Stellantis looks to AI to slash the number of chips in cars

Stellantis looks to AI to slash the number of chips in cars

Technology News |
By Nick Flaherty

Cette publication existe aussi en Français

Stellantis is looking to a new automotive architecture to significantly cut the number of processors in a vehicle.

The new Stellantis platform design includes a ground-up design for vehicle electronics and includes semiconductors from the SiliconAuto joint venture with Foxconn. This comes after Stellantis CTO Ned Curic highlighted the need to cut the number of chips.

“Right now in vehicles, there are 50 to 120 electronic control units (ECUs),” said Steve Rober, Senior Vice President and Global Head of Advanced Electronics & Semiconductors at Stellantis, tells eeNews Europe.

“We are looking to centralise the architecture with high-performance computing and zone controllers with a significant reduction in edge devices. The current architecture for most vehicles is almost entirely edge devices,” he said.

“We want to use AI and other methods to centralise the architecture. The algorithms want to get access to the raw signals so we want to centralise the data,” he said.

“In some instances, there is value in having the AI as close to the sensor as possible, but in many cases, you are looking for many different things in the data so you may want to bring it to the central processor. It also depends on the network architecture and the ability to move large amounts of data,” he said.

This simplification changes the vehicle architecture significantly, he says. This will require higher-speed Ethernet networks throughout the vehicle rather than the current CANbus implementations,  with controller chips in each of probably four zones in the vehicle feeding into a central AI processor.

This simplifies the sensors and the software development and builds on the acquisition of Hungarian AI technology developer aiMotive last year, which has developed hardware and software IP to add into automotive chips from Nextchip in Korea. This starts with Level 2 advanced driver assistance systems (ADAS) that are easily upgradeable to Level 3 and then to Level 4 autonomous driving.

“If you look at how autonomous driving has evolved, the current architectures are like a patch with additional solutions retrofitted on top, especially the L2 ADAS,” said Mustafa Imran Ali, product manager at aiMotive and an alumni of ARM in the UK.

“aiMotive software has evolved to be a centralized processing architecture to tackle this problem today for L2 and then offer L3 capabilities. If you are building on existing architectures, there is not a simple approach to do that.”

“So we are looking at the software architecture and marrying that with the sensor and network architectures. Stellantis is looking at a brand new platform for autonomous driving where we can do things better by starting from scratch.”

“As we take a clean sheet approach with higher capability multicore, multithreaded microprocessors with AI accelerators, we can design together with the joint venture and high-speed network,” said Rober.

This doesn’t necessarily mean an increase in automotive memory to support the higher amounts of data.

“There are instances where you need extra memory, particularly for AD data logging, but you don’t need to store it in the vehicle for very long before you send it to the cloud,” he said.

The designs will begin to come through “within the next few years,” he says.;


If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles