Available computing power for vehicle processors rising steeply
Development tools, simulation software, processor IP: The software manufacturer Cadence has arrived in the automotive industry. This became clear at Embedded World by numerous exhibits and application examples. For example, the Cadence software Clarity is now firmly established in the development process of automotive electronics, explained Robert Schweiger, Director, Automotive Solutions at Cadence. “EMI and EMC are a big topic in the car, also because mobile phones are firmly established as communications hub in the car,” says Schweiger. The “electromagnetic atmosphere” in the car must therefore be analyzed in detail. Many developers use Cadence Clarity tool for this task.
The software manufacturer is observing an increasing trend to outsource computationally intensive tasks such as finite element analysis to the cloud. “The design environment is in the cloud from the outset,” says the Cadence expert. The advantage is that several modules and simulations at different levels can be combined more easily, because the computing power is available practically without limit or can be switched on as required. “This is particularly interesting for start-ups and small and medium-sized companies,” explains Schweiger. “It gives them much more flexibility than if they had to build up the necessary computing capacity on site”.
The acceptance that Cadence’s Tensilica processor IP is now finding in the automotive industry is at least as striking. “More and more functions are migrating into the chips,” says Schweiger. At Embedded World, Cadence showed prototypes that focus on Tensilica processors. These include two hardware platforms for infotainment applications, computer vision, radar / lidar sensor fusion and real-time data processing as well as AI-based speech recognition with suppression of ambient noise.
Cadence supports collaboration with companies from the automotive value chain through various cooperation models. “Cadence has its own service unit to implement strategic products together with automotive customers,” says Schweiger. Customers could also acquire IP know-how in the process.
Parallelity speeds CNN computing
The AI start-up Kalray also is targeting the automotive market. At the Nuremberg fair, it showed its latest processor Cooldige, which relies on massive parallel processing and integrates five clusters with 16 cores plus 16 coprocessors each. The latter are designed as hardware accelerators for Convolutional Neural Networks (CNNs) and enable the processing of a corresponding number of AI tasks running in parallel. The cores can run standard operating systems such as QNX or Linux as well as real-time operating systems (RTOS). This constellation is ideal for use in driver assistance systems and autonomous driving, as Stéphane Cordova, Vice President Embedded Business at Kalray, explained.
According to Cordova, Kalray is involved in many development projects in the automotive sector – unfortunately, he was not allowed to mention the names of most of them. One exception is the French car manufacturer Renault, whose current concept study incorporates Kalray’s AI processor. Manufactured in TSMC’s 16 nm process, this processor has a particularly low power consumption of just of 20 to 30W. With this, it does not even require a very complex cooling system – a fact that is very important in automotive electronics. According to Cordova, Kalray is planning a further performance-enhanced version of the processor chip for next year. The Coolidge 2 should double the computing power to 50 TOPS.
The use of AI in safety-relevant applications in cars is a bit in the air, however. “When AI and functional safety meet, determinism is a major problem in industry,” confirmed Cordova. No one has yet announced a deterministic CNN. Therefore, the SOTIF design principle (Safety often he Intended Functionality) complements the ISO 26262 standard relevant here: “I don’t think it is possible to certify all the program code (generated by AI) according to safety criteria,” he explained. “Perhaps it would be possible to certify the result instead of the code,” Cordova suggested.
Related articles:
Machine learning for safety-critical functions? Kalray says yes
Tensilica P6 boosts image recognition performance for ADAS chip
Why chip design gets relevant for carmakers
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
