No abstract

Abstract-The impact of SRAM-based FPGAs is constantly growing in aerospace industry despite the fact that their volatile configuration memory is highly susceptible to radiation effects. Therefore, strong fault-handling mechanisms have to be developed in order to protect the design and make it capable of fighting against both soft and permanent errors. In this paper, a fully reconfigurable medium-grained triple modular redundancy (TMR) architecture which forms part of a runtime adaptive onboard processor (OBP) is presented. Fault mitigation is extended to the voting mechanism by applying our reconfiguration methodology not only to domain replicas but also to the voter itself. The proposed approach takes advantage of adaptive configuration placement and modular property of the OBP, thus allowing on-line creation of different medium-grained TMRs and selection of their granularity level. Consequently, we are able to narrow down the fault-affected area thus making the error recovery process faster and less power consuming. The conventional hardware based voting is supported by the ICAPbased one in order to additionally strengthen the reconfigurable intermediate voting. In addition, the implementation methodology ensures using only one memory footprint for all voters and their voting adaptations thus saving storing resources in expensive rad-hard memories.

Reliability is one of the key issues in space applications. Although highly flexible and generally less expensive than predominantly used ASICs, SRAM-based FPGAs are very susceptible to radiation effects. Hence, various fault tolerant techniques have to be applied in order to handle faults and protect the design. This paper presents a reconfigurable onboard processor capable of run-time adaptation to harsh environmental conditions and different functional demands. Run-time reconfigurability is achieved applying two different reconfiguration methodologies. We propose a novel selfreconfigurable architecture able to on demand duplicate or triplicate part of the design in order to form DMR and TMR structures. Moreover, we introduce two different approaches for voting the correct output. The first one is a traditional voter that adapts to different DMR/TMR domain positions whereas the second implies comparing the captured flip-flop values directly from the configuration memory read through ICAP. The comparison is done periodically by an embedded processor thus completely excluding the voting mechanism in hardware. The proposed run-time reconfiguration methodology provides savings in terms of device utilization, reconfiguration time, power consumption and significant reductions in the amount of radhard memory used by partial configurations.

No abstract