FuSE - a hardware accelerated HDL fault injection tool

The ongoing miniaturization of digital circuits makes them more and more susceptible to faults which also complicates the design of fault tolerant systems. In this context fault injection plays an important role in the process of fault tolerance validation. As a result many fault injection tools have emerged during the last decade. However these tools only operate on specific domains and can therefore be referred to as hardware- or software-, simulation- or emulation based techniques. In this paper we present FuSE, a single fault injection tool which covers multiple domains as well as different fault injection purposes. FuSE has been designed for usage with the SEmulator^reg-an FPGA-based hardware accelerator. The created tool set has been fully automated for the fault injection process and only requires a VHDL description and a test bench of the circuit under test. FuSE can then perform fault injection experiments with a diagnostic resolution that is known from simulation-based approaches, but at a speed that even handles long running experiments with ease.

1. INTRODUCTION

The increasing complexity of digital circuits makes their development more and more challenging and tricky. Especially the verification process becomes a crucial task. Simulations provide full access to internal signals but they can only cover short periods of real time (few milliseconds). A prototype implemented in an FPGA allows to observe the device for longer periods, but yields an extremely low visibility of internal procedures, which aggravates the debugging in the case of a malfunction. Furthermore, the pure observation does not suffice yet because the ongoing miniaturization of semiconductor structures makes digital circuits more and more error prone. As a consequence fault tolerance mechanisms are required to implement reliable systems. To verify these mechanisms, waiting for the occurrence of an actual physical fault would be impractical. Instead, a dedicated manipulation of the circuit has to be applied in order to artificially produce the desired fault. In addition to the trade-off between performance and visibility, a fault injection strategy has to find a balance between intrusiveness and reproducibility.