空天应用中自适应数据处理系统自主演化与容错机制研究

To meet the powerful adaptability, high reliability, miniaturization and advanced processing demand of the data processing system in space application, a novel adaptive data processing system based on field programmable gate array (FPGA) has been designed borrowing ideas from the evolvable hardware. Evolvable hardware can solve the weakness in adaptability and reliability of the conventional data processing system; but now it suffers from some shortcomings, such as long-evolving time, difficulties in the validation and evaluation of complex Circuits, and tardiness in online self-adaptiing and self-repairing. In this project, corresponding resolvent has been explored considering the system's framework, as well as the rapid self-evolving and fault-tolerant mechanisms of it. Firstly, a new system structure implemented on one FPGA has been proposed, which has more powerful adaptability, higher reliability, and more effective resource utilization comparing to the traditional one. Secondly, a rapid online self-evolving mechanism has been presented based on variable-size building blocks and the bit stream library of them, so as to increase the speed and efficiency of the on-line adaption. Finally, a fault-tolerant strategy of selective triple modular redundancy based on multi-objective optimization and evolvable hardware is presented, which can not only accelerate the speed of the online-repairing, but also be highly reliable and resource-effective. All research achievements can provide frame of reference for the design of new smart circuits and CMOS chips, as well as for the construction of self-organizing, self-adaptive and self-repairing hardware systems; and may offer new mechanisms to the research of reconfigurable computing and high performance computing.

为满足空天应用中数据处理系统对强自适应能力、高可靠性、小型化和先进处理能力的需求，借鉴演化硬件思想实现基于现场可编程门阵列的自适应数据处理系统。演化硬件可解决传统数据处理系统自适应能力不强，可靠性不高的问题，但目前仍存在演化时间长，复杂电路验证评估困难，在线自适应与自修复速度慢等问题。本项目从体系结构、自主演化与容错机制三个方面研究相应解决方法：（1）采用单芯片演化设计系统的体系结构，提高了自适应能力、可靠性和资源利用率；（2）提出了基于可变尺度演化构造块和构造块位流库的快速自主演化机制，提高了系统在线自适应速度和效率；（3）提出了基于演化硬件和多目标优化的选择性三模冗余系统容错设计方法，不仅提高了在线自修复速度，而且可以最少的资源代价，获取尽可能高的可靠性。研究成果可为新型智能电路和芯片设计及构建自组织、自适应、自修复的硬件系统提供理论依据，同时为可重构计算和高性能计算提供参考。

为满足空天应用中数据处理系统对强自适应能力、高可靠性、小型化和先进处理能力的需求，借鉴演化硬件思想实现基于现场可编程门阵列的自适应数据处理系统。演化硬件可解决传统数据处理系统自适应能力不强，可靠性不高的问题，但目前仍存在演化时间长，复杂电路验证评估困难，在线自适应与自修复速度慢等问题。本项目从体系结构、自主演化与容错机制三个方面研究相应解决方法：提出了基于演化硬件思想的单芯片自适应数据处理系统的体系结构，与传统结构相比，具有自适应能力强、可靠性高、资源节约的特点和故障情况下的自修复能力。提出了基于输入输出分解和构造块位流库的快速自主演化机制和基于快速混合重构的自适应数字系统架构，实现系统的快速演化。提出了基于EHW和多目标优化的高可靠、资源节约型STMR系统容错设计方法，以及配置刷新和演化修复相结合的强容错TMR系统设计方法，提高系统的可靠性和可用性。提出了基于工作输入和健康现态的自修复TMR系统状态恢复技术，可在一个时钟周期内实现修复后故障模块的状态与其他模块状态的同步，且不会造成系统功能的中断或延时。.以上研究成果不仅可以用于构建面向空天应用的自适应数据处理系统，实现其功能的快速自适应和故障情况下的自愈，而且可为新型智能电路和芯片设计及构建自组织、自适应、自修复的硬件系统提供理论依据，同时为可重构计算和高性能计算提供参考。