Single Particle Fault Injection Signal Generation Method Using Gaussian Cloud Model

In the traditional single particle fault injection experiments (SPFIE), when the target configuration register for fault injection is selected, the state of the configuration register is inverted directly without considering the actual operating environment of the target design, such as the distribution of particles around the orbit, the crash frequency of particles on the target design and the electronic characteristics of the design itself. So, the soft failure rate obtained based on this method cannot properly reflect the actual situation of the target design in a specific space environment. To address this issue, this work presents a fault injection signal generation method considering both particle distribution in space and the circuit electronic characteristics. Because the real particle distribution data is less and the information contained is single, a novel data expansion method based on the Gaussian cloud model is proposed. This method makes full use of small sample information and the expanded data using this method maintain the same distribution as the original sample data with more information. We take the particle data collected by the detector mounted on Tiangong-1 as the simulation scenario to validate this novel expansion method. Simulation results demonstrate the feasibility and effectiveness of the proposed expansion method. The SPFIE using the proposed fault injection signal generation method improves the authenticity of the experiment results and the SRAM FPGA design soft failure rate estimated according to experiment result is closer to the real situation in the space environment.