Method for predicting vulnerability of instruction SDC (Silent Data Corruption) based on support vector regression

Abstract

The invention discloses a method for predicting vulnerability of an instruction SDC (Silent Data Corruption) based on support vector regression. The method comprises the following steps: 1, carrying out a fault injection experiment on a program set to acquire a sample data set; 2, extracting inherent features characterizing the nature of the instruction; 3, transversing the sample data set, and generating a data propagation dependence path for a target instruction; 4, transversing the sample data set, and calculating fault screening factors for the target instruction; 5, extracting dependencyfeatures of the instruction relevant with data propagation dependence; 6, training a vulnerability prediction model of the instruction SDC based on the support vector regression; and 7, extracting thefeatures of the target program instruction, and predicting the vulnerability of the instruction. The method has the advantages of high prediction accuracy, and low performance overhead; and moreover,the method also can be effectively applied to predicting the vulnerability of the instruction SDC after the program is affected by transient faults.

Description

technical field [0001] The invention belongs to the field of soft reinforcement and trusted software, and in particular relates to a command SDC vulnerability prediction method based on support vector regression. Background technique [0002] With the continuous development of semiconductor manufacturing technology, processors continue to reduce the size of integrated circuits and reduce the operating voltage. However, due to the reduction of the sensitivity of the device, the chip is more susceptible to the influence of space radiation while the performance of the computer is greatly improved. In the harsh radiation environment, single event effects caused by high-energy particle radiation or electromagnetic interference are the main reasons for the failure of computer systems. Single event upset (Single Event Upset, SEU) is the most important manifestation of single event effect. SEU refers to high-energy particles bombarding the device to flip its logic state, so that a...

AI Technical Summary

Problems solved by technology

In the prior art, there is no report on instruction SDC vulnerability prediction using support vector regression

Images