High quality functional coverage based trace signal selection for post-silicon validation

Post-silicon validation;Signal tracing;PageRank

Ma, Sai ; Vasudevan ; Shobha

关键词: Post-silicon validation;    Signal tracing;    PageRank;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/73097/Sai_Ma.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 drastic growth of design complexity and also shrinkage of the time-to-market window, it is always impossible to capture all bugs during the pre-silicon verification phase; therefore, a small numberof bugs escape after a chip is manufactured. Once a chip is manufactured, it is considered as a black box since internal observability is lost. To increase observability for post-silicon validation, an effective silicon debug technique is to use an on-chip trace buffer to monitor and capture the circuit response of certain selected signals during its post-silicon operation. Since this aforementioned debugging trace buffer introduces area overhead, the amount of signals selected to be stored on this trace buffer is very limited. Therefore, a major challenge in this field is to select a powerful subset of internal signals to reconstruct the majority of the remaining signal values. Existing methods use a greedy selection process to converge to a locally optimal selection; this kind of method suffers from severe diminishing restoration ratio effect as more trace signals are selected. In addition, all of the previous publications have been focused on increasing restorability; none of them has ever been able to interpret the trace signals as high level meaningful debugging information. In this thesis, we formulate the trace signal selection problem into a data-mining problem and propose two approaches using PageRank and HITS algorithm. Our experimental results demonstrate that our algorithm can effectively alleviate the diminishing restoration ratio effect. Furthermore, we propose a new metric to evaluate the quality of selected trace signals instead of restorability, which is the number of functionalities they cover when debugging; this is an angle previous publications have not addressed. According to our experimental results, the two new algorithms proposed in this thesis select better and more meaningful trace signals in terms of debugging.