A Genetic Testing Framework for Self-Testing of Microprocessor Cores

摘要

As the use of general and special processors as embedded cores in SoC designs increases, developing high quality test programs becomes important. Our software-based self-testing approach is based on using macros that include a sequence of instructions to exercise all operations performed by each component of the processor being tested. Generation of the macros require instruction set and data transfer path information which is extracted from processor's RTL and ISA descriptions. We employ Genetic Algorithms (GAs) to choose a suitable sequence of macros for the test program and assigning good data values to the composing instructions of macros. In the former case, we utilize high-level fault coverage measurement metrics to alleviate the time overhead of simulations needed in GA for evaluation of the solutions. The method discussed here was applied to a benchmark processor to demonstrate steps involved in our approach.