Not just for the Cloud AI and ML for Microcontrollers

Maschinelles Lernen wird direkt in den Endgeräten stattfinden.
Machine learning will take place directly in the end devices.

In order to enable short reaction times, ML algorithms should run in the end devices. Alexandra Dopplinger from NXP knows what requirements are involved, both for the microcontrollers and for the tools.

At this year's SPS/IPC/Drives fair, NXP will be highlighting machine learning (ML) at its booth 10.1 - 229. Alexandra Dopplinger, Global Industrial Marketing Manager for NXP's Microcontroller Group, has bundled NXP's ML activities for the booth. She will be at the exhibition in Nuremberg and will inform about ML applications at the booth.


?     Ms. Dopplinger, machine learning (ML) is often equated with processing huge amounts of data that require cloud computing. NXP plans to implement ML applications on end devices in the future. How would this work?

!     Alexandra Dopplinger: Machine learning begins with building inference models from large amounts of data – typically done in the cloud or on a high-performance computer.

These inference models can then be optimized to run on a range of efficient embedded processors, with and without cloud access. NXP customers are already shipping edge-based ML applications such as MCU-based light switches, and MPU-based access controllers.


?     Which NXP microprocessors and controllers are suitable for ML applications, and what tools do developers need?

!     Dopplinger: NXP offers a continuum of microprocessors and controllers suitable for ML applications, including speech recognition, object recognition, face recognition, anomaly detection, and defect detection. The best device for each application is the one that makes decisions with the required level of accuracy, within the required timeframe, and for the best system cost.

After the inference model is created, several tool options exist to optimize their implementation on the appropriate NXP device. NXP’s complementary eIQ software development environment can optimize inference engines to run on the Arm Cortex-M7 cores in i.MX RT crossover MCUs, the Cortex-A cores in i.MX 8 applications processors, the GPUs in i.MX 8QuadMax applications processors, and the neural network units in future devices not yet publicly announced. Partners such as Au-Zone Networks offer commercial tools for even greater capability.


?     Can ML applications be tailored to the resources of a microprocessor? Is porting to other microprocessors possible?

!     Dopplinger: Yes, the ML inference engines can run on hardware engines such as Arm Cortex-A or Cortex-M7 cores, graphics processing units (GPU), digital signal processors (DSP), and neural network units (NNU) inside each device. The NXP eIQ tool supports porting between many NXP microprocessors and controllers. The commercial partner tools support porting between NXP devices, as well as other vendor solutions.


?     Which reference designs for ML does NXP offer specifically? What can developers start with?

!     Dopplinger: NXP offers a variety of sample code, demos and reference designs supporting voice recognition, face recognition, anomaly detection, and object detection. NXP also offers ready to use production grade MCU engines – Hardware plus Software – for Alexa voice (i.MX RT106A), pure edge Local Voice Recognition (i.MX RT106L), and pure edge Face Recognition (i.MX RT106F), applications with included shrink-wrapped production binary software and associated licenses. These off-the-shelf engines are based on the Cortex-M7 core in the i.MX RT1060 crossover MCU.

?     What does NXP show with regard to ML at SPS 2019?

!     Dopplinger: There are NXP-based ML solutions in partner and customer booths, such as the AWS solution in Amazon booth and in the Toradex booth, running on the GPUs in the i.MX 8QuadMax applications processor. And you can meet the experts on the NXP booth to address your interest. Or get more information about the eIQ software for instants at



Alexandra Dopplinger, P. Eng.,

is Global Industrial Marketing Manager for NXP's Microcontroller Group. She holds a patent for a redundant network technology, has more than 16 years of experience in the semiconductor industry and ten years of experience in telecommunications.

After graduating from Memorial University of Newfoundland, St. John's, Canada, with a bachelor's degree in electrical engineering, Dopplinger has held positions ranging from hardware and systems development to product management and marketing. Today, she leverages her extensive knowledge in industrial automation and control to expand NXP's industrial business.