Okay, so what the heck is Amazon doing in the car business? The straightforward answer is that Amazon Web Services has gone whole hog into “Connected Vehicles.”...
Okay, so what the heck is Amazon doing in the car business? The straightforward answer is that Amazon Web Services (AWS) has gone whole hog into “Connected Vehicles.”
NXP Semiconductors this week partnered with AWS, with the goal to enable car OEMs to collect and harness the voluminous streams of data generated by their own vehicles.
The auto industry, of course, has been talking up connected vehicles for a long time. Connectivity installed in vehicles, for example, already enabled carmakers to build and offer telematics services such as General Motor’s OnStar. It also allowed users to download apps and other content to in-vehicle infotainment systems.
“Let’s call them Phase I and Phase II of the connected vehicle,” said Brian Carlson, global marketing director for vehicle control and networking at NXP. In Phase III, the new NXP-AWS partnership seeks to make “vehicle-wide data available to car OEMs,” he explained.
But what kind of data?
By using a high-performance, high-bandwidth vehicle gateway processor such as NXP’s S32G, carmakers can now transfer vehicle data — from “sensory data to algorithmic and behavioral data” — to the cloud, Carlson claimed.
Some data will remain processed at the edge, right inside a vehicle. But NXP-AWS deal will open the door for OEMs to mine and examine — on the cloud — a lot of data previously unexplored.
In the IoT world where connectivity plays a crucial role, the design community has moved away from workbench to the cloud by using such services as AWS and Microsoft Azure.
Connected vehicles are next. “This will be a new trend,” said Egil Juliussen, a veteran automotive industry analyst. A lot of design/engineering work in automotive is moving to the cloud because “cloud services offer many tools that engineers can leverage,” he noted.
What sort of work will automotive engineers to do on the cloud?
The first order of their business is to improve AV/ADAS perception and machine learning (ML) algorithms in the cloud. Ideally, vehicle network security and the health of EV batteries can be better monitored from cloud by taking advantage of the real-time data coming from vehicles.
OTA Updates: Where Are We?
AI on the cloud
A lot of automotive capabilities are already being developed in the cloud with the availability of machine learning tools to develop, train, optimize and deploy models to vehicles for execution (ML/DL inferencing).
With cloud-based AI training already commonplace, what additional benefits might the NXP-AWS partnership offer? The key here is the use of real-world data, which lets developers improve performance and safety, said Carlson. Real-time connectivity empowers carmakers to detect and capture corner cases and anomalies, he added.
So, whose AI processors will NXP-AWS partnership target for algorithm improvement?
As it turns out, the AWS SageMaker Neo product can target multiple types of ML engines or even optimize for CPUs or SIMD engines. This includes x86, Arm, RISC-V others, explained Carlson. In parallel, NXP’s S32G vehicle gateway processor will use its PCI Express interface to extend ML inferencing to a host of powerful AV/ADAS application processors “ranging from Nvidia to FPGA and Qualcomm’s Snapdragon,” said Carlson. Soon, Google TPU, not enabled now, can join the crowd.
As a centralized gateway silicon provider, it’s crucial to be flexible and processor-agnostic, Carlson said. “We have to be able to support deployment of [AI] models throughout the vehicle,” regardless of whose processor that might be.
Meanwhile, let’s not forget that NXP also has its own ML targets, including S32V vision processor. NXP developed a toolkit called eIQ Auto, which will speed the “quantization, pruning and compressing” of the neural network. Much of that work gets done in the cloud by using the data coming from the connected vehicle.
With NXP’s automotive processing expertise combined with the cloud infrastructure by AWS, NXP outlined a host of new services that car OEMs can develop.
Sending the real-world data generated by lidar, radar, camera and other sensor information to the cloud for analysis, as noted above, is the important first step to improve the perception of AV and ADAS vehicles.
For NXP, Carlson pointed out that “electrification” is a primary use case showcasing NXP-AWS collaboration. The edge-cloud model enables real-time monitoring of battery, motor and other propulsion information. By driving digital twin models in the cloud, NXP is improving energy management and extending vehicle range of EVs, he claimed.
Security is another significant area where connected vehicle data will play a big role. Think “Intrusion Detection,” said Carlson. “With the data sent to the cloud, security can be improved with ML there and updates can be deployed to the vehicle fleet for Prevention.”
Many carmakers are also counting on the connectivity to make “vehicle health management” easy. Carlson explained that the real-time monitoring of vehicle data with edge, combined with cloud ML, enables carmakers to find problems “even before the vehicle knows or shows an error code or “check engine” light.
Real-time analysis of driver monitoring based on both edge/cloud processing will become a crucial element in supporting the future of connected, highly automated vehicles.
Changing role of ‘gateway’ processors
Expanded access to whole-vehicle data is vital for carmakers poised to offer more sophisticated Over-the-Air (OTA) updates.
Equally important is the ability of a gateway processor such as S32G to talk to a variety of in-vehicle ECUs, even though many of the chips aren’t necessarily supplied by NXP. From a gateway perspective, NXP’s goal is to support OTA updates of all the ECUs in the system, stressed Carlson.
Over time, as telematics has proliferated, carmakers have begun offering software updates wirelessly. Still, major software updates and repairs continue separately via OBDII ports. The emergence of more powerful gateway processors can make it possible for carmakers to develop cloud-based software services, observed Juliussen. This is an area where everyone wants to get a piece of the action in, because, potentially, it can generate lucrative after-sales revenues.
In theory, big automakers like GM and Toyota can build their own cloud platforms, said Juliussen, and they do.
But those same companies are also adding off-the-shelf cloud services such as AWS and Microsoft Azure, said Carlson. This is because with their tools, Amazon or Microsoft can make it easy for carmakers to developing cloud-based software and services.
Compared to the traditional gateway ECUs used in today’s vehicles, S32G is hugely different, claimed NXP. The role of traditional gateway ECUs has been pretty much limited to moving data securely around inside the vehicle, he explained. In contrast, “The S32G, for example, can provide high-speed networking acceleration to process the data, handle the high-bandwidth applications processing real time, and connect the data securely to the cloud,” Carlson noted.
Now that car OEMs are finding ways to make S32G perform more tasks, Carlson jokingly said, “We are now saying the G in the S32G stands for general-purpose processor.”
Just to recap, S32G is a vehicle network processor that combines ASIL D safety, hardware security, high-performance real-time and application processing, and network acceleration for service-oriented gateways, domain controllers and safety co-processors.
The processor includes: Quad Arm Cotex-A53 cores with Arm Neon technology organized in two clusters of two cores with optional cluster lockstep for applications and services. It also features triple Arm Cortex-M7 lockstep cores for real-time applications, a low latency communication engine for automotive network acceleration and packet forwarding engine for ethernet network acceleration.
As powerful as S32G might seem, Carlson stressed that the goal of NXP’s gateway processor isn’t to send terabytes of raw data to the cloud.
NXP is already working with companies such as Teraki and SafeRide to reduce the data that goes to the cloud by spotting specific anomalies and identifying intelligence. Noting that the company is aware of needle-in-the-haystack problems that plaque the big data, Carlson explained that the goal of the gateway processor is to transmit only what’s important without burning a lot of cycles.
The lesson is that the development of new cloud-powered services takes a village. Over at NXP, “we are developing a number of partnerships right now,” said Carlson. The NXP-AWS deal is “just the starting point” of what more to come from the new infrastructure, he added.