Correct Embedded Computing Futures

Abstract

Embedded computing systems have always been significantly more diverse than their mainstream microprocessor counterparts, but they have also been relatively more simple to design and validate. Given the current global fascination with ubiquitous information and communication services in a highly mobile world, simplicity is rapidly disappearing. Advanced perception systems such as speech and visual feature and gesture recognizers, 3G and 4G cellular telephony algorithms can not currently be done in real time on performance microprocessors let alone at a power budget commensurate with mobile embedded devices. This talk will describe an architectural approach to embedded systems which outperforms performance microprocessors while consuming less power than current embedded systems for the above applications. This approach will be used as a way to highlight new issues of correctness in embedded systems. Namely correctness applies to functional, energy consumption, and real time processing constraints. Given that these issues become even more critical as technology scales makes life even more complex. In order to deal with these hard problems, system architects are creating new system models where the application, operating system, and hardware interact in new ways to collaboratively manage the computational rates and energy consumption. This new system model generates a new set of validation problems that will become critical roadblocks to progress in advanced embedded systems of the future. The talk will conclude with a description of validation challenge problems inspired by the new system model.