text.skipToContent text.skipToNavigation



Future Electronics


John Robins
EMEA Vertical Segment Manager (Embedded Systems)
Future Electronics


The latest trend in embedded processing will be front and center at the Embedded World exhibition (Nuremberg, Germany, 26-28 February 2019): all the key players in our industry are investing in machine learning and Artificial Intelligence (AI) developments.

AI is not in fact a new technology: the term ‘machine learning’ was coined in 1959 by IBM’s Arthur Samuel. Initially its application had been mainly limited to demonstrations, such as Samuels’ own machine for playing the game of checkers. It is finally, however, becoming accessible to electronics product developers in all market segments.

Its use in autonomous vehicles attracts the widest publicity today. But AI offers exciting opportunities to improve the operation of, or add valuable features to, many kinds of end products. For instance, the addition of speech recognition can make a human-machine interface touch-free – potentially useful in hazardous or space-constrained operating environments. An ultra-low power FPGA, such as QuickLogic’s EOS S3, as shown on page 11, is the ideal hardware platform for such an application.

Similarly, the addition of a trained camera to access-control or security systems enables it not only to detect the presence of a person, but also to indicate a person’s age and identity for security purposes, or to apply age-defined comfort settings. Future Electronics’ Avalanche development board, which is based on a mid-range PolarFire FPGA from Microchip, can recognise up to 20 different objects such as birds and cars.

Why now, after nearly 60 years, is the rate of adoption of AI suddenly accelerating? Two changes have occurred: first, processors and FPGAs have become powerful enough to train neural networks. According to Dr Shane Legg, a co-founder of Google’s DeepMind subsidiary, a training run that takes one day on a current tensor processing unit – an AI accelerator chip developed by Google for neural network machine learning – would take 250 million years on an 80486 microprocessor from 1990.

Second, there is now a vast volume of data available to generate models. Estimates suggest that 90% of the data available today was generated in the past two years.

Technologies supporting the use of AI in mainstream industrial and consumer systems will be open for viewing and evaluation at the Future Electronics stand 3.225 at Embedded World. For instance, a Future Electronics example project, for identifying street signs, shows how microcontrollers can accelerate machine learning through the use of Arm’s CMSIS-NN, a collection of efficient neural network kernels which maximise the performance and minimise the memory footprint of neural networks on Cortex®-M processor cores.

Predictive maintenance demonstrations will also be on view, based on technologies from STMicroelectronics and QuickLogic.

Real-world applications of AI and machine learning can be implemented today, and your local branch of Future Electronics will be pleased to tell you more about the technologies and demonstrations available from franchised suppliers, or call 1.800.FUTURE.1 for help.


FTM NA SideNav SubscribeTile EN
FTM NA Issue2-2019 SideNav Download