ARM has developed a compute sub-system (CSS) with 16 high performance processors to speed up the development of the next generation of AI chips for automotive designs.
This is the first use of the Zena branding and provides 16 Cortex A720AE automotive processor cores with Cortex-R82AE real time microcontroller cores.
A key difference is that the Mali GPU is optional, as is an AI accelerator, indicating that ARM expects chip makers to add additional IP, whether in the chip as RTL or as a chiplet.
This is marking a more to more AI in software defined vehicles, driving up the processing requirements
“We have over 94% of auto makers using ARM technology and the top 15 silicon suppliers use our technology,” said Dipti Vachani, senior vice president and general manager of the automotive and IoT product lines at ARM.
“All the new entrants in electric vehicles use ARM technology and this allows us to see where the markets are going, from software defined vehicles (SDV) with centralised compute to AI-defined vehicles. What we are seeing going into the future is building on SDV for the AI-defined vehicle with models in the cloud as well as models in the car.”
“What is different is we are seeing AI here in automotive before we see it in other areas,” she said. “This AI-defined vehicle has unique requirements, with increased demand for compute, and this has driven the Zena CSS.”
- ARM to re-brand its core IP
- Faraday licenses its first ARM automotive core
- ARM moves Neoverse V3 core into automotive
The ARMv9 720AE core was launched last year and the cores will sit alongside a safety island with real time processing via the Correx-R82AE cores and a security enclave, which is key to protecting the data in the vehicle. A Runtime Security Engine provides secure over-the-air updates and the system-wide security and root-of-trust is enabled by ARM TrustZone.
There is a higher performance core, the NeoverseV3AE, that is used in the Nvidia Drive Thor chip that is expected to launch shortly.
Future versions of the Cortex-A520AE, A720AE and the NeoverseV3AE are expected to be branded as the Zena Pro, Premium and Ultra cores as part of the overall re-branding of the product lines announced last month.
Using the CSS means fewer engineers are needed for projects, speeding up development times. ”This means 20% fewer engineers per project and 12 month saving, that’s a whole model for our customers,” said Vachani. “This allows customers to focus on the differentiation, whether an OEM or silicon provider or tier one supplier.”
The interface to the GPU or neural processor is based on the UCIe standard which uses PCI Express modulation to simplify the integration. This also allows scalability, with the ability to add several AI accelerators or NPUs to the CSS for high end processing, for example for a self driving vehicle, but use the same underlying chips and software for vehicle models throughout a range.
One key partner, Cadence Design Systems, has developed chiplet designs and IP for the ARM Chiplet System Architecture (CSA) alongside the Automotive Chiplet Program (ACP), a collaborative initiative led by imec in Belgium, where ARM and Cadence play significant roles. Supporting the CSA standard is crucial for widespread industry adoption of the chiplets using UCIe for multi-die chiplet connectivity. The Cadence UCIe IP recently achieved ISO 26262 ASIL-B certification for automotive functional safety for chiplets.
“You can have multiple chips with separate NPUs for scalability in silicon but also in software,” she said. “You can use the exact same Zena CSS and build around it for consistency.”
ARM has worked with Amazon AWS, Cadence Design Systems, Siemens EDA and Synopsys on the tools to run software on the CSS in the cloud to start development long before the silicon is available. Synopsys for example works with over 50 OEMs and Tier 1 suppliers using virtual prototyping technologies and more than 1,000 automotive virtual models developed.
The Synopsys Platform Architect allows automotive hardware and software architects can use early models to explore and predict key indicators such as latency, bandwidth, contention, and utilization of their application workloads executing on specific Zena CSS configurations.
The electronic digital twin includes Virtualizer Development Kit (VDK) for the Zena CSS, to simulate domain-based and zonal E/E architectures. With these eDTs, teams can develop and test software for applications in simulated domains, reducing the need for physical hardware. This approach enables faster software delivery, helps reduce defects, and streamlines hardware/software co-design processes.
It is also working with Vector Informatik on a a SIL Kit open-source library, as well as with Continental, Elektrobit, IPG Automotive, and others to integrate virtual prototyping capabilities in their solutions.
Software is becoming the critical factor for development, and for the AI-defined vehicle, developers can use virtual platforms to validate AI workloads and edge inference behaviour in realistic simulations combined with RTL emulation. The cloud tools also support software testing in the cloud before deployment in the car, accelerating the development and verification of silicon and software.
Helium Studio from Cadence Design Systems works with the Palladium Emulation and Protium Prototyping systems for early software validation using actual SoC hardware on hybrid platforms as designs progress. Helium Studio also enables development with multiple chiplets for automotive and the Scalable Open Architecture for Embedded Edge (SOAFEE). The SOAFEE community is developing Blueprints for digital cockpit and entertainment system designs on the Zena CSS to use AI in the cabin for detection and voice control. These Blueprints use the SystemReady certification and compliance programme with industry standard APIs to work on top of the CSS.
Siemens Digital Industries Software is also adding support for the Zena CSS in its PAVE360 software, designed for software-defined vehicles.
“The era of AI-defined vehicles is an opportunity to bring new in-vehicle experiences to life, but it will require a much faster speed of development and deployment,” said Suraj Gajendra, vice president of automotive products and software solutions, Automotive Line of Business, Arm. “With the help of virtual platform solutions like PAVE360 from Siemens, Arm is enabling our partners to begin software development on Zena CSS before physical silicon is available, significantly reducing development time for new software solutions.”
“Our work with ARM demonstrates that it’s no longer enough that vehicle development is software defined – the process now needs to be systems-aware with the full vehicle system developed in parallel to help ensure that the entire system meets requirements and will require continuous verification,” said David Fritz, vice president, Hybrid and Virtual Systems at Siemens Digital Industries Software. “We enable customers to develop multi-domain digital twins across electronics, hardware and application development for validation and integration that encompasses the whole System-on-a-Chip (SoC), electronics/electrical (E/E) system and vehicle development flow.”
- Cognizant, Siemens team on SDV digital twins
PAVE360, as part of Siemens’ SDV framework, brings together the Innexis software environment, Veloce hardware-assisted verification and validation system, Teamcenter software for Product Lifecycle, Polarion for Application Lifecycle Management (ALM), and Simcenter Prescan and Simcenter Amesim software for simulation to provide a more integrated approach to software-defined development.
This allows developers to use PAVE360 to develop software for Zena CSS before silicon availability, and within the SOAFEE community, the virtual prototyping environment will become a key technology to enable SOAFEE Blueprints. Developers can then functionally validate software in-system and accurately model SoC algorithms and hardware/ software interaction, helping to mitigate the inevitable challenges posed by software-defined and systems-aware vehicle development.
ARM says it is working on extensions to SystemReady for the end of 2025 for operating systems and software stacks for automotive applications working across different hardware.
www.arm.com; www.synopsys.com; www.cadence.com; sw.siemens.com
Leave a comment