CSR: Medium: Modeling and synthesis for application-specific systems-on-a-chip

点击翻译按钮获取中文摘要

Popular electronic devices such as e-book readers, automotive entertainment systems, digital TVs, and medical devices, are internally powered by a state-of-the-art computer chip. Such a chip may have hundreds of components, including many microprocessors, graphics processors, video processors, audio processors, memories, and more. Just a few decades ago, so many components would have required a table-sized box, but today they fit on a tiny fingernail-sized chip. Because such a chip costs tens of millions of dollars to build, companies that build these chips make them as configurable as possible, so the same chip that is used in a digital TV application may also be used in a medical device application. However, engineers have a hard time setting all the configurable features to get the chip's performance and power optimal for their specific application. The features are like the ingredients of a recipe; an imperfect balance can ruin the performance/power (or the meal). As a result, today's chips run much slower than they could, get hotter, and consume more battery than they could as well. This project's intellectual merit includes developing techniques to automatically tune a state-of-the-art chip's many configurable features to serve any one particular application well, thus substantially improving performance and power for any application implemented on such a chip. The project will develop a unified computation model that ideally supports such techniques, empowering engineers to make best use of state-of-the-art chips. The project's broader impacts not only include improving the performance and power of a wide variety of popular devices, but also includes a web-based tool for capturing the unified computation model. Such a tool will positively impact university education as well as practicing engineers, and can catalyze research.