Why it changes
A vehicle from around 2010 could contain 70–100 ECUs, each built for one function and wired to its own sensors and actuators. Every new feature meant a new box, new wires and new supplier software. The harness became one of the heaviest and most expensive parts of the car, and updating software across dozens of suppliers’ ECUs was slow.
Software-defined vehicles need the opposite: few, powerful computers that can be updated as a whole, and a harness that does not have to change every time a function changes.
What the three steps look like
| Step | Idea | Typical result |
|---|---|---|
| Distributed | One ECU per function | Many ECUs, long harness, many CAN buses |
| Domain | One controller per domain (powertrain, chassis, body, ADAS, infotainment) | Fewer ECUs, but domains still span the whole vehicle |
| Zonal + central | Zone controllers per location, functions on central computers | Short harness, Ethernet backbone, software separated from hardware |
How a zonal architecture works
- Central compute (HPC) runs vehicle functions as software services, often split into partitions of different criticality.
- Zone controllers sit in each area of the vehicle (front, rear, left, right). They connect local sensors and actuators, distribute power through electronic fuses, and translate between local buses and the backbone.
- The backbone is switched Ethernet, increasingly with TSN for deterministic timing.
- Local networks stay cheap: CAN FD and classical CAN for MCU-based ECUs, LIN for simple slaves, 10BASE-T1S for smart edge nodes, and hardwired I/O for switches, lamps and simple sensors.
- Safety-critical ECUs (brakes, steering, airbag) usually remain dedicated MCU-based ECUs with their own sensors, because their timing and certification needs are strict.
When it makes sense
Zonal architectures pay off with a new vehicle platform: harness savings and software flexibility outweigh the cost of new zone controllers and central compute. For an existing platform, migration is gradual — one domain or one zone at a time.
Best practices
- Separate where hardware is (zones) from where functions run (central compute) in your models and requirements.
- Keep hard real-time and highest-ASIL control close to the actuator unless the backbone and compute are designed for it.
- Plan power and communication redundancy for fail-operational functions from the start.
- Define clear interfaces between zone and central compute (services or signals) and version them.
- Expect a long mixed period: legacy CAN ECUs and new zonal concepts coexist for years.
Common pitfalls
- Moving functions to central compute without budgeting latency across zones and the backbone.
- Treating zone controllers as “simple I/O” and underestimating their gateway, power and network management load.
- Ignoring variant complexity — especially in trucks and buses with body builders.