The Future of the Automotive E/E Architecture Is Mixed-Criticality Compute

September 2022 saw automotive product announcements from NVIDIA and Qualcomm, two leading System-on-Chip (SoC) suppliers that are leveraging the technologies and capabilities refined in the cloud, data center, PC, and mobile industries to address the needs of a rapidly transforming automotive market. While there are important differences in the automotive strategies of both companies, a clearly identifiable trend, common to both NVIDIA GTC and Qualcomm’s Automotive Investor Day, was mixed-criticality compute.

NVIDIA DRIVE Thor: The NVIDIA DRIVE Thor was announced as a replacement for the anticipated NVIDIA DRIVE Atlan, delivering twice the anticipated performance of the replaced Atlan at 2,000 Trillion Operations per Second (TOPS)/2,000 Trillion Floating Point Operations per Second (TFLOPS). This extraordinary headroom, combined with innovations, such as Multi-Instance GPU (MIG), will allow Original Equipment Manufacturer (OEM) customers to configure DRIVE Thor to simultaneously run multiple domains, including digital cockpit and autonomous driving applications.

Qualcomm Snapdragon Ride Flex: At its automotive investor event, Qualcomm announced its “graduation day” in automotive, detailing the products and pipeline that will see the SoC giant expand out of telematics and deeper into the digital cockpit and autonomous driving domains. Alongside design wins and launches for Snapdragon Cockpit and Snapdragon Ride, Qualcomm also announced Snapdragon Flex, a mixed-criticality compute platform that will consolidate multiple domains, including digital cockpit and autonomous driving into a single platform. While more details on the capabilities of Snapdragon Flex are expected in 2023, Qualcomm was able to confirm existing OEM customers for the forthcoming SoC.

The typical automotive Electrical/Electronic (E/E) architecture has become bloated and unscalable, with each new functionality requiring the addition of a new Electronic Control Unit (ECU) to a flat topology defined by aging, low bandwidth protocols. This results in cars that are heavier (weighed down numerous ECUs and a complex wiring harness), lengthy design cycles (with each ECU delivered by a fragmented network of Tier One and Tier Two suppliers), and a static vehicle functionality (as the Controller Area Network (CAN) bus chokes Over-the-Air (OTA) update potential). Therefore, automakers and their Tier One and Tier Two suppliers are actively pursuing compute consolidation, leveraging more capable hardware and virtualization/hypervisors to define more of the vehicle architecture in software and reduce weight and complexity, and deliver the headroom necessary to accommodate lifecycle maintenance.

To date, consolidation has been limited to domain controller-dominated architectures, whereby ECUs delivering similar applications of the same level of safety criticality are consolidated onto a more powerful domain controller. This is due to automotive design practice regarding mixed-criticality compute—a malfunction in a non-safety-critical application must never be allowed to interfere with the operation of a safety-critical application, requiring adequate isolation mechanisms between applications of differing criticality. OEMs have largely depended on separate ECUs or separate domain controllers to deliver the isolation required, but with the support of SoCs, such as DRIVE Thor and Snapdragon Flex, automakers have a feasible path to take compute consolidation to the next level.

Most automotive technology trends follow a trickle-down trajectory. A novel technology is first adopted on, for example, an Audi model, before proliferating throughout that brand into the Volkswagen brand, and then into the cost-sensitive SEAT and Škoda lineup, with growing scale bringing the necessary price reductions to proliferate into the higher-volume market segments. However, market demand for rapid and extensive compute consolidation is coming from all corners of the market—from mass market and premium OEMs alike.

For premium OEMs, the primary motivation for domain consolidation is the opportunity to exploit software-driven revenue and to explore the development of applications that straddle the currently isolated automotive domains. In contrast, mass-market OEMs are looking for the most cost-effective way to satisfy the burgeoning consumer demand for rich infotainment experiences, while also delivering on often mandatory active safety applications, such as Intelligent Speed Assistance (ISA) and Driver Monitoring Systems (DMS).

Therefore, it is essential that mechanisms for effective isolation of safety-critical and non-safety-critical applications are not restricted to top-end SoCs targeted at fully unsupervised automation, but that SoC suppliers opt for a modular approach, satisfying the demand at the low end for single platform fulfillment of Advanced Driver-Assistance Systems (ADAS) and digital cockpit applications.