Contents

Virtualisation & partitioning on the HPC safety island (MCU)

The safety island runs its own real-time software on dedicated cores, with hardware boundaries that limit interference from the HPC’s application processor.

3 min read · Discipline stake: Safety Hardware BSW Software Update

What runs here

“Physically separate cores” describes the execution boundary: the safety island runs on dedicated real-time cores inside the same system-on-chip (SoC), outside the application processor’s hypervisor. Its AUTOSAR OS or safety RTOS executes directly on those cores.

This cell therefore contains relatively little standalone software. Isolation depends on hardware protection configured during startup and maintained during operation.

Component Role
Dedicated real-time cores Execute safety control and monitoring independently of application-processor scheduling.
Protection configuration Defines which processors and bus masters may access island memory, registers and peripherals.
Local OS protection Separates software within the island where its safety architecture requires this.
Controlled communication boundary Restricts how the application processor exchanges commands, data and status with the island.

Hardware mechanisms vary by SoC. Local memory, interconnect firewalls and fault-containment logic may support this boundary; their exact guarantees must come from the selected device’s documentation. An example of hardware isolation around shared resources appears in TI’s MCU-island architecture description.

Why it looks like this

Safety control needs bounded execution and a defined response to faults. Dedicated cores remove competition with application-processor guests for processor execution time. Local memory can also reduce dependence on shared-memory traffic, although accesses through shared interconnects still require timing analysis.

Lockstep cores execute corresponding instructions and compare results to detect certain hardware faults. They usually form one logical execution resource; they do not provide two independently developed safety applications. Lockstep also cannot detect every software error.

The ASIL D context describes the intended safety requirements, not an automatic property of everything running on the island. The safety argument must establish the required freedom from interference and examine dependent failures, including shared clocks, power, reset, memory and interconnect resources. ISO 26262-9:2018 covers ASIL-oriented and safety-oriented analyses.

Integrating the island can reduce board components and external communication compared with a separate safety MCU, while introducing dependencies within the chip. A small, controlled software baseline also makes changes easier to assess. Application-processor software may evolve more frequently, but its updates still need compatibility and interference checks against the island.

How it connects

Above this boundary sit the island’s real-time control functions, safety monitors and AUTOSAR Classic software. Below it sit the RTOS, processor protection mechanisms and SoC access controls. Within the island, AUTOSAR OS can provide memory and timing protection when supported and configured; these protections complement the hardware boundary. See the AUTOSAR Specification of Operating System.

The application processor commonly communicates through mailboxes, interrupts and shared-memory buffers. Define ownership, message formats, cache handling, timeouts and restart behaviour explicitly. Treat incoming commands as inputs requiring validation, even when both processors occupy the same chip.

Connections to other vehicle nodes use the island’s available communication peripherals or a controlled gateway path. Safety-relevant exchanges need appropriate end-to-end protection and freshness monitoring. Any fault response that must survive application-processor failure needs an actuation or signalling path that remains available during that failure.

Compared with other node classes

  • HPC application processor: A Type-1 hypervisor separates guest operating systems; the safety island has its own execution cores outside those guests.
  • Zone controller and MCU-based ECU: MPU-based partitioning separates software sharing an MCU execution environment. An island can use this internally as well as its separation from the application processor.
  • Smart node / LIN slave: This layer is typically not applicable because the software has no comparable virtual-machine or partition architecture.

Best practices

  • Map all shared resources and document which application-processor faults the island must tolerate.
  • Configure access controls for every relevant bus master, including DMA and accelerators; protect their configuration against unauthorised changes.
  • Verify timing under representative worst-case shared-resource traffic and interrupt load.
  • Specify bounded communication, malformed-message handling and independent restart sequences.
  • Test application-processor hangs, resets and unavailable services while checking the required safe or degraded behaviour.
  • Review island and application-processor updates together for interface compatibility and changed resource usage.

Common pitfalls

  • Equating dedicated cores with complete independence despite shared power, clocks or reset controls.
  • Assuming lockstep detects incorrect algorithms or establishes ASIL D compliance by itself.
  • Protecting CPU accesses while leaving DMA access or shared-buffer ownership uncontrolled.
  • Making safety monitoring or fault reaction depend on a service that fails with the application processor.

What it means for your discipline

  • Functional safety Core

    Defines the required freedom from interference and fault response. Reviews dependent failures and verifies that the island can meet its safety requirements when the application processor misbehaves.

  • Hardware & silicon Core

    Checks the SoC’s memory, interconnect, power, clock and reset boundaries against the safety concept. Establishes which resources remain available during each relevant fault.

  • Base software Partial

    Implements startup protection settings, local OS protection and communication drivers. Ensures configuration and recovery behaviour preserve the intended boundary.

  • HPC & application software Partial

    Designs bounded interfaces between application-processor software and island functions. Handles stale data, rejected commands and independent restarts without breaking safety behaviour.

  • Software update & OTA Partial

    Checks host and island version compatibility when planning updates and rollback. Coordinates activation and restart sequences so programming one side preserves the required isolation and safety supervision.

No direct stake: E/E architecture · Cybersecurity · Networks & communication · Diagnostics & lifecycle

Where it sits in the software stack

Generic layered software of a central HPC (application processor with hypervisor, plus a safety island) and of a classic MCU-based ECU (AUTOSAR Classic). Highlighted modules matter for this topic. Click any module to open its page.

HPC · application processor (MPU) and safety island (MCU)
  • Own cores — Keeps island execution on dedicated cores outside the application processor’s hypervisor, with controlled access to shared resources.
MCU-based ECU · AUTOSAR Classic

Not directly involved on this node type.