System-level trace signal selection for post-silicon debug using linear programming

Trace signal;System-on-chip (SoC);Network-on-chip (NoC);post-silicon;Validation

Amrein, Matthew Curtis

关键词: Trace signal;    System-on-chip (SoC);    Network-on-chip (NoC);    post-silicon;    Validation;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/78532/AMREIN-THESIS-2015.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Due to the increasing complexity of modern digital designs using NoC (network-on-chip) communication, post-silicon validation has become and arduous taskthat consumes much of the development time of the product. The process offinding the root cause of bugs found in post-silicon validation has proven to bemuch more difficult than in pre-silicon because of the lack of the observabilityof all signals on chip. Trace buffers are a often used structure in post-silicondebug that stores the state of a selected signal into an on-chip buffer, wherethey can be offloaded for a debugger to observe. However, because of arealimitations for debug structures on chip and routing concerns, the signalsthat are selected to be traced must be a very small subset of all availablesignals. Traditionally, these trace signals were chosen manually by systemdesigners who determined what signals may be needed for debug once thedesign reaches post-silicon. However, because modern digital designs havebecome very complex with many concurrent processes, this method is nolonger reliable as designers can no longer fully understand the complexitiesthat are involved within these designs. Recent work has concentrated onautomating the selection of low level signals from a gate-level analysis. Inthis work, we present the first automated system-level, message-based traceselection where the guiding principle is functional coverage of system-levelmessages. We use a linear programming formulation to find multiple so-lutions that allow tracing of the high-frequency messages and then furtheranalyze these solutions using a message interval heuristic. This method pro-duces traces that allow a debugger to observe when behavior has deviatedfrom the correct path of execution and localize this incorrect behavior for fur-ther analysis. In addition, this method drastically reduces the time neededto select signals, as we automate a currently manual process.