Multi-Purpose Parallel Process for Biosequence Analysis

本项目旨在建立一个生物序列比较专用计算机系统。该结构适用于各种序列比较方法，包括Smith- Waterman算法、BLAST、剖面搜索和字典方法。此外，该体系结构正在与使用隐马尔可夫模型等技术对序列进行统计分析的软件集成。该体系结构的核心是SIMD处理元素的线性阵列，每个元素都有一个小的本地内存。一个芯片可以包含64个这样的元素，所以一个有20个芯片的电路板可以容纳1280个处理器。处理元素通过合并单个周期加最小指令和用于快速记录指令结果的数据路径而调整为序列比较。在该体系结构的设计中考虑了软件、I/O和算法；由此产生的平衡实现应该以低成本获得高性能。

This project is building an application-specific computer system for biosequence comparison. The architecture is suitable for a wide range of sequence comparison methods, including the Smith- Waterman algorithm, BLAST, profile searches and dictionary methods. In addition, the architecture is being integrated with software for the statistical analysis of sequences using techniques such as hidden Markov models. The core of the architecture is a linear array of SIMD processing elements, each with a small local memory. A single chip can contain 64 of these elements, so a board with 20 chips will hold 1280 processors. The processing elements have been tuned to sequence comparison by incorporating a single cycle add-and-min instruction and a data-path for quickly recording the results of the instruction. Software, I/O, and algorithms have been considered in the design of this architecture; the resulting balanced implementation should result in high performance at low cost.