Neural-network execution layer for safety-critical embedded systems
AICORX
Deploy, control and record neural-network execution across embedded platforms through one consistent interface.
AICORX sits between neural-network applications and the target platform, with engineering controls, traceability and supporting evidence designed for safety-critical development and certification. For aerospace, defence and automotive programmes under DO-178C/ED-12C, ED-324/ARP6983 and ISO 26262.
- For
- Aerospace, defence and automotive programmes
- Certification context
- DO-178C/ED-12C · ED-324/ARP6983 · ISO 26262
- Availability
- Design-partner programme
Overview
Neural networks are entering airborne, defence and vehicle functions whose surrounding software is developed and certified under DO-178C or ISO 26262. Changes to the model, data, silicon or operating environment can introduce new verification and certification work.
AICORX introduces a consistent execution layer between neural-network applications and supported target platforms. It helps programmes maintain control, traceability and verification as models and deployment platforms evolve, without forcing a new hardware or operating-system architecture.
What AICORX changes for a programme
Reduce repeated integration
Use one consistent interface as models and supported target platforms evolve.
Designed for certification
Engineering controls, traceability and supporting evidence are designed in from the start for safety-critical development and certification.
Simplify platform migration
Reduce engineering effort when moving between supported hardware and operating environments.
Bring neural-network execution under control
Apply defined execution, monitoring and lifecycle controls appropriate for mission-critical development.
Build evidence into the development lifecycle
Structured runtime records support verification and certification activities as the programme progresses.
Protect model independence
Integrate proprietary and third-party models through the same interface. Your models remain yours and are not tied to a particular platform.
Target platforms & integration
| Category | Automotive | Aerospace & Defence |
|---|---|---|
| Processors / SoCs | TI Jacinto TDA4 · NVIDIA DRIVE · Qualcomm Snapdragon Ride · NXP automotive SoCs | Arm-based SoCs · NXP Power Architecture · NVIDIA Jetson · AMD Xilinx adaptive SoCs |
| Operating systems | QNX · Embedded Linux · other RTOS | VxWorks and VxWorks 653 · QNX · Embedded Linux · other RTOS |
| Inference environments | TensorRT · TI TIDL · ONNX Runtime · Vendor-specific runtimes | TensorRT · ONNX Runtime · Vendor-specific runtimes |
| Integration interfaces | Native C/C++ · POSIX · DDS · AUTOSAR Adaptive | Native C/C++ · POSIX · DDS · ARINC 653 |
| Certification context | ISO 26262 | DO-178C/ED-12C · ED-324/ARP6983 |
AICORX is designed for integration across heterogeneous embedded platforms. Platform availability, integration requirements and certification scope are assessed for each target configuration through the design-partner programme.
Who it is for
- Aerospace and defence programme integrators working under DO-178C/ED-12C and ED-324/ARP6983.
- Engineering and functional-safety teams bringing neural-network functions into ISO 26262 scope.
- Programmes facing silicon second-sourcing, obsolescence or imported-to-indigenous migration.
Evaluation
AICORX is offered through a design-partner programme. Architecture, platform support and target-hardware demonstrations are shared under NDA.
Tell us your target hardware, operating environment and certification requirements so we can assess the appropriate AICORX evaluation path.