Evaluation of a Commercial Microprocessor

Evaluation of a Commercial Microprocessor by Robert Yung Doctor of Philosophy in Computer Science University of California, Berkeley Professor David A. Patterson, Chair After twenty-five years of development, the field of microprocessor design is still more art than science. A commercial microprocessor design is a fine balance between many conflicting goals, such as higher performance, lower design and manufacturing costs, shorter design time, and higher reliability. Effective planning and management of the entire design process are crucial to the success of a microprocessor, but are often overlooked and under-estimated. This dissertation is a first attempt to record and analyze the development of a commercial microprocessor—the UltraSPARCTM—from its inception to product shipment, from the point of view of an original architect. It shows how traditional computer design theory is extended with much more detailed cost analysis, how time to market and volume are key cost multipliers that must be factored in, and how the design process must be tailored toward rapid development of a performance and costeffective design. A dozen design principles and pitfalls were derived from this analysis. Two case studies further illustrated the design approaches and specific solutions to two major design issues in the UltraSPARC. Two major contributors to the success of the UltraSPARC project are an early integration effort and a robust verification environment. Full-chip integration exercised the design flow and verified the interoperability of tools used in the design. Extensive verification with real applications reduced time-to-volume of a product because critical mistakes were found and corrected before tape-out. This was made possible with a hardware emulation system, which exercised billions of verification cycles prior to tape-out, and allowed software and system co-development with the microprocessor.