Engineering Creativity Award: Optical and Electronic Implementations of Neural Models

该研究项目旨在探索利用硅器件和光学技术实现大型并行通用神经网络架构的可能性。新架构的基本原理是利用先进成熟的硅信号处理技术这一事实，并将硅失效的光学器件（即互连问题）纳入其中。这些新架构使得利用现有技术构建具有1000个神经元的完全集成的可变网络成为可能。下一个重要的突破将是在单个设备上实现具有本地记忆的完整学习结构。描述了包含简单学习算法的大型、完全并行的体系结构，并指定了它们的初始设计。

This research project is directed to exploring the possibilities of implementing large parallel generic neural network architectures using both silicon devices and optical technology. The underlying principle of the new architectures is to take advantage of the fact that signal processing in silicon ins an advanced and mature technology and to incorporate optics where silicon fails, namely, the interconnectivity problem. These new architectures make possible the construction of fully integrated, alterable networks with 1000 neurons using existing technologies. The next important breakthrough will be to implement a complete learning structure with local memory on a single device. Large, fully parallel architectures that incorporate a simple learning algorithm are delineated and their initial design specified.