Building Efficient and Reliable Emerging Technology Systems

Passcode: 942039

The semiconductor industry has been reaping the benefits of Moore’s law powered by Dennard’s voltage scaling for the past fifty years. However, with the end of Dennard scaling, silicon chip manufacturers are facing a widespread plateau in performance improvements. While the architecture community has focused its effort on exploring parallelism, such as with multi-core, many-core and accelerator-based systems, chip manufacturers have been forced to explore beyond-Moore technologies to improve performance while maintaining power density. However, the infancy of the manufacturing process of these new technologies, such as carbon-nanotube and tunneling-based field effect transistors, impedes their usage in commercial products.

This dissertation combines both architectural and device-level efforts to provide solutions across the computing stack that can overcome the reliability concerns of emerging technologies. This allows for beyond-Moore systems to compete with highly optimized silicon-based processors, thus, enabling faster commercialization of such systems.

This work analyzes yield issues in carbon-nanotube transistors, designs a carbon-nanotube based system using pass transistor logic that provides high energy efficiency, designs a fault-tolerant reconfigurable architecture that enhances yield and performance, and provides insights into designing function-level and device-level heterogeneous systems and schedulers.

Chair:  Professor Ronald Dreslinski Jr.