Pushing AI to the "edge"

Big players such as Qualcomm, Huawei, Ericsson, Nokia and others are pushing the 5G network that will enable transferring huge mass of information to the cloud for big data analysis, machine learning and AI while others such as Broadcom, Amazon and Google are working on ultra-fast backbones, infrastructure and processing power that will need to handle everything in the really near future.

In the meantime, other major players such as NXP, ST, Renesas and other Micro Controllers (MCU) and Micro Processors (MPU) manufacturers straggle to move some of the processing power and decision making to the end nodes.

Applications like voice recognizing doorbells, face recognition in small locks, image recognition in small robotic vacuum or pool cleaners and many other battery operated low power IoT devices can't afford massive data transmission over the cellular network and sometime just need an immediate YES/NO decision making on low power budget.

For that, different level of multi-dimension matrix calculation is required, and while an MPU usually incorporates graphic processor that could be used to do exactly that, the requirement for low power and low cost solutions is pushing the MCU manufacturers to implement improved DSP and mathematical co-processors in what used to be "simple" MCUs. These MCU will need to support new applications like the uTensor (TensorFlow Lite), TensorFlow 2.0, Python, and other neural networks based applications.

NXP introduced the i.MXRT cross platform processors family that runs bare metal or Real Time Operating System (RTOS) on ARM Cortex M7 MCU currently running up to 600MHz and 1GHz on soon to be released devices with camera interface and loads of peripherals as well as the ultra-low power LPC55xx family that uses the new ARM Cortex M33 up to 180MHz.

ST Microelectronics also have it's own solutions, starting with the F7 and H7 families with ARM Cortex M7 up to 400MHz with camera interface and integrated huge flash and RAM and families like the L5 with the new ARM Cortex M33 and the new G family with G0 and soon to come G4 with Mathematical co-processing. ST's open source libraries along with their support with MATLAB, Python and hopefully soon also Keras and others allow fast adaptation and implementation of new features and capabilities.

The Japanese giant Renesas also pushes forward the endpoint intelligence technology with new solutions and a full echo-system based on their Synergy platform.

All in all, with the growing market of sensing machines with decision making capabilities, expect to see them in unexpected places, from your homes to the palm of your hand.