US deep learning and edge technology company Kneron has unveiled a range of low-power artificial intelligence (AI) processors.
According to the company, the devices each consume less than 5mW of energy, unlike most similar processors that require several watts to function.
The Kneron NPU IP Series consists of the KDP 300 ultra-low power version, the KDP 500 standard version, and the KDP 700 high-performance version.
The new processors are designed to run AI applications for smartphones, and for smart-home, security, and other IoT devices.
With them, users are able to operate deep learning networks, such as ResNet and YOLO on edge devices, said the company.
Read more: MIT develops low-power high-speed chips for IoT security
According to Kneron, they offer “complete hardware solutions for edge AI, including hardware IP, compiler, and model compression”.
The processors work with neural network models such as Resnet-18, Resnet-34, Vgg16, GoogleNet, and Lenet.
The new chips use so-called decomposition technology to “divide a large-scale convolutional computing block into a number of smaller ones to compute in parallel”.
Kneron explained: “Together with the reconfigurable convolution accelerating technology, the computing results from the small blocks will be integrated to achieve better overall computing performance.
“Through Kneron’s advanced model compression technology, the size of the unoptimised models can now be shrunk a few dozen times. The multi-level caching technique reduces the use of CPU resources and further improves overall operational efficiency.”
The NPU IP- KDP 300 enables 3D live facial recognition, while the NPU IP- KDP 500 can be used for analysis and deep learning. Finally, the NPU IP- KDP 700 high-performance chip can handle more advanced AI computing.
Albert Liu, founder and CEO of the company, said: “To run AI computation on edge devices, the biggest challenge is to fulfil the performance while keeping the power consumption low.
“Kneron’s NPU IP Series provides high performance at ultra-low power, which is a revolutionary breakthrough in edge AI.”
Read more: ARM launches scalable chips for IoT machine learning
Read more: Why you could soon have a neural network on your smartphone
Internet of Business says
The edge environment is increasingly important to the spread and success of the IoT, as more and more critical decisions will need to be made, and functions carried out, at speed and at or near the point of origin. As ever greater intelligence and computing power is pushed into edge networks and devices, speeds need to increase, and power consumption needs to fall – along with costs.
Kneron’s ongoing innovation in this space is part of this sector-wide trend, along with the development of low-power security chips and hardware that is increasingly optimised for AI and machine learning applications.
IoTBuild is coming to San Francisco, CA on March 27 & 28, 2018 – Sign up to learn all you need to know about building an IoT ecosystem.