Ambient intelligence: gigascale dreams and nanoscale realities

Ambient intelligence (AmI) is a vision of a world in which people will be surrounded by networks of intelligent devices that are sensitive and adaptive to their needs. This concept implies a consumer-oriented industry driven by software from the top, and enabled (and constrained) by nano-scale physics at the atomic level. "More-Moore" will be needed to deliver the giga-ops computation and GHz communication capabilities required for stationary and wearable devices. But AmI also creates new challenges at the architectural and physical levels. To make AmI possible, AmI devices need two-orders-of-magnitude-lower power dissipation than today's microprocessors, at one-twentieth of the cost. These requirements conflict with the embedded programmability needed for personalization and adaptation to new services. We focus on emerging techniques for bridging this gap between software-centric systems and power-efficient platform architectures for AmI devices. A second challenge is to cope with the fact that nano-scale physical phenomena are in conflict with the traditional "nice and predictable" digital abstractions now used for complexity management. Techniques for coping with uncertainty and signal degradation, while achieving better-than-worst-case design, are discussed. Finally, one must realize that, to facilitate AmI, "more-than-Moore" technology is required, for example, for the design of autonomous wireless sensor networks. Correspondingly, attention is paid to novel combinations of technologies above and around CMOS, for the design of ultra-low-power, ultra-simple sensor motes for AmI.