Der Blätterkatalog benötigt Javascript.
Bitte aktivieren Sie Javascript in Ihren Browser-Einstellungen.
The Blätterkatalog requires Javascript.
Please activate Javascript in your browser settings.
International 2020 Elektronik 17 Microcontroller Arm NN SDK is a good example of a simplified software development an open source tool that runs on Linux and executes inference engines based on models that have already been trained As such it allows ML to be deployed on any architecture able to run Linux such as Cortex-Aprocessors For deployment on Cortex-M MCUs developers can move to CMSIS-NN As part of the CMSIS framework this software library of neural network kernels covers specific functions as shown in the block diagram in Figure 4 These functions are supported by the CMSIS-NN library with the help of the CMSIS-DSP library to implement AI more efficiently in MCUs as shown in Figure 5 TensorFlow Lite is another framework gaining popularity in the embedded sector This deep learning framework comprises tools designed to help port models created using TensorFlow to target architectures with fewer resources such as MCUs It achieves this through optimizing the models using a dedicated converter for example taking 32-bit floating point numbers and turning them into 8-bit integers Part of the effort here involves creating tools that understand what the model is doing well enough to be able to prune parts of the network without impacting accuracy This includes using the science of causality and counterfactuals to discern what may have happened had the outcome of a previous stage been different and then deciding if that would have made any difference to the eventual outcome Clearly the data science behind ML particularly in endpoints is quite intense and for this reason there are more tools and frameworks being developed As an example NXP has developed its eIQ which stands for edge intelligence software environment that integrates with CMSIS-NN TensorFlow Lite OpenCV Computer Vision and the Glow compiler Like others now available the approach taken by NXP with eIQ is to take a pretrained model and modify it for easier or more optimized deployment on resourceconstrained hardware platforms STMicroelectronics has recently introduced its own solution to helping engineers develop AIempowered endpoints based on low power MCUs The STM32 Neural Network Developer Toolbox or STM32Cube AI also takes pretrained neural networks and generates Ccode that can be directly deployed on the companys STM32 MCUs It currently supports IDEs from Arm Keil MDK IAR and System Workbench Figure 4 CMSIS-NN provides a library of functions for implementing AI and ML in Arm Cortex-Mbased MCUs photo Arm www elma com info@elma de VISIT US AT THE EMBEDDED WORLD 25 02 - 27 02 20 Halle1 - Stand 473 WE ARE READY FOR THE FUTURE Customer satisfaction is top priority at Elma Our engineers have by your side for over 30 years in the fields of SYSTEMS ENCLOSURES and ROTARY SWITCHES We advise you - individually and to your needs and projects Contact us! WE ARE READY FOR YOUR PROJECT! Elma_EK-Bus_01_20 pdf S 1 Format 72 00 x 297 00 mm 27 Jan www elma com info@elma de VISIT US AT THE EMBEDDED WORLD 25 02 - 27 02 20 Halle1 - Stand 473 WE ARE READY FOR THE FUTURE Customer satisfaction is a top priority at Elma Our engineers have been by your side for over 30 years in the fields of SYSTEMS ENCLOSURES and ROTARY SWITCHES We advise you - individually and adapted to your needs and projects Contact us! WE ARE READY FOR YOUR PROJECT!