Cell-based architecture is gaining increasing attention, particularly in automotive software-defined approaches, due to its inherent fault resilience. A clear parallel exists between vehicle domains—currently distributed across numerous electronic control units (ECUs)—and cells in a cell-based architecture. This analogy remains valid when applied to a centralized vehicle compute architecture. Each domain must provide fault boundaries, ensuring that failures in one domain (or cell) do not affect others. For example, a failure in the infotainment system should not impact the vehicle’s safety-critical ADAS system.
Each domain manages its own workloads, much like a cell in cloud architecture handles its application workloads. Domains vary in their reliance on the control plane, which dynamically allocates resources such as processing power, memory, and I/O bandwidth, and manages functions like deployment, task prioritisation, and calibration. The data plane, as seen in common software-defined architectures, handles integration between domains (or cells). For instance, in autonomous driving, data from cameras, radars, and LIDAR systems must be routed to a central ADAS processing unit for real-time decision-making, while motion management, which processes smaller amounts of data, works in parallel. This data plane must be robust and continue functioning even if the control plane fails, ensuring that critical systems like steering, braking, and safety features remain operational, as guided by the cell partitioning strategy.
The effectiveness of failure management largely depends on domain (or cell) sizing. Domain sizing plays a crucial role in the overall robustness of the system. Smaller, more isolated domains—similar to smaller cells in IT systems—limit the impact of a failure to a narrower segment of the vehicle’s functionality, while larger domains benefit from economies of scale in processing, resource, and capability utilization. For example, dividing ADAS functions (such as perception, decision-making, and actuation) into separate domains can limit failure scope. However, overly small domains may lead to inefficient resource usage or difficulties in managing complex or high-volume data in real-time.
In the automotive sector, electrical and electronic (EE) architecture faces constraints due to physical resource limitations, such as connectivity bandwidth and access between physically separated zones. Future vehicle features will demand even more robust and scalable domain management.