Metasurface Optical Neural Network for Object Recognition

We propose to investigate a novel approach to artificial neural computing using the principles of wave dynamics. Specifically, we will demonstrate an optical neural network (ONN) in the form of a smart glass that can recognize objects placed in front of it. The smart glass receives light scattered by the object and directly processes the optical information using its internal nanostructures. Our smart glass utilizes metasurfaces, which are two-dimensional arrays of meta-units that manipulate optical amplitude and phase with subwavelength resolution. By collectively operating millions of meta-units, our ONN enables efficient parallel computing with high expressive power, accomplishing tasks that typically require complex digital neural networks. Importantly, our ONN operates as a passive computer without the need for a power supply or digital processor, functioning at the speed of light. Moreover, this physics-based computing approach offers enhanced security beyond digital encryption. During Phase I Base of this project, we have theoretically demonstrated that our ONNs can recognize optically coherent and incoherent images, realizing classification of handwritten digits and verification of human facial photos with ~90% accuracy. This high accuracy is achieved with an extremely compact device (a stack of thin metasurfaces and a small planar array of photodetectors), with power consumption at least four orders of magnitude lower than digital ANNs and computing speed at least five orders of magnitude faster than ANNs.  In this context, we propose in Phase II to investigate multimodality ONNs that are tailored to “read” or “interpret” the spatial patterns of optical amplitude, phase, polarization, and spectrum for high-accuracy classification or verification of complex 3D objects, such as facial profiles, fingerprints, underwater mines, and watercrafts, which are of high relevance for Navy applications. We further generalize our definition of “optical” so that it is equivalent to “electromagnetic”; that is, the operational spectrum of the ONNs can range widely from the ultra-violet, to the mm-waves and microwaves, depending on the characteristics of the targets to be analyzed. Three specific objectives of Phase II include: Objective 1 (Phase II Base): Demonstration of a single-modality ONN for analyzing the phase information of electromagnetic waves reflected from 3D objects, targeting applications in secure personal identification based on facial profiles. Objective 2 (Phase II Base): Demonstration of a dual-modality ONN for classifying materials according to their characteristic spectral and polarizational properties, targeting applications in underwater mine identification. Objective 3 (Phase II Option): Demonstration of multimodality ONNs capable of analyzing a complete set of optical information (i.e., phase, amplitude, polarization, and spectrum), targeting applications in surveillance of watercrafts in the open sea.