Ultimately, if you’re efficient in terms of energy, it will lead to efficiency in another allimportant area: cost. Needless to say, then, that energy efficiency will be a key driver for electronicsystems designers going forward. With designers under ever-more pressure to miniaturise, add new functionality, as well as cut costs, they find that greater integration is the only way to achieve all of these goals simultaneously. In other words, they’re turning to system- on-a-chip (SoC) design.
However, conventional SoC architecture based around a single high-speed processor with complementary on-chip peripherals is not the answer. This approach denies designers the flexibility they need to adapt devices for use in a range of products—something that’s now an economic necessity for many applications. It also compromises system performance and eats power. Just look at the size of the heatsinks on some PC processors. Contrast this with an approach that uses multiple low-power processors coupled with programmability.
This is fast becoming the norm rather than the exception in functions such as control, DSP, multimedia, and security. Each processor is optimised to its task. For example, cell phones often contain six or more processors with multimedia and signal processors optimised by the IP, silicon, or phone vendor for the individual platform’s power and cost requirements. Compared with single-processor design, power consumption can be reduced by a factor of five or more when using multiple processors, particularly in “always on” phone functions like audio and video decoding. At the other end of the spectrum, Cisco’s fastest router uses 188 energy-optimised processors. Thus, the highest performance no longer means the highest energy consumption.
Of course, while processor-centric design can be lower power, faster, and lower risk than hardwired designs, other design challenges still must be addressed in the next few years. Faster multiple- processor modeling, programming, and interconnect is needed. Also, to make tools easier to use requires better automation. But these challenges will be met, turning more designers to multiprocessor designs.