As carmakers integrate 100-150+ chip-based electronic control units in each car, a look at the challenges of an auto industry that is more dependent on software
The trend toward self-driving and electric vehicles will add hundreds of millions of lines of code to cars. Can the auto industry cope? Tweets: @davewiner , @bertrandduflos , @azaidman , @trengriffin , @sub8u , @drorpoleg , @jamescham , @andwor , @azaidman , and @azaidman Tweets: Dave Winer / @davewiner : The trend toward self-driving and electric vehicles will add hundreds of millions of lines of code to cars. Can the auto industry cope? https://spectrum.ieee.org/... Bertrand Duflos / @bertrandduflos : Managing complexity: The electric/electronic stack(s), multiple ECUs, and zillions of lines of code, of modern vehicles. Very nice article. https://spectrum.ieee.org/... Andy Zaidman / @azaidman : Deloitte estimates that >=40% of a vehicle's development budget can be attributed to systems integration, testing, verification and validation. [...] Unsurprisingly, efficiently managing variant complexity is a significant concern for the auto industry. https://spectrum.ieee.org/... Tren Griffin / @trengriffin : 1/ Volkswagen Chairman Herbert Dies says, “Software will account for 90% of future innovations in the car.” ..hardly a line of software code comes from us." VW estimates that only 10% of the software in its vehicles is developed in-house. https://spectrum.ieee.org/... Subrahmanyam Kvj / @sub8u : Code on wheels! Good read. “...as of 2020, Volvo has a superset of about 120 ECUs... they comprise a total of 100 Mn lines of code. This code contains 10 Mn conditional statements as well as 3 Mn functions, which are invoked some 30 Mn places in the code.” https://spectrum.ieee.org/... https://twitter.com/... Dror / @drorpoleg : Attention, real estate owners and operators. https://twitter.com/... James Cham / @jamescham : All over different industries, the center of gravity for business value is shifting. https://twitter.com/... Andreas Wortmann / @andwor : Interesting: “Run a job assessment with all top managers at VW, Audi, Porsche, BMW, and Daimler tomorrow and ask them to code [...] if they are not able to do so, fire them immediately, because they are not fit for the job.” Former head of Audi's R&D. https://spectrum.ieee.org/... Andy Zaidman / @azaidman : Low-end vehicles contain 100M LOC, while high-end vehicles with advanced driver-assist systems may contain 150 electronic control units and 150M LOC. https://spectrum.ieee.org/... Andy Zaidman / @azaidman : “Once, software was a part of the car. Now, software determines the value of a car” —Manfred Broy https://spectrum.ieee.org/...
Context & Ripple Effects
The IEEE Spectrum piece lands after a year of evidence that legacy automakers' software capacity is the binding constraint on their EV and autonomy plans. Volkswagen is the case study: its ID.3 launch was delayed and beset by faulty in-car software, forcing a reboot of its software organization, followed by a $30B software push hampered by internal battles and technical glitches. VW itself estimates only ~10% of its vehicles' software is developed in-house, against a stated goal of building 60% of it internally by 2025 with a Car.Software unit expanded to thousands of engineers.
The structural backdrop the article documents: cars now carry 100–150+ chip-based electronic control units, Volvo's 2020 superset alone ran about 100 million lines of code across ~120 ECUs with ~10 million conditional statements, and Deloitte estimates at least 40% of a vehicle's development budget goes to systems integration, testing, verification and validation — costs that scale badly as code volume grows.
First-order effects
- Volkswagen and peers face direct engineering-cost pressure: with ≥40% of development budgets consumed by integration and testing of distributed ECUs, every added line of autonomous-driving code raises validation cost faster than feature value.
- VW's Car.Software unit must close the gap between ~10% in-house software today and its 60%-by-2025 target while simultaneously debugging the architecture that failed on the ID.3 — hiring and reorg speed becomes the product roadmap.
Second-order effects
- The fragmentation problem forces an architecture response: the industry's effort to replace the aging CAN networking standard with a higher-bandwidth backbone for sensor data (the CAN replacement quest) becomes a prerequisite for consolidating dozens of ECUs into fewer compute platforms.
- Suppliers and tier-1s shift from selling discrete ECUs to selling integrated software stacks, and automakers that cannot build in-house — most of them, per VW's own 10% estimate — compete for a limited pool of software engineers, bidding up labor and pushing smaller players toward licensing rather than owning their stack.
Third-order effects
- If the pattern holds, the industry bifurcates into a handful of vertically integrated software-defined-vehicle makers (Tesla-style) and contract manufacturers running licensed stacks — with the integration-and-validation share of budgets deciding who can afford vertical integration at all.
- Regulatory attention follows the failure mode: as hundreds of millions of lines of code control safety-critical functions, homologation regimes built around mechanical certification will be forced toward continuous software verification standards, raising the compliance bar that already consumes the largest slice of development budgets.
The trend: Car manufacturing is becoming a software-integration business first, and the automakers' ability to consolidate 100+ scattered ECUs into fewer owned software platforms will sort winners from licensed-stack renters over the next decade.