三值光学计算机乘法器及计算例程平台关键技术研究

Using the thousand-bit Ternary Optical logic computing system developed by our own research group as experimental platform, the project is focused on the study of the principles and key techniques about new optical multiplier. Based on the research results, the principles and key techniques about how to build computing platform for Ternary Optical Computer (TOC) will be studied. All the studies will much benefit to solve the problems in multiplication and other numerical computing applications with the characteristics and advantages of the optical computing system. In the process of the study of the project, the parallelism of Modified-Signed Digital (MSD) number system and optical computing is used. The principles of the optical multiplier that suits to process huge-number-of-bit data together with its architecture and implementation methods will be studied. Using the obtained results, the model of the optical multiplier will be built and the implementation methods of the key components for the multiplier with low energy consumption and huge-number-of-bit data processing capability will be discussed. After that, fully considering the hardware resources of the TOC, the general data-bit allocation strategies and techniques, auto-reconfiguration of optical processor technology, joint working mechanisms between different hardware resources will be studied. Further, the interfaces and protocol between the computing applications and the TOC will be designed. Meanwhile, the methods, strategies and algorithm on how to use the resources on TOC through the interfaces will be developed. When the project will have been finished, the complex application process used on electronic computer will be much simplified, the applicability of the TOC will much increase and energy consumption will be much reduced. This will provide a new thought for the high performance computing applications.

本项目以自主设计开发的三值光学计算机系统为实验平台，研究避免进位的新型光学乘法器理论及关键技术，并以此为基础，探索建构公共计算例程平台的基本理论及关键技术，以期充分利用光学计算的特点与优势解决乘法运算及其他数值计算领域中的基本科学问题。项目拟将MSD数的并行性与光学计算的并行性相结合，研究适合于众多位数数据处理的光学乘法器理论，探讨其架构设计原理及其实现方法，建立相应的数学模型，探索低能耗、高数据宽度光学乘法器关键部件的设计方法；在此基础上，充分考虑光学计算机的硬件资源、研究适用于不同计算应用的公共数据位分配策略和技术、处理器自动重构技术及光学计算机硬件资源之间的协同工作机制，设计与第三方计算交互的通信接口与协议，探讨利用公共接口调用系统资源进行数据访问的科学方法、基本算法与应用策略。项目完成后将大大提升光学计算机的应用能力，显著降低能耗,为高性能计算开辟新的途径。

结合光学计算机数据为众多，但仅能在加法与逻辑运算领域应用，不能充分发挥三值光学计算机在高数据宽度应用场合优势，且没有应用平台使得用户应用光学计算机时需了解其实现细节，不利于光学计算机发展的现状，从充分提升与发挥三值光学计算机在乘法应用中的效率的角度出发，提出了避免进位的新型并行光学乘法理论，探讨了其关键实现技术，给出了基于对称MSD数的一步式加法的实现方案和基于二分法的MSD加法实现方案，提出并研究了部分积求和全并行实现机制和流水线实现机制，设计了基于对称MSD数的一步式加法和基于二分法的MSD加法实现光路、M变换实现光路，给出了全并行实现和基于流水线实现的光学乘法器实现方案及基本架构；设计实现了基于光学计算机的应用平台，以确保用户在无需了解光学计算机实验平台实现细节的情况下，高效的利用光学计算的特点与优势，完成基本的数值计算应用；同时以MSD（Modified-Sign-Digit）并行乘法、离散傅里叶变换和小波变换等典型应用为例，对其基于应用计算平台的实现细节进行了深入研究，给出了相应的资源分配方案、同步控制策略、数据回馈实现机制及专用数据区的设计机理等，并进行了大量实验、分析与验证，实验结果证明了基于光学计算机的应用平台正确可行，用户可以方便的应用该平台完成数值计算而无需了解其实现细节，有利于光学计算机的推广与发展，同时也为适合于光学计算特点的应用提供了一种新的解决思路。此外，通过大量的实验及对结果的对比分析发现，在光学计算机上实现DFT/FFT应用、小波变换应用所需要的时钟周期数远远少于电子计算机处理同等应用所需要的时钟周期数，而且随着问题规模增大，光学计算机的优势越明显，这使得光学计算机在需要进行DFT/FFT、小波变换等应用场景下具有十分广阔的应用前景。