AI is changing the cockpit as well as the brains of the car

LAS VEGAS — According to exhibitors at CES this year, the car is a mobility device, a living room and an office.

A driver can instruct an endlessly patient artificial intelligence-enabled voice assistant to map routes and make calls, while invisible mechanisms use lidar, radar and other sensors to navigate the complexities of hectic city driving, rendering beautiful and functional graphics in seconds. Meanwhile, electric vehicles factor charging times and locations into route calculations. All of this happens through the magic of AI.

But there is neither sleight of hand nor fairy dust behind the magic. The secret to AI-enabled vehicle tools is the computer chips that can handle the Herculean task of processing and responding to reams of data in real time.

All the flashy tech visible to consumers requires a computer platform composed of hundreds and sometimes even thousands of semiconductors. But now, some of the most cutting-edge companies are trying to squeeze much of that onto a single chip.

Intel, Nvidia and Qualcomm all highlighted new forms of this consolidated vehicle architecture that can combine many of the microprocessors into two or three system-on-chips.

The push for vehicle architecture consolidation has set off a debate over whether a car's brains should be this new, more compact technology or the existing system. There are implications for the financial health of Tier 1 suppliers that have banked on the growth of software-defined features through these tiny chips. The old architecture used about $540 worth of semiconductors per car in 2022, according to Yole Group, amounting to billions of dollars.

Proponents of the current system say the new AI architecture isn't ready for prime time.

Tier 1 suppliers are "very worried," said Alex Oyler, director of SBD Automotive North America. The switch "is happening now."

Before emissions standards, vehicles needed very little processing power.

The first automotive microcomputer engine control unit on production vehicles was a single-function controller in the 1977 Oldsmobile Toronado, according to the Institute of Electrical and Electronics Engineers. It was used to control the timing of the electronic spark for the engine.

Since then, vehicle features relying on computers have grown in number and complexity. This has created the need for more and more little computers, including microprocessors, a type of semiconductor. Those chips govern certain functions — for example, buttons on the driver's door — all around the vehicle.

Polar Semiconductor estimates that a modern automobile contains between 1,000 and 3,000 semiconductors, including microprocessors.

"Each feature is a new electronic control unit, so as each new feature was developed, a new box would be added," said Jack Weast, vice president of Intel Automotive. "And these are fixed functions. They're not upgradable, and in many cases the teams that originally built them are long gone."

Intel and the other big chip companies argue that rather than using many chips for different vehicle functions, automakers should migrate to system-on-chips. They say this enables not only greater efficiency and integration of the vehicle features, but also creates cost savings.

The existing system requires "the extra boxes, the extra development, and sometimes, you're paying three different people," said Danny Shapiro, Nvidia's vice president of automotive. "Time is money. It's the physical expense of those boxes. It's extra cabling and connectors and things like that, and then it's the ongoing support of it as well."

The chipmakers are also highlighting that a system-on-chip can support "mixed criticality workloads," which include functions that are safety-critical, such as braking, alongside functions related to digital infotainment.

That would allow automakers to run different operating systems for different functions, such as driver-assist technology or infotainment, on the same chip. In that case, automakers would have an easier time partnering with different software designers or using different software development kits for features and running them on the same architecture.

The industry needs a new path that includes "silicon-enforced separation for mixed-criticality workloads," Weast said.

There are limitations to the consolidation of vehicle architecture. For example, there is only so much physical space on a system-on-chip. It's also not clear when efficiency savings begin to plateau.

"You're still going to need electric windows, and memory in your seat — that won't be controlled by" a system-on-chip, said Kevin Cassidy, a semiconductor analyst at Rosenblatt Securities. "It's just one little memory chip."

The consolidation of vehicle architecture creates a quandary for Tier 1 suppliers such as Panasonic, Denso, Bosch and Harman, among many others.

Until this point, Tier 1 suppliers have cashed in on the growth of microprocessors in vehicles tied to the laundry list of new features and gadgets in the car.

"Every year they get four more $2 chips in there," Cassidy said. "It doesn't matter if the number of vehicles sold doesn't change;" they're "getting 5 percent growth because we're just adding more and more components."

Now, the chipmakers pushing the system-on-chip transition intend to reduce the number of smaller chips in the vehicle, which shrinks the Tier 1 suppliers' electronics volume. Automakers are also generating their own intellectual capital by building their own software.

Tier 1 suppliers are "selling five modules with very little intellectual property on" them, so "you're losing volume and you're losing value," Oyler said. How can the big Tier 1 companies "survive in that environment? It's frankly a question that I don't think the industry has answered."

The consolidation of vehicle architecture onto two or three system-on-chips raises some complex safety questions, Cassidy said. If vehicle operations are consolidated onto fewer chips, that makes those systems more vulnerable to viruses or other compromise. That may not be a huge deal for infotainment, but it could be catastrophic for high-priority functions such as braking and steering.

Placing more information on the chip could lead to reliability issues, said Thomas Coughlin, the 2024 Institute of Electrical and Electronics Engineers president and the president of a data storage consulting firm, Coughlin Associates. Because they are consolidated, some types of the new architecture require smaller versions of the patterns that define the layout of the chip. Those patterns are called lithography.

"There are reliability issues that show up at the smaller lithography features," Coughlin said.

At the very least, companies such as Infineon Technologies, Germany's largest semiconductor manufacturer, are confident that microprocessors will remain an important source of redundancy.

"Some of these systems are responsible for the control of the vehicle and will require redundant compute, power and sensing systems for safety," Bill Stewart, Infineon's vice president of Automotive Marketing for the Americas, said in an email. Even "the central compute needs that include safety and security features will continue to need backup watchdog features in place."

Still, system-on-chip "is the future," said Harald Wimmer, the leader of PwC Global Automotive. "And I think everybody's very clear on that."

Hannah Lutz contributed to this report.