OTERA-III: Online Test Strategies for Reliable Reconfigurable Architectures - From Reliability to Guaranteed System Performability: A Multi-Layer Approach

点击翻译按钮获取中文摘要

The OTERA Project has been funded since 2011 and is now applying for the third funding phase. In a fruitful cooperation, the two groups (CES and ITI) combined their expertise in runtime systems and reconfigurable architectures (CES), as well as efficient online test, diagnosis and fault-tolerance (ITI). This lead to 8 papers published at renowned peer-reviewed conferences and transactions. Funding Phase I and II have targeted permanent structural faults (stuck-at and transition delay fault models) and transient faults (soft errors in configuration bits, bit-flip/RAP fault model) in the reconfigurable fabric.In Funding Phase III the target architecture is a multi-tasking system with heterogeneous runtime reconfigurable resources shared among multiple tasks. The reconfigurable fabric is fine-grained and implemented on SRAM-based FPGAs. We target to improve dependability by increasing system lifetime and availability (i.e. via repair, graceful degradation, or controlled stress distribution), raising system and task reliability, and guaranteeing system performability.The architecture of this project allows to dynamically adapt the system at multiple layers. Our project spans five layers (from highest to lowest): System, Task, Accelerated Functions, Accelerators (placed into containers), and Logic. At each layer, dependability actions can be applied with impact on other layers, which spans a vast search space to achieve the dependability requirements of tasks. Thus, the OTERA project offers a platform for exploration of different multi-layer dependability strategies.The multi-layer dependability optimization aims to (i) guarantee a specified level of system performability over the system mission time (i.e. reliable system operation at a desired level of performance), (ii) satisfy the dependability of concurrently executing tasks under runtime-changing environment, functional and non-functional conditions, (iii) provide self-adaption of the system in a multi-dimensional parameter space (controllable parameters, e.g. degree of redundancy, voltage level etc.).