UNCONVENTIONAL PROCESSOR DESIGNS FOR
HIGH PERFORMANCE, LOW COST, LOW POWER CONSUMPTION
Sotirios G. Ziavras
This work introduces radically new architectural
concepts for uniprocessor system designs. They result in a uniprocessor design that
conforms to the data-driven (i.e., dataflow) computation paradigm. It is shown that usage
of this, namely D2-CPU (Data-Driven) processor,follows the
natural flow of programs, minimizes redundant (micro)operations, lowers the hardware cost,
and reduces the power consumption. We assume that programs are developed naturally using a
graphical or equivalent language that can explicitly show all data dependencies. Instead
of giving the CPU the privileged right of deciding what instructions to fetch in each
cycle (as is the case for CPUs with a program counter), instructions are entering the CPU
when they are ready to execute or when all their operand(s) are to be available within a
few clock cycles. This way, the application-knowledgeable algorithm, rather than the
application-ignorant CPU, is in control. The CPU is used just as a resource, the way it
should normally be. This approach results in outstanding performance and elimination of
large numbers of redundant operations that plague current processor designs. The latter,
conventional CPUs are characterized by numerous redundant operations, such as the first
memory cycle in instruction fetching that is part of any instruction cycle, and
instruction and data prefetchings for instructions that are not always needed. A
comparative analysis of our design with conventional designs proves that it is capable of
better performance and simpler programming.
Keywords -- Processor design, data-driven
(dataflow) model of parallel computation, comparative analysis, conventional processors,
concurrent operations, distributed CPU.
Return to the
"summary of recent research contributions" page