Continued advancements in semiconductor technology make it possible to integrate hundreds of processing cores into one silicon die. However, as the transistor size continues to shrink, increasing manufacturing defects have resulted in severe yield loss and made chip products? profit to drop substantially. Process variations also cause the performance and power characteristics differ significantly from chip to chip. In addition, this non-determinism greatly exacerbates the complexity of developing, validating, and maintaining software built upon these chips, especially for mission-critical real-time embedded applications.
This project explores the hypothesis that architectural virtualization is an effective strategy and will be the norm to address extensive process variations and manufacturing defect problems on many-core platforms. This collaborative research effort seeks to develop effective methods and techniques to virtualize hardware resources on many-core platform and thus isolate the underlying hardware non-determinisms without changing the operating system and application software.
This project seeks to improve the robustness and reliability of software and systems developed on many-core platforms. Techniques developed in the research may also help to address the component upgrade and obsolescence issue, which has been challenging military and avionics industry for decades. The collaborations between PIs and their international collaborators provide a unique opportunity to build an international research and educational collaboration structure, to share research expertise, to enhance the educational opportunity, and to promote cultural exchange. In addition, this project provides opportunity to nurture, encourage and attract students from under-represented groups to engage in this exciting research.
