Software-defined car takes shape
Kurt Sievers, General Manager of NXPs Automotive business unit, said he is seeing an automotive industry that is moving at different speeds of innovation: There is the traditional automotive industry with evolutionary innovation steps. “Both approaches will coexist – and we will serve both”, Sievers said. In any case, chips and software play an increasing role in cars, up to the extent that some experts like Glen De Vos, Chief Technology Officer of automotive tier one Delphi, talk about the “software-defined car” – analogous to other software-defined systems in the area of technology, like software-defined radio or software-defined networks. De Vos left no doubt that change is in need for the car of the future. “We have to change the architecture to accommodate new technologies and applications,” De Vos said. “What we see right now is only the beginning.”
He sketched an image of cars with “more sensors, more actuators and more (software) content – and all at lower cost. By the same token, De Vos demanded to “centralize driver assistance systems” – they would morph to virtual software entities, running on multi-domain controllers. “This would reduce complexity and weight”, De Vos announced. His company already has experience in this sector – among others, it will produce the zFAS central computer for future Audis. At the same time, centralizing the electronic architecture of the cars will eventually support new advanced functions. Features of such a vehicle include centralized computer platforms with distributed sensors and actuation. These building blocks will be connected across high-speed data paths like HDBaseT. To support the electronic functions of (more or less) autonomous vehicles, fault-tolerant power distribution will also be a feature of such cars.
Hadi Nahari, CEO and founder of startup company Cognomotiv, got into more details about the software-defined car. From his point of view, conventional cyber security won’t be enough to make such a car a safe and secure one. In addition, the industry needs ways to handle software faults. “We know that software always has faults. Experience shows that 1000 lines of code typically contain 5 to 50 software faults,” adding that the diverse computers in today’s cars typically run software that amounts to dozens of millions of lines of code. “The Ford F150 – a popular pickup truck in the USA – has about 150 million of LOC,” (for comparison: the entire software of the Facebook platform contains just about 16 millions LOC). For cars running at autonomy level 5, Nahari expects some 500 million lines of code. Making such amounts of software safe obviously is not a trivial task. Cognomotiv, while still in stealth mode, might offer tools and processes to handle this problem. How these tools will look light is still a matter of the future. Cognomotiv will focus, towards this end, on deep learning and big data analytics to find these faults and neutralize them.
At the event, NXP announced collaborations with automotive software vendor Elektrobit and infotainment company Harman (see separate article). With regard to NXPs BlueBox computing platform, NXP Automotive CTO Lars Reger said in an interview with eeNews that from his perspective the Bluebox platform certainly has enough computing horsepower for current and next-generation ADAS. For Level 4/ Level 5 autonomy, it will be necessary to further develop the platform. “An autonomous car can only be as good as its environmental sensing,” he said. For more powerful computing platforms, NXP now will rely on the merger with Qualcomm. “High performance systems will be an asset Qualcomm will contribute to the marriage,” he said.