Abstract
This work focuses on testing methodologies to analyze the radiation sensitivity of FPGA-based systems. Due to their flexibility, the reliability analysis on these components is a challenging task as the radiation sensitivity is entirely conditioned by the implemented system. Indeed, it depends on the one hand on the intrinsic sensitivity of the component (to both TID and SEEs) and, on the other hand, on the way the different induced perturbations can impact the operation of the implemented system. State-of-the-art methodologies have shown a number of limitations in bridging the intrinsic sensitivity of the FPGA and the one of the implemented systems. The objective of this thesis is to improve radiation testing methodologies to overcome these limitations.Concerning TID effects, a new testing methodology is proposed. Its main contribution is to extend the evaluation of parametric degradations to all logical and routing resources of the component. For this purpose, specific benchmarking structures have been developed to measure the propagation delay deviation of each type of logical and routing resource. A new technique to measure the propagation delay in real time and with limited external instrumentation is also proposed. X-ray radiation tests have been performed on three FPGA families to highlight the benefits of this methodology.As for SEE, the proposed testing methodology lies between the two traditional accelerated particle beam testing approaches (primitive level testing and final application testing) by proposing a sensitivity evaluation at a higher level of granularity. The basic idea is to instantiate a set of dedicated benchmarking structures, simple enough to provide a good testability (low error masking, traversable state spaces) while sufficiently complex to provide a good representativity of the circuits effectively implemented on FPGAs. The benchmarks selected in this study are based on arithmetic operations. By using different implementations of the same arithmetic functions with a large diversity in the circuit parameters, and in the use of resources, the radiation tests fulfill a multifaceted purpose. First, the test results provide extensive information to identify and understand the different failure mechanisms and their predominance; second, it allows to qualitatively evaluate the impact of different types of resources on the global system sensitivity and to quantitatively compare the sensitivity of different implementations of the same logic function and the effectiveness of mitigation solutions. Finally, it provides a set of guidelines for designers to improve the reliability of FPGA-based systems. Several neutron and proton beam tests have been performed to demonstrate the advantages of this approach.The main limitation of radiation testing lies with the difficulty to extrapolate the results of tests performed with a given implemented circuit to estimate the sensitivity of any other circuit. To address these limitations, a new software-based approach has been developed to estimate the susceptibility of circuits implemented on SRAM based FPGA to configuration memory corruptions. This analytical approach uses the physical netlist of the circuit and explores the different nodes and logical resources that compose it to extract all the configuration bits that are critical for the system operation. The main contribution of this approach is to take into account the workload of the circuit, extracted from logic simulation, to analyze the propagation of errors and thus filter among the set of potentially critical configuration bits, those that actually modify the output signals of the system. The efficiency of this approach is validated through fault injection and proton experiment.