NXP has introduced the i.MX 9 applications processors, which integrate dedicated neural processing units (NPUs) to accelerate machine learning (ML).
The i.MX 9 series is NXP’s first implementation of the Arm Ethos U-65 microNPU, which makes it possible to build low cost, highly efficient AI in embedded devices. They also implement NXP’s Energy Flex architecture which can be used by developers to optimise energy efficiency and reduce carbon footprint while extending battery life, says the company.
The i.MX 9 series is intended to expand machine learning capabilities at the edge for performance-intensive systems which require low power connectivity and machine learning acceleration options. Typical uses will be factory and process automation, transportation, robotics and industry 4.0, together with machine learning with time sensitive networking (TSN) technology, they will enhance performance graphics, vision and enable emerging technologies.
Specific i.MX 9 applications processors will also enable advanced machine learning, including multi object, multi-face recognition with inference in milliseconds and voice based system which will recognise natural language and accents. There will also be sequence analysis for gesture recognition and processors will be able to detect anomalies in industrial systems for predictive maintenance and in synthetic sensors for smart homes and other IoT applications.
The first families in this series will be built in 16/12nm FinFET process technology with specific low power optimisation. The i.MX 9’s Energy Flex architecture combines heterogeneous domain processing (independent applications processor and real-time domains with a separate low power, multimedia domain), design techniques, and process technology to maximise performance efficiency, explains NXP.
For example, low power audio use cases can be powered by the real time domain, where audio can run when the rest of the processor is switched off. The real time domain is also suited to industrial applications such as the CAN network which requires a fast boot (less than 100 milliseconds).
Also included in the i.MX 9 processors is the EdgeLock secure enclave. This is a self-managed, autonomous on-die security IP in the SoC architecture. It simplifies the implementation of complex security with autonomous management of security functions, including silicon root of trust, run-time attestation, trust provisioning and SoC secure boot enforcement. There are also crypto services for advanced attack resistance and to simplifying security certifications. The EdgeLock secure enclave intelligently tracks power transitions when end-user applications are running on the device to help prevent new attack surfaces from emerging.
For edge to the cloud security, specific members of the i.MX 9 series will be Microsoft Azure Sphere-certified.
The i.MX9 application processors will find applications in smart homes, smart cities and public safety systems, fleet management, farming and agriculture, consumer audio, healthcare and energy applications where low power connectivity and machine learning acceleration options are necessary.
http://www.nxp.com
