We already know that we won’t see fully autonomous vehicles until after 2030. But the systems that will form the brains of a driverless car are already being introduced piecemeal. The result is a whole new profit center for chipmakers.
According to a new report from analysis firm IHS Technology, microcontroller and processor units for autonomous vehicles will be a half-billion-dollar market by 2020, up 625 percent from $69 million in 2013.
By the end of this decade, Google and other robocar makers will start rolling out vehicles that operate without human intervention—but only under certain conditions: at low speeds, under certain traffic conditions, on approved roads in the handful of areas that they’ve rigorously mapped by then and for which they've managed to gain local government approval the way they have in California. Even so, the demands inherent in having the car run the show will require the car to get smarter. The report notes that in order to run the algorithms that keep a driverless car from being a menace, the number of cores and the speed of each processor will need to increase.
Though the report doesn't specifically say how much faster the chips need to be or how many electronic control units will be transmitting and receiving data, we get a sense that the answer is "a lot" when the report's authors note that cars will likely rely on FlexRay, an automotive network communications protocol that governs automotive computing. It is designed to be faster and more reliable than the CAN bus message-based protocol that runs the electronics in most of today's cars. FlexRay supports data rates up to 10 megabits per second (10 times as fast as CAN), and allows dual independent data channels for fault tolerance.
And to keep the location of every traffic signal, stop sign, lamp post, crosswalk, and fork in the road in mind, a car will need a vast amount of mass storage. The onboard memory will serve as the template against which it compares what its sensors are capturing as it cruises along the road. Here, stability trumps access speed. Cars will also need much more volatile memory—perhaps several high-capacity RAM drives—in order to deal with the torrent of data delivered by sensors as they keep track of the environment, monitor the position, speed, and direction of other cars, and keep an eye out for pedestrians. That is where, say, a RAM drive's speed would come in handy (not to mention the fact that it would lose, the instant the car is turned off, the data which might otherwise be misused by a hacker or an intrusive government agency).
If chipmakers stand to reap a windfall as cars progress to what the U.S. National Highway Traffic Safety Administration refers to as Level 3 autonomy, where a human "is available for occasional control,” I can only imagine what they'll rake in after we reach the point when the human’s only role is to tell the car where to go.