There is little argument that Tesla Motors changed the conversation around automobiles in 2015, or that NVidia caught a ride on Tesla’s coattails. Tesla has set the stage for the automotive future by rolling out new autopilot features — such as lane keeping and self-parking — via over-the-air (OTA) software upgrades. Tesla showed a glimpse of the future in which consumers don’t need to buy a new car to add features. The presumptive car of tomorrow, behaving like a smartphone, is software upgradable. Of course, OTA isn’t a foreign concept to the automotive industry. Some car makers like Nissan have been sending software patches over the air. Ford is partnering with Microsoft to provide continual updates to its next-generation infotainment systems.
But none of the automakers has added software upgradable features for engines, transmissions, brakes or suspensions — like what Tesla did in enabling some autonomous driving functions via software.
To put it mildly, Tesla is freaking out car OEMs and Tier Ones. Today, none of the conventional carmakers can offer anything close to what Tesla does — “without changing the entire hardware and software architecture in a car,” explained Danny Shapiro, Nvidia’s senior director of Automotive.
Armed with the company’s DRIVE PX platform based on its own Tegra X1 processor, Nvidia is coming to Las Vegas next month for CES, pitching its centralized CPU platform to “make cars better and improve their value,” Shapiro explained.
Acknowledging Tesla’s halo effect, Jeff Bier, founder of the Embedded Vision Alliance, said that software upgrades “will create huge opportunities — to save people’s lives and improve efficiency.”
Nvidia, a relative newcomer to the automotive field, has nothing to lose in prompting carmakers to start from scratch and embrace a brand new centralized CPU platform like its PX platform for their new models.
In contrast, neither NXP nor Renesas Electronics – two leading automotive chip suppliers – can afford a grandstand move like Nvidia’s. A lot of their chips are already designed into millions of cars.
Digital networking processor inside a car
In an interview with EE Times, Kurt Sievers, executive vice president and general manager of NXP’s automotive business unit, said, “Nvidia certainly knows how to speak high-tech language” that gets people’s attention.
A modern car already deploys more than 50 ECUs inside a vehicle, with each tasked to dedicated functions, much like a distributed computing architecture.
Since modern cars use many sensors, each ECU also has to pre-process sensory data, which needs to travel via secure connections to a central processing unit for sensor fusion, explained Sievers. Fully aware of the need for a powerful platform to perform software upgrades and complex sensor fusion, NXP, freshly merged with Freescale, is offering car OEMs high-performance multicore networking processors — originally developed by Freescale’s digital networking group. “We are letting our customers try these samples,” explained Sievers.
Sievers, however, does not agree with Nvidia’s approach, which is reminiscent of Intel’s brute-force CPU-centric push for PC improvements. Nvidia today makes no bones about leveraging sheer processing power to revolutionize vehicle architecture and improve the car’s capacity for deep learning. Nvidia is applying its GPU-accelerated deep learning expertise to computer vision. In medicine, for example, the same technology is used to detect cancer cells, said Shapiro. “Frankly, if your only application is automotive, I don’t think you can match the resources you need to explore deep learning as the way we’ve been able to do.”