15. Virtualization: System-Level Fault Simulation of SRAM Errors in Automotive Electronic Control Systems

In the coming age of self-driving cars, system-level testing of electronic control will become much more important to ensure dependable operation of automated functions. Modern VLSI devices are not always totally reliable. They can fail due to aging, electromagnetic excitation and many other reasons as described in the other chapters in this book. Therefore, dependable electronic systems must be tested against possible VLSI device failures. This may not be a common practice for meeting the ISO 26262 functional safety standard today, but deemed necessary for full-fledged self-driving cars in future. In this chapter, we demonstrate system-level simulation of SRAM errors and their impact on the design of electronic control. Automotive engine control is chosen as a test bed for this study. Model-based development techniques for automotive control systems are described first as the background and virtual electronic control units are introduced. A dependable SRAM architecture is proposed, and to test it in a practical use, a multilayer simulation modeling of an electromechanical system, its control software, and the SRAM design built-in microcontroller are discussed. To run a fault injection analysis in the SRAM chip at a large scale, a public cloud computing is used. The virtual computer machines in the cloud computing carry out the virtual engine control system simulation in which an instruction set simulator for the microcontroller executes the control software code step by step. The simulation system traces the outcome of the engine control system behavior upon a fault injection into SRAM to evaluate the dependable SRAM design. The large-scale fault analysis proposed here allows us to evaluate quantitatively the impact of the quality design of components on the entire system failure rate.