Run-time Soft Error Injection and Testing of a Microprocessor using FPGAs

Nowadays, soft errors in logic circuits are becoming increasingly important. This is especially true in the ever shrinking nanometer technologies, and aerospace applications where soft errors are more prevalent. Previous soft error injection simulation methods using FPGAs have usually been limited to small test circuits such as ISCAS89, or parts of a larger circuit. Other software approaches have been proposed that simulate soft errors by adding extra instructions that the processor must execute. In this paper, we introduce our FPGA based transient fault injection system that can handle all memory elements of an entire microprocessor (MIPS32) while connected and running within a complete system. This allows us to perform fault injection and analysis when the system is live and actually running arbitrary applications on the processor. We test our implementation, and the effects that soft errors have on a software filter used in aviation for probabilistic sensor data fusion, as this algorithm would be run at higher altitudes where soft errors are more frequent. Our experimental results not only show that our method is extremely fast and versatile, but that it also allows us to test how software applications perform under a wide range of fault conditions.