In the last ten years, two main factors have fueled the steady growth in sales of mobile computing and communication devices: a) the reduction of the footprint of the devices themselves, such as cellular handsets and small computers; and b) the success in developing low-power hardware which allows the devices to operate autonomously for hours or even days. In this review, I show that the first generation of mobile devices was DSP centric – that is, its architecture was based in fast processing of digitized signals using low- power, yet numerically powerful DSPs. However, the next generation of mobile devices will be built around DSPs and low power microprocessor cores for general processing applications. Mobile devices will become data-centric. The main challenge for designers of such hybrid architectures is to increase the computational performance of the computing unit, while keeping power constant, or even reducing it. This report shows that low-power mobile hardware architectures design goes hand in hand with advances in compiling techniques. We look at the synergy between hardware and software, and show that a good balance between both can lead to innovative lowpower processor architectures.
