Future nano age devices will face a dramatic increase of soft and hard errors, degradation and process variability. Addressing these problems with worst-case design techniques will lead to a suboptimal performance and will introduce a huge overhead and an unacceptable low yield. A solution to this problem is to adapt the system parameters at runtime to current external requirements on performance and reliability as well as on current device dependent parameters such as error rates and chip temperature. But due to noise and time-variant system properties, measured chip properties may not represent the real system state. For the same reasons, control parameters of thermal and load management may not always show the same physical effects. To avoid worst-case design further, it is our approach for a trustworthy operation to explicitly model the vagueness, ambiguities and uncertainties of sensor values and actor settings. The concept of a respective methodological framework for such a trust management is motivated and presented in this paper. Its main features are to model the uncertainties of sensor readings and actors explicitly by dynamic trust level attributes. These parameters are exploited by the internal control of the device operation, e.g. for load and thermal management. In this way, robustness and performance are increased without sacrificing the lifetime of the device.
|2010 International Conference on Dependable Systems and Networks Workshops (DSN-W)
|Veröffentlicht - 20.09.2010
|2010 International Conference on Dependable Systems and Networks Workshops - Chicago, USA / Vereinigte Staaten
Dauer: 28.06.2010 → 01.07.2010