Development of Highly Parallel Optical Multiplex Computing Systems

在本计画中，我们研究了光多工运算系统--以光波长元件作为可多工资讯载体的平行运算系统--在高度平行运算架构中克服互连问题的潜力。1.建立了一套利用集值逻辑闸设计光多工运算系统的系统方法。在此基础上，利用分立光电器件设计并实现了一个三波长光并行加法器，验证了波长空间并行处理的可能性。2.研究了光复用计算对高度并行互连网络设计的影响。针对一类由BPC排列定义的互连网络，研究了互连拓扑在波长信道中的最优填充问题。结果表明，它们的平均互连密度可以减少1/伽马使用伽马波长组件的因素。设计并实现了一个8输入4波长的光学双调分选网络。3.作为集成光学多路复用计算的关键部件，采用介质多层薄膜滤波器制作了单片波长检测器，成功区分了635- 830 nm范围内的4个波长分量。通过实验研究，估计在目前的技术状态下，8波长复用是可能的。这一数字提供了将互连网络的面积复杂性平均降低1/64的潜力。

In this project, we have investigated the potential of optical multiplex computing systems-parallel computing systems that employ optical wavelength components as multiplexable information carries-for overcoming interconnection problems in highly parallel computing architectures.1.A systematic method for designing optical multiplex computing systems using set-valued logic gates was established. On the basis of the above result, an optical parallel adder using three wavelengths was designed and implemented with discrete opto-electronic components to demonstrate the possibility of wavelength-space parallel processing.2.The impact of optical multiplex computing on the design of highly parallel interconnection networks was investigated. Optimal packing of interconnection topology into wavelength channels was studied for a wide class of interconnection networks defined by BPC permutations. It was shown that their average interconnection density can be reduced by the factor of 1/gamma using gamma wavelength components. To demonstrate the impact of the proposed method, an 8-input 4-wavelength optical bitonic sorting network was designed and successfully implemented.3.As a key component for integrated optical multiplex computing, a one-chip wavelength detector employing dielectric multilayr thin-film filters was fabricated, demonstrating successful discrimination of four wavelength components within 635-830nm range. Through experimental study, it was estimated that 8-wavelength multiplexing is possible at the present state of technology. This number offers the potential for reducing area complexity of interconnection networks by the factor of 1/64 on an average.