Dynamic Partial Reconfiguration for Dependable Systems

Moore's law has served as goal and motivation for consumer electronics manufacturers in the last decades. The results in terms of processing power increase in the consumer electronics devices have been mainly achieved due to cost reduction and technology shrinking. However, reducing physical geometries mainly affects the electronic devices' dependability, making them more sensitive to soft-errors like Single Event Transient (SET) of Single Event Upset (SEU) and hard (permanent) faults, e.g. due to aging effects. Accordingly, safety critical systems often rely on the adoption of old technology nodes, even if they introduce longer design time w.r.t. consumer electronics. In fact, functional safety requirements are increasingly pushing industry in developing innovative methodologies to design high-dependable systems with the required diagnostic coverage. On the other hand commercial off-the-shelf (COTS) devices adoption began to be considered for safety-related systems due to real-time requirements, the need for the implementation of computationally hungry algorithms and lower design costs. In this field FPGA market share is constantly increased, thanks to their flexibility and low non-recurrent engineering costs, making them suitable for a set of safety critical applications with low production volumes. The works presented in this thesis tries to face new dependability issues in modern reconfigurable systems, exploiting their special features to take proper counteractions with low impacton performances, namely Dynamic Partial Reconfiguratio