The theme at the Spring Processor Forum, held in San Jose last month, was cool—as in low power. Many products that were presented use different techniques to deliver more performance for fewer watts, such as a novel Forth-based multicore mesh from IntellaSys (see "Cores That Share Chores," June 8, 2006, p. 37).
Element CXI utilizes a mesh of computing elements that fit between the basic FPGA functionality and computing blocks of complete processors (see the figure). The Element CXI block operates asynchronously using a dataflow architecture.
PA Semi's PWRficient 64-bit PowerPCbased core uses almost an order of magnitude less power. A single core uses a little-over 3 W, and multicore solutions use up to 16 processors. The design employs a large number of clocking regions to minimize the number of active regions based upon the threads running on the system.
Handshake Solutions takes a different approach by eliminating clocks. ARM's ARM996HS processor uses Handshake Solutions' unique clockless IC design technology. On average, the clockless design uses 2.8 times less power than the comparable ARM968E-S. The ARM996HS has lower electromagnetic emissions as well.
Tarari highlighted other approaches to high-performance, low-power solutions with its Windows Media accelerator. The company has a number of solutions along these lines, including its RAX4 pattern matching processor, which offloads chores such as XML processing. Specialized processors like Tarari's T9000 can deliver one order to two orders of magnitude improvement, allowing a conventional host processor to concentrate on other chores like moving data around.