Microprocessors: The New LUT

Hardware design for high performance computing appears to be reaching its limits on several fronts. In the desktop microprocessor world, clock speeds seem to have reached their peak somewhere below 10 GHz. Power consumption has begun to approach, and in some cases surpass, 100 W. These issues, along with the problems of managing designs of approximately a billion transistors have caused one of the most profound changes in microprocessor architecture in decades. Every maker of desktop microprocessors has abandoned the traditional performance scaling approaches which have driven the industry for decades and have opted for so-called multicore designs. Today, it is widely accepted that no future commercial highperformance microprocessors will be built using a single CPU core. It is difficult to overestimate the significance of this change. A similar trend can has been seen in the ASIC and FPGA worlds. As both power and design complexity issues converge on designers, multiple microprocessor cores are increasingly found in integrated circuit designs. This trend in desktop microprocessors, ASIC and FPGA designs points toward a new reconfigurable architecure based on multiple microprocessor cores working togther in parallel. Such a system has many advantages over more traditional system design approaches. The design is inherently easier to design, uses a familar programming model, provides high levels of performance with modest power consumption. In this paper a single chip multiprocessor architecture will be explored in more detail. A programming model will be demonstrated using standard peocessor design tools and implementing a flexible, multiprocessor version of the Advanced Encryption Standard (AES). This implementation will serve as a motivation to explore the limits of exploitable parallelism in such a system, as well as issues of programmability and power consumption.