Self-driving cars require artificial-intelligence (AI) support, but such technology is needed well before this type of vehicle becomes the norm. Neural networks are a subset of machine learning (ML) that’s in turn a subset of AI. Though ML algorithms have been used in all sorts of other application areas, few require the safety certifications necessary for use in the automotive space.
NXP’s eIQ Auto platform (see figure), which targets the automotive space, is Automotive SPICE (ASPICE) compliant. ASPICE is a German standard derived from the generic SPICE (Software Performance Improvement and Capability dEtermination) (ISO/IEC 15504) standard.
NXP’s eIQ Auto targets all of NXP’s automotive platforms with its ASPICE-qualified inference engine.
The eIQ Auto platform is built around NXP’s ASPICE-qualified inference engine. This supports models from TensorFlow as well as those compatible with the ONNX interchange format. The system is designed to work in heterogeneous environments. It includes optimized libraries for NXP accelerators, CPUs and GPUs. Among the target platforms is NXP’s S32V234 automotive SoC.
The platform supports all automotive applications, from identifying objects detected via radar, vision, or LiDAR, to monitoring facial expressions of the driver. Deep learning can also be used to improve motor control and battery management.
Developer support ranges from training of models to evaluation and deployment. Optimization tools provide critical performance improvement. This can improve performance by a factor of 10X to 30X over non-optimized performance.