Intel shifts to chiplets for automotive AI

Intel has made a significant advancement in utilizing chiplets for its second generation software-defined vehicle system-on-chip (SoC) device. Unveiled at the Shanghai 2025 Auto Show, this SoC integrates chiplets constructed on various process technologies to offer extensive language model AI support, enhanced graphics capabilities for user interfaces, and 12 camera channels for ADAS driver assistance systems.

While Intel has shared some performance comparisons with a Core i7 processor without specific figures, more technical details regarding the chiplets within the device are expected to be disclosed in the future. This development places Intel and its partners notably ahead of other chip suppliers in the automotive industry.

One of the key advantages of the chiplet approach is the ability to produce different chiplets using diverse technologies. For instance, a 3nm digital processor could be utilized for AI or self-driving functionalities, while an optimized high-speed analog process could be employed for camera interfaces.

Earlier this year, Intel introduced an Adaptive Control Unit (ACU) as an alternative to a microcontroller, designed to circumvent dataflow bottlenecks in real-time, safety-critical automotive designs. The ACU features a flexible logic area that offloads real-time control algorithms from the CPU core.

Additionally, Intel has developed the second generation of its standalone B-series Arc GPU for Automotive, slated for production by the end of 2025. The company has also inked strategic software agreements with ModelBest in Beijing and Black Sesame Technologies in Wuhan to operate on the SDV SoC.

Black Sesame Technologies is working on a central compute platform that integrates ADAS and immersive cockpit functionalities with its autonomous driving technology. This platform combines the SDV chip with Intel’s discrete Arc graphics chip, with the Black Sesame HuaShan-2 A1000 chip being adopted by numerous Chinese automakers for autonomous driving applications.

ModelBest has developed a GUI Intelligent Agent that enables on-device LLM operation with the SDV SoC and Arc graphics chip. This facilitates offline, AI-enhanced voice control and personalized interaction without network connectivity, enhancing voice interaction by accurately interpreting natural language in complex scenarios for an intuitive cockpit experience.

Jack Weast, Intel Fellow, vice president, and general manager of Intel Automotive, expressed, “Intel is redefining automotive compute with our second-generation SDV SoC, combining the flexibility of chiplet technology with our proven whole-vehicle approach. Together with our partners, we’re addressing real industry challenges — from energy efficiency to AI-driven experiences — to bring the software-defined vehicle revolution to fruition for all.”