Reliability challenges for electric vehicles: from devices to architecture and systems software

Today, modern high-end cars have close to 100 electronic control units (ECUs) that are used to implement a variety of applications ranging from safety-critical control to driver assistance and comfort-related functionalities. The total sum of these applications is several million lines of software code. The ECUs are connected to different sensors and actuators and communicate via a variety of communication buses like CAN, FlexRay and now also Ethernet. In the case of electric vehicles, both the amount and the importance of such electronics and software are even higher. Here, a number of hydraulic or pneumatic controls are replaced by corresponding software-implemented controllers in order to reduce the overall weight of the car and hence to improve its driving range. Until recently, most of the software and system design in the automotive domain -- as in many other domains -- relied on an always correctly functioning or a zero-defect hardware implementation platform. However, as the device geometries of integrated circuits continue to shrink, this assumption is increasingly not true. Incorporating large safety margins in the design process results in very pessimistic design and expensive processors. Further, the processors in cars -- in contrast to those in many consumer electronics devices like mobile phones -- are exposed to harsh environments, extreme temperature variations, and often, strong electromagnetic fields. Hence, their reliability is even more questionable and must be explicitly accounted for in all layers of design abstraction -- starting from circuit design to architecture design, to software design and runtime management and monitoring. In this paper we outline some of these issues, currently followed practices, and the challenges that lie ahead of us in the automotive and electric vehicles domain.