Despite significant efforts in testing and validation, CPUs are still riddled with bugs when they make it to the end users. Most of these bugs are impossible to fix post-production, and remain with us for as long as we use the systems built with these CPUs. To remedy this, we systematically study microprocessor errata published by Intel and AMD over many years to find commonalities between these bugs. The results is a structured database of errata, which we call RemembERR, that can be used to identify gaps in existing design testing and validation techniques.
RemembERR is published at the 55th IEEE/ACM International Symposium on Microarchitecture (MICRO’22). You can read the paper if you like to know more about the study and its conclusions, and use the open-source RemembERR database to study the errata further.