Low power AI chip has quantised deep neural network engine
Edge computing capabilities, performance and functionality are expanded in a prototype chip developed by Socionext.
The chip incorporates newly-developed quantised deep neural network (DNN) technology, with a quantised DNN engine, optimised for deep learning inference processing at high speeds with low power consumption.
Today’s edge computing devices are based on conventional, general-purpose graphics processor units (GPUs), which are generally not capable of supporting the growing demand for AI-based processing requirements, such as image recognition and analysis, explains Socionext. Larger devices incur higher costs due to increases in power consumption and heat generation.
Socionext has developed a proprietary architecture based on quantised DNN technology for reducing the parameter and activation bits required for deep learning. The result is improved performance of AI processing along with lower power consumption, reports Socionext. The architecture incorporates bit reduction including one-bit (binary) and two-bit (ternary) in addition to the conventional eight-bit, as well as the company’s original parameter compression technology. This enables a large amount of computation with fewer resources and significantly less amounts of data, adds the company.
In addition, Socionext has developed a novel on-chip memory technology that provides highly efficient data delivery, reducing the need for extensive large capacity on-chip or external memory typically required for deep learning.
Integrating these new technologies, Socionext has prototyped an AI chip with its DNN engine and has confirmed its functionality and performance. The prototype chip achieved object detection by YOLO v3 at 30 frames per second, while consuming less than 5W of power. This is 10 times more efficient than conventional, general-purpose GPUs, reports Socionext. The chip is also equipped with a high-performance, low power, Arm Cortex-A53 quad-core CPU. Unlike other accelerator chips, it can perform the entire AI processing without external processors, notes Socionext.
Socionext has also built a deep learning software development environment. Incorporating TensorFlow as the base framework, this allows developers to perform original, low-bit quantisation-aware training or post-training quantisation. When used with the new chip, users can choose and apply the optimal quantisation technology to various neural networks and execute highly accurate processing.
According to Socionext, the chip will add the most advanced computer vision functionality to small form factor, low power edge devices. Target applications include advanced driver assistance system (ADAS), security camera, and factory automation.
Socionext is currently conducting circuitry fine-tuning and performance optimisation through the evaluation of this prototype chip.
The prototype is a part of a research project on Updatable and Low Power AI-Edge LSI Technology Development, commissioned by the New Energy and Industrial Technology Development Organization (NEDO) of Japan.
The company will continue working on research and development with the partner companies towards the completion of the NEDO-commissioned project, to deliver the AI Edge LSI as the final product.
