SAN JOSE, Calif. – Fully autonomous cars will drive to the market by 2022, carrying a $10,000 price premium. Their arrival will help double by 2025 the size of today’s $10 billion automotive semiconductor market, fueling new segments like vision processors.
That’s one of the conclusions analyst Linley Gwennap draws in preparation his annual Linley Processor Conference next week. Cadence aims to ride the trend with a new vision DSP its former Tensilica group will describe at the event.
Among other expected news items at the conference:
- Startup Soft Machines will describe its first CPU core and SoC platform
- The HSA Foundation will reveal adopters of its heterogeneous SoC specifications
- MoSys will disclose a third-generation Bandwidth Engine
- Arteris and Synopsys will introduce new memory interfaces
- NetSpeed will describe a tool for designers of coherent systems
- And Marvell may provide an update on it MoChi SoC initiative
This year’s event is the first to shine a light on the growing market for automotive chips, thanks to the rise of self-driving cars.
“I’ve talked to several people working in the field, and the general agreement is the technology [for self-driving cars] is there and being tested,” said Gwennap, principal of the Linley Group (Mountain View, Calif.), noting car makers are already working on 2018-2020 model cars given their long design cycles.
Today’s cars already have lane detection, that’s a fundamental capability you need to keep the car on the road. Adaptive cruise control and features to maintain a safe driving distance are being tested -- those are two other main pieces for highway driving.
Companies such as Mobileye have solved a lot of the complex urban driving problems. But there are still a lot of corner cases R&D has to deal with. People talk about wanting to take over driving in a time of emergency, but that’s the time you don’t want to take over because a [self-driving] car is most likely going to deal with an emergency better than a person.”
Today’s cars are already adding many cameras for assisted-driving features, spawning a new class of vision-processing units (VPUs). “It’s become a whole new sector, and we see a lot of the development going on here in Silicon Valley,” said Gwennap, citing Cadence’s rival Synopsys among others.
“You need to interpret all the visual data coming from these cameras -- a standard CPU can’t handle it,” he said. “These new VPUs are designed for this large amount of pixel data, resolving it into objects and mapping everything out,” he said.
The expected $10,000 premium for self-driving cars “pays for lots of cameras and processing horsepower,” he added.
Asked about the next steps in the current consolidation of the semiconductors industry, Gwennap predicted shifts among vendors of networking chips. Applied Micro could be ripe for an acquisition and companies including Broadcom, Freescale and Marvell might spin out their network-chip divisions, he said.
“Given the way Avago chopped up past companies it acquired, it could spin out groups such as Broadcom’s NetLogic network processor business -- the $100 processors and Broadcom’s high-end Ethernet switches require big engineering commitments that don’t fit the Avago model,” Gwennap said.
“Qualcomm is a potential buyer because it has been somewhat acquisitive lately and is looking for new markets, so a big embedded networking business” could help it diversify beyond its base in mobile systems, he added.
— Rick Merritt, Silicon Valley Bureau Chief, EE Times