • Zhiwen Liu

Organizations

  • Air Force Office of Scientific Research
  • Pennsylvania State University
  • United States Air Force

Tags

Fields of Study

  • Physics

Readers

  • Parallel and Distributed Computing.
  • Quantum Dot Semiconductor Device Photonics and Graphene Optoelectronic Materials and THz Physics.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks