Techniques for Tolerating Soft Errors in Brake-by-Wire Systems
This thesis addresses the problem of tolerating soft errors in brake-by-wire systems. Soft errors are caused by ionizing particles and manifest themselves as bit-flips in memory registers, latches, and flip-flops. As transistors are shrunk to sustain Moore's law, the frequency of such radiation-induced faults increases. The overall goal of the thesis is to design and evaluate cost-efficient techniques to tolerate soft errors in control systems.
To emulate the effects of soft errors, we inject single bit-flips into the registers and the main memory of a microcontroller running a prototype brake controller. We first evaluate a version of the brake controller in which hardware exceptions are the only means for detecting errors. These experiments show that although many of the injected errors are masked by the brake controller program, a non-negligible proportion of the errors caused the controller to produce erroneous outputs to the brake actuator.
We designed two software implemented mechanisms for detecting and recovering soft errors in the brake controller. These mechanisms protect the stack pointer and the brake controller's integrator state. Using a second error injection campaign, we evaluate the effectiveness of these mechanisms. Results from this campaign show that the software mechanisms reduced the proportion of errors causing critical controller significantly.