Arm introduces IP for machine learned in edge devices

The suite includes scalable machine learning and object detection processors, claimed to enable trillions of machine learning operations per second on mobile devices.

As well as machine learning, the suite is intended to deliver neural network functionality for a new class of device with advanced compute capabilities including object detection.

The initial launch focuses on mobile processors, although the company says future machine learning products will be designed for use in sensors and smart speakers, to mobile and home entertainment.

The processors take on the GPU for compute performance and object identification in machine learning and artificial intelligence projects. They deliver more than 4.6 trillion operations per second. The company reports that there is a further increase in effective throughput in real-world uses through intelligent data management. For thermal and cost-constrained environments the processors have an efficiency of over three trillion operations per second per watt (TOPs/W).

The object detection processor delivers real-time detection with full HD processing at 60 frames per second.

Arm neural networks software, when used alongside the Arm Compute Library and CMSIS-NN, is optimised for neural networks and bridges the gap between TensorFlow, Caffe, and Android NN frameworks, and Arm Cortex CPUs, Arm Mali GPUs, and ML processors.

The IP suite will be available for early preview in April of this year, with general availability in mid-2018.