Doctor of Philosophy (PhD)
Electrical and Computer Engineering
The shrinking processor feature size, lower threshold voltage and increasing on-chip transistor density make current processors highly vulnerable to soft errors. Architectural Vulnerability Factor (AVF) reflects the probability that a raw soft error eventually causes a visible error in the program output, indicating the processor’s susceptibility to soft errors at architectural level. The awareness of the AVF, both at the early design stage and during program runtime, is greatly useful for designing reliable processors. However, measuring the AVF is extremely costly, resulting in large overheads in hardware, computation, and power. The situation is further exacerbated in a multi-threaded processor environment where resource contention and data sharing exist among different threads. Consequently, predicting the AVF from other easily-measured metrics becomes extraordinarily attractive to computer designers. We propose a series of AVF modeling and prediction works via using advanced statistical techniques. First, we utilize the Boosted Regression Trees (BRT) scheme to dynamically predict the AVF during program execution from a variety of performance metrics. This correlation is generalized to be across different workloads, program phases, and processor configurations on a single-threaded superscalar processor. Second, the AVF prediction is extended to multi-threaded processors where the inter-thread resource contention shows significant and non-uniform impacts on different programs; we propose a two-level predictive mechanism using BRT as building blocks to characterize the contention behavior. Finally, we employ a rule search strategy named Patient Rule Induction Method (PRIM) to explore a large processor design space at the early design stage. We are capable of generating selective rules on important configuration parameters. These rules quantify the design space subregion yielding lowest values of the response, thereby providing useful guidelines for designing reliable processors while achieving high performance.
Document Availability at the Time of Submission
Secure the entire work for patent and/or proprietary purposes for a period of one year. Student has submitted appropriate documentation which states: During this period the copyright owner also agrees not to exercise her/his ownership rights, including public use in works, without prior authorization from LSU. At the end of the one year period, either we or LSU may request an automatic extension for one additional year. At the end of the one year secure period (or its extension, if such is requested), the work will be released for access worldwide.
Duan, Lide, "Analyzing and Predicting Processor Vulnerability to Soft Errors Using Statistical Techniques" (2011). LSU Doctoral Dissertations. 2033.