Qualcomm, long the world's largest vendor of smartphone chips, is looking ahead to the looming age of autonomous cars and entering the ring against rivals Intel and Nvidia. The San Diego, California-based company last month rolled out its first supercomputer for cars, called Snapdragon Ride. It can be used to handle all sorts of functions from advanced driver assistance systems (ADAS) to fully autonomous cars (AVs).
Snapdragon Ride has three core components: system-on-chips (SoCs) that act as the heart of the system, accelerators (ASICs) for handling artificial intelligence (AI) chores, and all the self-driving software. It is capable of carrying out up to 700 trillion operations per second (TOPS), with more than double the speed of current solutions on the market sold by Nvidia and Intel. It also consumes less power and vents less heat, Qualcomm said.
Qualcomm said it would be used to integrate all the data from the camera, radar and other sensors, meticulously model the car's surroundings and react to danger on the road ahead. Snapdragon Ride is also significantly more compact than the water-cooled servers crammed in the trunks of autonomous cars being tested today, saving space and weight. Qualcomm plans to put Snapdragon Ride behind the wheel of a car by 2023.
Nakul Duggal, corporate vice president of product management at Qualcomm, said that Snapdragon Ride aims to ease engineering challenges at every level of autonomous driving. He added in a statement that it "aims at accelerating the deployment of high performance autonomous driving to mass market vehicles." Qualcomm said Snapdragon Ride supports the highest standard of functional safety in electronics—called ASIL-D.
The San Diego, California-based company has been trying to steal more market share in cars following its failed bid to buy NXP Semiconductors. Qualcomm is one of the world's largest vendors of chips used in digital dashboards and cellular modem chips that can be used to connect cars to the cloud. Despite starting from the back of the pack, Qualcomm said last month that it had won a new ADAS system contract with GM.
Qualcomm is looking to gain ground on the competition. Intel has jumped out to the lead in ADAS with its $15.3 billion Mobileye deal, which sells chips for the front-facing cameras in cars paired with its computer vision software, HD road maps and other tools. Nvidia has also bet billions of dollars to advance its ambitions in self-driving cars, which has helped catapult it into a very serious challenger to Intel's Mobileye.
NXP has also been fighting fears of being left behind by Qualcomm, Intel, Nvidia, and others that leapfrogged it in the race for autonomous cars. NXP recently rolled out its latest line of networking chips to combat the swelling communications load in autonomous cars. It is also selling chips to manufacturers adding advanced ADAS to cars. Texas Instruments, Infineon, and Renesas are other firms trying to secure a spot in ADAS..
Snapdragon Drive is a modular, scalable system based on the ADAS processors and AI accelerators. The chips can be chained together in different way to cover all the needs of ADAS all the way to AVs. Packaging all these chips together also adds redundancy to the system, which will help guard against catastrophic failures due to faults or other glitches. Qualcomm said it offers car manufacturers three levels to select from.
A single ADAS chip is designed to deliver 30 TOPS of performance, giving it the processing grunt to handle advanced driver assistance systems (ADAS). That ranges from lane control systems that gently guide the car back into the lane when it starts to drift over the line to blind spot alerts that are tripped every time a car passes in another lane. These electronic safety features are installed in nearly all cars now and are key components of NCAPs.
The ADAS SoC is based on a heterogeneous compute architecture. It has image signal processors (ISPs) and digital signal processors (DSPs) for sorting through all the data from the cameras, radar, and other sensors around the car, and graphics processing cores (GPUs) for mapping its surroundings. The central processing cores (CPUs) are used to plan out the path of the car. The chip also has AI and computer vision (CV) engines.
A second ADAS processor can be added to the Snapdragon Ride to handle more advanced safety systems. These Level 2 systems—including Tesla's AutoPilot and Volvo’s Pilot Assist—can be used to control the speed, steering, signaling and other electronic domains to keep the car centered in its lane at a set speed and distance from other cars. But the driver must be prepared to intervene if road conditions degrades or the system cuts out.
The chips can also be used to handle Level 3 systems that completely take control of the car—from using turn signals and changing lanes to speeding around slow drivers on the road—in very limited conditions. You can take your hands off the wheel, your feet off the pedals and your eyes off the road. But the system requires the driver to grab the wheel when the car is stumped. Today, the highest level of automated driving in cars is Level 2.
Customers can add the AI accelerator for a big burst of performance. Qualcomm said that the Snapdragon Ride computer scales to four chips: two ADAS processors paired with two AI accelerators that can run 700 TOPS at 130W. The high-end system can be used to design and test the Level 5 autonomous cars of the future that never have to hand over the wheel. These types of cars could come without a steering wheel or pedals.
The Snapdragon Ride computer can also be passively cooled, the company said. The trunks of autonomous cars today are crammed with cooling fans and thermal solutions that pipe water or other coolants to the electronics to avert overheating. These cooling systems are all potential points of failure in the car and burn through large amount of power, which is why they are not scalable, Qualcomm said. Passive cooling is also cheaper.
Qualcomm is also following in the footsteps of its rivals by supplying advanced self-driving software. The software stack in Snapdragon Ride covers all the core functions of autonomous driving, from integrating data from sensors around the car ("sensor fusion") to charting its surroundings ("perception") to planning the route of the car ("path planning"). Car manufacturers can also host their own software stacks to Snapdragon Ride.
On top of Snapdragon Ride, Qualcomm has introduced a new Car-to-Cloud service so that manufacturers can safely and securely update their cars from the cloud. It also rolled out blueprints for adding short-range radio technology to cars so that they can chat with each other and infrastructure. Using the C-V2X standard, the cars can alert each other to danger on the road by sharing location, speed, and other data over cellular networks.
