Design And Implement Of Software-Implemented Fault Injection And Test System For Intelligent Processing Platform

Artificial intelligence chips are currently a necessary reference standard for measuring the level and strength of a country's scientific and technological development.In addition,there are a large number of high-energy particles and radiation in outer space.These high-energy particles and radiation are likely to cause a series of single event effects in special chips in the aerospace field,thereby affecting the regular operation of space equipment.Therefore,it is of great significance and value to study the single event effect and analyze the reliability of intelligent chips.Based on the above two main reasons,this paper studies the intelligent processing platform and software fault injection technology,and designs and implements a software fault injection and test system based on the KM22 intelligent processing platform.This paper focuses on the KM22 intelligent platform,and first studies the KM22 intelligent platform architecture and operating system.Then,based on the classic FARM fault injection model,a KM22-FARM fault model based on the KM22 intelligent platform is designed.The KM22-FARM fault model defines the KM22-F fault set,KM22-A active set,KM22-R readout set,and KM22-M measurement set according to the characteristics of the KM22 intelligent platform.The four t uple definition of KM22-F set,the workload design principle of KM22-A set,the reaction parameter definition of KM22-R set,the test index of KM22-M set are introduced in detail.After designing the KM22-FARM fault model,this paper introduces the software simulation fault injection system based on the KM22 intelligent platform(KM22-SWIFIS).The system is designed and implemented based on the overall structure of client+server and adopting TCP protocol.The server mainly includes the KM22-FARM model module,fault injection module,fault detection module,fault log module,and network communication module.Among them,the KM22-FARM model module explicitly explains how to inject a fault in the form of four tuples;the fault injection module describes in detail how to implement fault injection through the PTRACE system call provided by Linux.The client introduces the user management module,the UI interface module for visualizing the interface and the communication module for communicating with the server.Finally,this paper introduces how to evaluate the KM22 intelligent platform through KM22-SWIFIS.The test indicators include coverage test and sensitivity test.The coverage test is aimed at the injectable range of KM22-SWIFIS fault injection and proves the availability of KM22-SWIFIS;the sensitivity test is aimed at the sensitivity of different injection positions to fault injection,and the situation after the fault occurs.