Accurate and Scalable RTL-level Fault Injection Simulation for Industrial and Automotive Standard | Consortium for Embedded Systems | SIU

Abstract: Traditional fault injection-based gate-level simulation is non-scalable at the SoC level but industrial and automotive standards require gate-level fault diagnostic coverage. Existing fault simulators that accelerate gate-level fault simulation by considering RTL modules do not meet fault diagnostic coverage standards for industrial and automotive applications. A scalable metric-based fault simulator is proposed. A framework will be developed to evaluate whether the proposed metric-based module-level simulation achieves accurate fault diagnostic coverage. Experimental evidence will be provided. Classes of circuits that require gate-level simulation in order to achieve accurate fault diagnostic coverage will be investigated.

Problem: Fault injection-based techniques depend on gate-level simulation but are not scalable due to the size of existing systems on a chip (SoC). At the same time, industrial and automotive standards require that the fault diagnostic coverage be reported at the gate-level. Existing fault simulators that accelerate gate-level fault simulation by considering RTL modules do not meet fault diagnostic coverage standards for industrial and automotive applications. This effort proposes a scalable RTL-level simulator that provides accurate fault diagnostic coverage. Each RTL module may have drastically different netlist representations, and the test set may have drastically different fault diagnostic coverage for alternative netlist representations. The RTL-level simulator should consider an appropriately defined abstract quantity (metric) for each RTL module to derive the fault diagnostic coverage much faster. Appropriate statistical evaluation must be performed to ensure that the approach performs accurately for any test set at any RTL instance. A separate but related problem is to investigate whether there exist RTL-module implementations (netlists) that adopt effective metrics for RTL-level simulation and subsequent conversion to fault diagnostic coverage.

Rationale / Approach: In order for the fault simulator to be scalable it should avoid explicit fault simulation. It is proposed to consider an abstract quantity (metric) for each RTL module which provides sufficient information for the underlying netlist so that accurate fault diagnostic coverage estimation is obtained. The values obtained after all patterns are simulated are then translated to gate-level fault diagnostic coverage. Existing RTL-level simulators are either non scalable or consider metrics that do not provide fault diagnostic coverage. The proposal carefully reviews existing approaches and analyzes their weaknesses. The proposed solution relies on a metric quantity that abstracts each RTL netlist while considering the test set. Thorough statistical evaluation will be performed to ensure that the estimated fault diagnostic coverage is accurate and meets industrial and automotive standards. The framework will then be applied to different netlists in order to determine whether certain netlist properties ensure more accurate fault diagnostic coverage. The approach will be evaluated on existing benchmarks available to the academia as well as benchmarks provided to the investigators by the sponsor member company.

Novelty:
(a) A new metric quantity for the problem under investigation.
(b) The statistical evaluation framework for the proposed metric

Potential Member Company Benefits:
1. A fast and accurate grading tool to estimate the quality of a given test set in order to assess whether automotive and industrial standards have been met.
2. Thorough experimental evaluation on industrial benchmarks provided by the sponsor company

Deliverables for the proposed year:
1. Software tool to perform RTL-fault grading.
2. Detailed experimental evidence on benchmarks provided by the sponsor company.
3. Determine whether there exist netlist properties that support RTL metric-based accurate simulation.
4. Detailed report and documentation