IN Brief:
- ECARX and Arm signed a non-binding agreement to explore high-performance computing platforms for vehicles and other physical AI systems.
- The planned co-engineering covers processor architecture, performance, software tools, functional safety, and cybersecurity validation.
- Their existing work through SiEngine includes the production 7nm SE1000 and development of a 5nm SE2000 platform.
ECARX and Arm have agreed to examine high-performance AI computing platforms for vehicles and other autonomous equipment. Their non-binding memorandum of understanding sets out joint architecture and software work, extending an established relationship in automotive semiconductors. The planned effort covers system definition, performance optimisation, functional safety, development tools, and cybersecurity validation; it does not yet establish specifications for a commercial processor.
Engineering teams will explore how processing hardware, software and safety functions can be combined in configurations suitable for different markets. China and international vehicle programmes may require different software stacks, supplier interfaces, and validation arrangements, making early architecture choices significant. ECARX and Arm envisage joint research and product definition under the memorandum, with project deliverables and launch dates still to be established.
London-headquartered ECARX develops automotive computing systems for digital cockpits, connectivity, and intelligent driving. Arm supplies processor architecture and a substantial development ecosystem. The proposed collaboration brings these capabilities together around systems in which high-performance inference must coexist with communications, real-time vehicle functions, and software that may be updated over the product’s operating life.
The partners have experience turning an earlier architecture collaboration into silicon through SiEngine. Its 7nm SE1000 automotive-grade system-on-chip entered volume production in March 2023, and ECARX says millions of units have been produced and deployed. SiEngine is also developing a 5nm SE2000 successor. These are existing programmes; the new memorandum concerns further platform research rather than a relaunch of either chip.
More vehicle functions are being consolidated from separate electronic control units into central computers and domain controllers. That approach reduces the number of independent computing nodes but increases demands on shared memory bandwidth, peripheral interfaces, power distribution, and isolation between workloads. An in-vehicle display, for instance, has different timing and failure requirements from a function that monitors a driving system, even when both depend on the same computing assembly.
Adding AI workloads makes those boundaries more demanding. Sensor streams must be captured and transferred to processing engines, inference results delivered to other software, and conventional control and diagnostics maintained while computing demand varies. The underlying architecture consequently has to account for scheduling, data movement, and fault response alongside peak throughput. The companies’ decision to include performance optimisation and functional safety in early co-engineering reflects those linked design tasks.
Functional safety concerns the detection of faults and the defined response when hardware or software does not operate as intended. Cybersecurity, meanwhile, concerns access to software, interfaces, and data and the integrity of communications across connected systems. Work on both can affect hardware partitioning, secure boot and update processes, monitoring functions, and the evidence needed before a platform is adopted by a vehicle maker.
Software enablement and toolchain validation are part of the agreed scope because an automotive computing architecture must be usable by the teams developing the finished vehicle. Drivers, middleware, compilers, debugging facilities, and test environments determine whether applications can exercise the hardware predictably. Early testing of shared resources can also identify contention between inference pipelines, communications, and other processing tasks before product design becomes fixed.
ECARX and Arm are also examining applications beyond vehicles, including other forms of physical AI. Industrial robots and autonomous machines likewise connect sensing, inference, and physical actuation, but their operating cycles, communications, and safety requirements differ from automotive systems. A common computing approach would therefore need application-specific integration and validation rather than a single interchangeable design.
Ziyu Shen, ECARX’s founder and chief executive, said the companies would pursue a joint technical roadmap and co-engineering to develop solutions for international markets. Drew Henry, Arm’s executive vice-president for physical AI, identified the combined demands of computing, software, and safety as a focus of the collaboration. The next concrete steps are to define platform configurations, evaluate the supporting software, and establish how proposed designs will be validated for their intended deployments.
