Challenges in Scaling Software-Based Self-Testing to Multithreaded Chip Multiprocessors

Functional software-based self-testing (SBST) has been recently studied by leading academic research groups and applied by major microprocessor manufacturers as a complement to other classic structural testing techniques for microprocessors and processor-based SoCs. Is the SBST paradigm scalable to testing multithreaded chip multiprocessors (CMP's) and effectively detect faults not only in the functional components but also in the thread-specific and core interoperability logic? We study the challenges in scaling existing software-based self-test capital (uniprocessor self- test programs and se Nest generation techniques) to real, multithreaded CMPs, like Sun's OpenSPARC T1 and T2. Since this type of CMPs is built around well studied microprocessor cores of mature architecture (like SPARC v9 in the OpenSPARC case), tailoring, enhancing and scheduling of existing uniprocessor self-test programs can be an effective methodology for software-based self-test of CMPs.