Your browser is not supported. Please update it.

19 May 2016

STMicro, NXP, CEVA step up connected car hardware war

This week, three silicon vendors have announced wins that position them well in the connected car market – but less than 1% of consumers will recall ever having heard their names. STMicroelectronics, NXP, and CEVA have all released deals that should further their automotive ambitions.

The market is still emerging, and there are a lot of niches left to contest. Similarly, there are a number of players that have shown no real interest in certain areas – such as Nvidia’s apparent exclusive focus on the industry-leading machine learning and AI functions, but absence in the less headline-grabbing airbag or ADAS applications.

First up, STMicro has announced a new phase in its partnership with Mobileye, an image-processing and machine learning focused vendor that develops a range of automotive chips called EyeQ. The news this week sees Mobileye pick STMicro to bring the fifth generation of that chipset to market, with the EyeQ5 being co-developed by the two – with ST having developed the previous four generations. It should be noted that the first engineering samples are a long way off – expected in the first half of 2018.

The EyeQ5 is being built on a 10nm FinFET process, with eight multithreaded CPU cores, paired with eighteen Mobileye vision processors – and ST notes that the 10nm could be shrunk further, but doesn’t give a lower limit. Collectively, the new package claims to offer 18x the performance of the gen-four EyeQ SoC, at 40Gbps bandwidth. Its claimed 5W power consumption is rather impressive, which allows the SoC to use passive cooling – which removes the need to incorporate air-flow or additional fan circuitry into its designs.

The latest in a string of cooperation with STMicro, the new EyeQ5 “is designed to serve as the central processor for future fully-autonomous driving, for both the sheer computing density, which can handle around 20 high-resolution sensors and for increased functional safety,” said Amnon Shashua, Cofounder, CTO, and Chairman of Mobileye. “The EyeQ5 continues the legacy Mobileye began in 2004 with EyeQ1, in which we leveraged our deep understanding of computer vision processing to develop highly optimized architectures to support extremely intensive computations at power levels below 5W, to allow passive cooling in an automotive environment.

Mobileye notes that its new SoC confirms with that all-important ISO 26262 safety standard, as well as the stringent security standards required by those Tier 1 automakers. Those who purchase the SoC will also be receiving a full suite of hardware-accelerated algorithms and automotive-grade OS (unnamed, but surely RTOS), as well as an SDK for prototyping and deploying the neural networks that will leverage the image-recognition capabilities of the chip – which process the inputs from cameras, LiDAR, and radar.

The second announcement this week sees Autotalks license and deploy the CEVA-XC DSP in its upcoming second-gen V2X (Vehicle-to-X) chipset – which will allow the chipset to communicate with other vehicles (V2V), smart city traffic infrastructure (V2I), and even pedestrians.

The pair note that their announcement comes before the US Department of Transport is expected to announced a Notice of Proposed Rulemaking (NPRM) that would see it mandate the installation of V2X communications technology in new vehicles.

The new chip includes support for IEEE 802.11p, a WiFi-based standard that is specifically designed to operate in a protected part of the 5.9GHz band – which is currently facing attack from WiFi-proponents desperate for new spectrum to alleviate frequency congestion. If the protected vehicular spectrum is removed from the FCC regulations, many in the automotive space will mourn its early passing – and likely fear their reliance on the more crowded ISM bands.

The CEVA DSP will allow Autotalks to “implement the PHY layer and specific layers of the MAC layer on the DSP, allowing scalable V2X use cases. The implementation supports concurrent dual-channel operation; either two different 802.11p channels or a combination of 802.11p and 802.11a/b/g/n/ac. The solution outperforms all SAE2945/1 requirements, and also incorporates an innovative power scaling unit (PSU), that enables application-specific low power modes, ensuring the chipset is finely tuned for power efficiency.”

The final announcement is a little different, in that NXP has announced a new strategy, rather than a specific chip or product. The silicon designer has launched BlueBox, a modular way of designing connected car platform using a combination of electronics from NXP that augment the car itself.

NXP says BlueBox is being trialed by four major automakers, and will make it first appearance in consumer-facing self-driving cars in 2020. The platform comprises two NXP chips, the S32V vision-processor, and its eight-core, 64-bit LS2088A processor, in a 40-watt power package – higher than the 5-watt Mobileye-STMicro chip, but substantially lower than the 250W+ Nvidia Drive PX 2 package that the GPU-specialist was showing off at CES.

The major benefit of the modular approach that NXP is pushing is that OEMs can pair other chips from different electronics manufacturers, without NXP completely locking itself out of the vehicle – as it might if it rigidly stuck to a closed model. Given that cars typically feature dozens, if not hundreds, of separate CPUs and sensors, it would be a little foolish for NXP to assume that an automaker would use a pure-NXP solution – and so its BlueBox offering allows it to ensure that it has the option of being included in every connected car blueprint, and consequently increasing its volume shipments.

It’s somewhat shrewd, but it makes a lot of sense. It also means that NXP doesn’t have to focus on a particular niche in the market, and can offer its chips to those looking for ‘simple’ parking-assist applications or near-autonomous platforms. With regular ADAS, to full self-driving, NXP is aiming for non-vision applications too, as opposed to the pure image-processing options of some of its rivals.