The Internet of Cars Is Paved With Silicon


EETimes

September 21, 2021

Technologists often overuse words such as “disruptive” and “revolutionary,” but the recent evolution of automobile design and manufacturing is undoubtedly both disruptive and revolutionary. These changes represent the most significant upheaval in transportation design over the past 100 years, the greatest advancement since the early days of cars, when steam, gasoline, and even battery-powered vehicles mingled on the same dusty roads. The entire car is being overhauled, this time with electronics, to re-make cars as supercomputers on wheels.

One BMW engineer lamented that his team had spent decades designing cars with superb handling and performance, yet the first thing buyers want to know is whether the vehicle integrates with their mobile phone. You can understand the frustration of traditional car designers as electronics and software replace mechanical design as customer selection criteria.

It is no accident that Mercedes-Benz, Nissan, Volvo, and others are running ads that highlight their cars’ wireless phone chargers, all-digital dashboards, collision-avoidance systems and 24-channel stereos. No one mentions horsepower, RPM, ride, handling or even “rich Corinthian leather.” Even Ford truck commercials tout the number of 12V outlets and the built-in generator, not the truck’s cargo capacity or towing ability. GVWR (total maximum weight) has given way to Teraflops and MHz.

Electric motors, self-driving technology, infotainment electronics, subscription-based services, after-sales wireless updates and emerging vehicle-to-anything communication have exponentially increased the demand for semiconductor chips and software. Industry analysts now treat automotive electronics as a distinct category, a sub-industry unto itself, like computing or aerospace.

Where is this all headed? The now-mandatory connectivity to the internet and cloud datacenters shifts automakers’ core competencies from mechanical design to software and silicon. The Internet of Cars will not run on gasoline or even electricity but on data. That data is gathered, processed, communicated and stored by system-on-chip (SoC) semiconductors.

Auto SoCs drive demand

The world’s automakers produce about 100 million cars and trucks every year. According to research firm IHS Markit (Oct. 2020), more than half of those will include 15–24 complex electronic SoC devices by 2026, with an average of 23 SoCs per vehicle. Multiply those 23 SoCs by about 60 million cars, and you end up with nearly a billion and a half complex SoCs a year for the automotive market alone—and that is only five years from now. By the early 2030s, 10 years from now, the demand will increase to over 2 billion complex automotive SoCs.

The Internet of Cars represents the most significant semiconductor opportunity since the smartphone. Some of that silicon goes into the high-profile self-driving technology of advanced driver-assistance systems (ADAS). Still, most goes into infotainment and telematics, the more familiar and widespread features of nearly every car: radio, GPS navigation, electronic instrumentation, mobile phone integration, engine management, and so forth. These are all must-have features in an automotive retail market driven by consumer demand—hence, the feature-forward car commercials.

But that technology is only for the cars themselves—the “endpoints” in engineering jargon. What about the infrastructure behind them? Automakers require cavernous rooms full of computer servers, not unlike Amazon, Google, Facebook, or Microsoft.

If we assume one server per 200 cars, that amounts to 300,000 servers per year. Furthermore, municipalities around the globe are building out infrastructure. Already, “smart city” technology is being deployed in millions of traffic signals, highway signs, cameras, streetlights, roadside sensors and vehicle-to-infrastructure networks, supported by countless miles of wired and wireless access points. Cars will communicate with each other and roadside base stations for latency-sensitive reactions. All of this needs additional silicon.

Then there is the software. Conservative estimates put the in-car software stack at about 30 million lines of code. (For comparison, older versions of Microsoft Windows had about 50 million lines of code.) Some estimates run over 100 million lines for a new car with all the bells and whistles. Either way, it is a complex, networked, high-reliability multiprocessor system.

Hardware reliability is crucial, as automakers will be on the hook legally for any dangerous failures. Chip designers respond with fail-safes and functional redundancy, in which an SoC has two or more internal processors running side by side, continually checking each other. This adds even more silicon and more complexity to the SoC design.

Although commercial off-the-shelf chips will often suffice, automakers can squeeze more performance from their hardware and add differentiation to their software if they design their own custom chips. Image-processing neural networks, for example, need very specific machine-learning features which are large and complex. An example of such a development is the latest Tesla Dojo SoC for turning millions of videos coming from the Tesla fleet into neural net learning to improve automated driving user experience.

Automotive decision systems work best with either fully custom chips or intellectual property (IP) blocks incorporated into semi-custom chips. That means many of those 2 billion SoCs coming in the early 2030s will be fully or partially customized to meet the demands of the new automotive ecosystem.

Automotive companies worldwide are looking to develop their own solutions, which is why many, if not most, end up working with companies such as ourselves (Arteris), Arm, and CEVA — companies that license IP for SoCs. Our view of where the automotive market is heading is based on our experience over more than a decade working with more than 30 automotive industry customers.

Data is king

It is hard to say whether Tesla’s share price is overvalued or undervalued, but it is indisputable that the company exemplifies the electronics-first approach to automaking. It is no coincidence that Tesla’s market capitalization is valued as if it were a Silicon Valley tech company. Following the lead of many consumer electronics companies, it designs many of its chips and develops its software. Its core competence is in electronics, not metal bending. The company also manages its datacenter servers. Its cars are always connected to the internet and Tesla’s servers, with vast amounts of data flowing in both directions in the form of maps, entertainment, traffic information, visual data collected by onboard sensors and software updates. This data is of immense value to Tesla, and other Internet of Cars companies—which is to say, all of them—will need to build similar systems.

Mobileye is another example of a company that uses data coming from its customers to constantly improve its technology based on leading-edge software, SoC silicon and machine learning techniques. Automakers do not make everything from scratch, of course, but rely on well-established networks of suppliers and OEMs—Bosch, Continental, Denso, ZF, Magna, and many others—that design and build major subsystems. These are mainly electromechanical devices, such as automatic transmissions or air conditioners. In the future, these top-tier suppliers will have to build SoC design and software development teams to continue supplying high-value subsystems to automakers. If not, other firms with more electronics experience will likely take their place. Many custom SoC developments will involve partnerships between traditional semiconductor design companies that know how to build SoCs and automotive OEMs and Tier 1s that have the application knowledge.

Although the Internet of Cars will take a long time to mature, it has already emerged, fueled not by electricity, hydrogen or gasoline but by data. Data drives the ongoing improvement and evolution of the user experience, and consumers are willing to pay for that revolution. Automakers’ core competencies are moving from the mechanical design of engines and chassis to software and silicon. The Internet of Cars is paved with silicon.