- Dspace has announced the launch of Neural Net Coder, a software solution designed to automate the integration of neural networks into embedded systems.
- This tool generates production-ready C code directly from trained artificial intelligence (AI) models, independent of the original development environment used during the training phase.
Manually implementing neural networks is often time-consuming and error-prone when deploying to target hardware. The automated process delivers deterministic code with predictable execution behavior, meeting the requirements of resource-constrained control units. Regarding software safety, the generated code complies with MISRA coding guidelines—a critical criterion for safety-critical industrial and automotive applications.
To validate the generated code’s compliance, the tool incorporates an automated verification protocol using “back-to-back” testing. These comparative tests confirm that the generated C code maintains strict functional equivalence with the original neural network, ensuring result transparency for certification processes.
Hardware Resource Optimization and Real-Time Applications
The tool offers post-training optimization functions that act directly on the code structure without requiring the AI model to be retrained. Developers can adjust parameters to reduce memory footprint and required computing power, making it easier to adapt the application to the constraints of various target hardware platforms. Furthermore, the ability to estimate memory and execution time requirements during the early design stages helps minimize iteration loops throughout the project.
Typical use cases for this technology include the deployment of virtual sensors—for instance, to calculate battery state-of-charge or to evaluate physical parameters that are difficult to measure directly. The tool enables the use of AI in settings subject to real-time and safety constraints.
In terms of software integration, the system is designed to interface with existing code-based toolchains as well as model-based development environments, such as the Dspace TargetLink platform.
