State Recovery for Coarse-Grain TMR Designs in FPGAs Using Partial Reconfiguration

The operation of field-programmable gate arrays (FPGAs) in harsh environments like space entails the need for suitable fault-tolerance techniques of which Triple-Modular Redundancy (TMR) is most commonly deployed. While TMR is undoubtedly effective in masking faults, state recovery remains a problematic issue: Fine-grain TMR allows safe recovery, but incurs prohibitive area and performance penalties. In contrast, coarse-grain TMR has little overhead, but cannot safely provide recovery without roll-back or reset. We use the dynamic reconfiguration feature of modern FPGAs to augment an initially coarse-grain TMR with the ability of temporarily loading a fine-grain TMR design for forward-state-recovery. Therefore, we can seamlessly resume correct (fully redundant) operation in case of data-as well as configuration faults that occurred in the FPGA. As a proof of concept, the paper presents a showcase design and discusses distinctive properties of this new approach.