Study of novel enabling technologies for optoelectronic neurocomputing

This dissertation deals with the study of novel enabling technologies for neurocomputing with optoelectronic hardware. Optoelectronic neurocomputing is attractive because it combines the best of two worlds, i.e., the inherent parallelism and massive interconnectivity of optics, and the versatile and nonlinear processing ability of electronics. However, to realize large-scale optoelectronic neurocomputers for real-world applications, dedicated neural hardware that is more inherently suited for neural computation and is more expandable toward large-scale integration is needed.;The main thrust of this dissertation concerns studying the dynamics of a new class of optical memory materials, electron trapping materials (ETMs), for optoelectronic neurocomputing applications. The novel approach we have taken is to use ETMs under simultaneous illumination by both the blue and IR light. Two theoretical models governing ETM dynamics with varied levels of sophistication are developed. Experimental study confirms the validity of these models and thus pave the way for applying ETMs to optoelectronic neurocomputing. Based on these results, it is shown that not only the large linear dynamic range of ETMs can be utilized in implementing the synaptic weight matrix of an optoelectronic neural network, but also the unique dynamics of ETMs are well suited for building large-scale optoelectronic pulsating neural networks. Schemes of employing ETMs to produce dense arrays of optoelectronic pulsating neurons and dense arrays of synapto-dendritic responses are studied and devised. A key proof-of-concept experiment, that involves an ETM with two nonlinear switching elements (a programmable unijunction transistor (PUT) and a single pulse generator, emulating the switching dynamics of the excitable membrane) in an optoelectronic feedback loop, is demonstrated to produce an optically controlled pulsating neuron. To further enhance the capability of ETMs, issues of writing information onto ETMs with electron beam (e-beam) are studied, and a general design framework for e-beam addressed ETM devices is developed. Based on this framework, two e-beam addressed ETM devices, namely, the field emitter micro-cathode ETM spatial light modulator and the ETM-based image intensifier, are presented including their structures, design considerations and expected performances. With their enhanced performance and capability, these devices form the essential building block of an ETM-based large-scale optoelectronic pulsating neural network.;We also studied neural computing architectures based on optoelectronic neural chips. Because of their integrated nature, neural chip architectures are free from problem associated with neural systems based on discrete components, and are also more expandable toward large-scale integration. Results of a feasibility study of an optoelectronic neural chip based on waveguide grating-coupled emitter/receiver arrays (WGCERAs) are presented. The structure and design of the WGCERAs draw heavily from advances in the technologies of integrated optics, optoelectronic integrated circuits (OEICs), and micro-fabrication techniques. To achieve uniform emitting/receiving characteristics that are required for neural computation, the use of grating coupler with tapered grating height profile is proposed and the required profile expression is derived. Methods of fabricating tapered grating and choices of waveguide materials that facilitate optoelectronic integration are discussed and suggested in light of today's advanced microfabrication technologies. Finally, compatibility issues of WGCERAs with potential spatial light modulators (SLMs) in forming a bidirectional neural network are discussed. (Abstract shortened by UMI.).