Geely Subsidiary SiEngine Begins Mass Production of ADAS Chip Capable of 2,048 TOPS

SiEngine, a Chinese automotive R&D firm, has announced the commencement of mass production of its latest AI-capable chips for autonomous vehicles and ADAS. The new AD1000 and AD800 SoCs (System-on-chip) are fabricated on a 7nm process for thermal and power efficiency. They are targeted at car companies who are looking to implement L2-L4 autonomous driving, ie hands-free, attention-free driving. 

The company is a subsidiary of Geely Holding, one of the country’s largest privately owned automotive conglomerates, and was co-founded by ARM China. It is known for its smart cockpit solutions, which have been adopted by multiple carmakers as well as Volkswagen. The two new chips are part of a new series called StarLight. While the AD800 is a more basic solution, the AD1000 is intended for flagship-class models with industry-leading CPU and image signal processing performance, according to the company.

The AD1000 chip on its own is rated for 512 TOPS (trillion operations per second) but systems can be designed to scale with up to four of them working in tandem, for up to 2,048 TOPS. Memory bandwidth can also scale up to 204GBps using 256-bit LPDDR5 memory. Support for the PCIe 4.0 bus plus 10-Gigabit Ethernet and CAN-FD standards ensure high-bandwidth, low-latency connections to various other in-vehicle components. A hardware “security island” is implemented for safe encryption, and the chip is fully compliant with the EVITA (E-Safety Vehicle Intrusion Protected Applications) standard. 

The architecture is optimised for AI and vision processing for current and next-gen models. It supports up to 20 camera feeds. The company will be working with its developer ecosystem to support major AI frameworks and software libraries. The new solutions will compete with Nvidia's Drive series both in China and abraod.

As in-vehicle technology gets more sophisticated, SiEngine says its hardware can handle large-scale world models, vision-language-action systems, and complex urban autonomous driving conditions.