Kneron's core technology is to develop a high-efficiency, low-power neural network unit (NPU) that transfers artificial intelligence from the cloud to terminal equipment for real-time identification and judgment analysis, and expands the application of artificial intelligence. Unlimited possibilities at different levels. Kenneth founder and CEO Liu Juncheng stated: 'We are very honored to have Mr. Li Ka-shing's support and are confident of becoming the next company to lead the industry in innovation and change.'
Liu Juncheng revealed that it will launch a revolutionary 3D artificial intelligence solution in the second half of the year: 'Kneron provides industry-leading technology, including a complete terminal artificial intelligence solution, artificial intelligence integrating video and voice, and the introduction of revolutionary 3D labor in the second half of the year. Smart Solutions. The new funds will be used to accelerate product development and, through cooperation with strategic partners, deepen vertical industry applications, including smart homes, smart security, and smartphones.
The most important feature of off-line artificial intelligence is security, high-speed, and the key to popularization of artificial intelligence applications. Jonathan Tam of Victoria Harbor Investment stated: 'Kneron's breakthrough lies in the seamless integration of hardware and software. We believe that Kneron will continue to develop its terminal labor in the future. Intelligent technology and deepening applications, bring AI to everyone's life.
The founder of Kneron Liu Juncheng told Xinzhi that this financing will be mainly used to expand the layout of the Asia-Pacific region. Kneron's current team members have a group of overseas Tsinghua electronics department members who will form the backbone of Kneron Greater China. Team leaders have With more than 20 years of work experience, there are also leading pilots of VP ranks of international companies such as former Qualcomm and Huawei.
Liu Juncheng is a UCLA Ph.D. who has participated in R&D projects of NASA JPL, IARPA, and Bell Labs. He also worked at Samsung R&D Center and Qualcomm.
Liu Juncheng stated that Kneron's core competitiveness lies in the integration of hardware and software, with its ultra-low power consumption NPU (lowest power consumption below 100mw) and its lossless compression technology. Liu Juncheng said: 'Using our restructuring technology ( (patents have been patented), most multi-dimensional matrix techniques can be made with a less-dimensioned, more streamlined architecture. For example, a 5×5 convolution can be achieved by stacking two 3×3 filters with low-dimensional operations, 7×7 convolution You can stack three.'
Kneron has mass-produced AI chips and is mainly used in mobile phones, security, and smart homes. Kneron is one of the few companies in the world with AI chips NPU in 2017 and has achieved significant revenue. Recently, Kneron and mobile phone chip supply chain The key producers, together with the main accelerator chip in the specific scene of the mobile phone.
Specifically, Kneron NPU IP is a dedicated artificial intelligence processor designed for terminal equipment. The terminal equipment can run ResNet, YOLO and other deep learning networks in offline environment. Kneron NPU is a complete terminal artificial intelligence hardware solution. , including hardware IP, Compiler, and model compression. It can support various mainstream Convolutional Neur al Networks (CNN) models, such as Resnet-18, Resnet-34, Vgg16, GoogleNet, and Lenet, as well as supporting mainstream deep learning frameworks, including Caffe, Keras, and TensorFlow.
The Kneron NPU IP consumes 100mw (milliwatts) and the ultra-low power version of the KDP 300 (the main mobile 3D face recognition application) is even less than 5mw. The full range of products have a performance of 1.5 TOPS/W or more per watt. Using a number of exclusive technologies, it is possible to achieve the requirements of low power consumption and high computational power. In the architecture design, the convolution kernel partitions the large convolution kernel convolution operation block into a convolution kernel decomposition (Filter decomposition) technique. Multiple small convolutional operation blocks are respectively operated, and then combined with Reconfigurable Convolution Acceleration technology, the operation results of multiple small convolutional operation blocks are merged to accelerate the overall computing performance. By Kneron Advanced compression technology can compress unoptimized models by tens of times. Multi-level caching can reduce the CPU resources and data transfer, further improving overall operational efficiency.
In addition, Kneron NPU IP can be combined with Kneron image recognition software to provide real-time identification analysis, fast response, not only more stable, but also to meet the security and privacy requirements. Because the hardware and software can be tightly integrated, the overall scheme is smaller and the power consumption is lower. To assist in the rapid development of products.
Liu Juncheng said: 'To perform artificial intelligence operations on the terminal equipment while meeting power consumption and performance requirements is the primary consideration. The Kneron NPU IP has achieved such a goal and brought revolutionary development to the terminal artificial intelligence. Kneron has been launched since 2016. After the company's first artificial intelligence processor NPU IP dedicated to terminal equipment, it continuously improved its design and specifications, and optimized for different industrial applications.
Liu Juncheng stated that Kneron focuses on immediate revenue applications, mainly smart security, smart homes and mobile phones. The 3D artificial intelligence solution to be launched in the second half of this year will focus on high-precision identification, such as Face ID applications on mobile phones. 3D manual The specific applications of the smart program include face-to-face payment for mobile phones, face-to-face payment for new retails, high-precision smart door locks for replacement of keys on smart cars, gesture recognition for human-computer interaction, and high-precision quality control for industrial Internet of Things.
Kneron focused on the terminal AI chip from its inception. When asked whether there is any plan for the development of cloud AI chips in the future, Liu Juncheng told Xinzhi that the cloud chips can be technically implemented, but it needs commercial opportunities. Kneron team The DNA has always been in the terminal, and with the large number of cloud companies in the world, their ecological layout is also relatively complete. It is unrealistic for small companies to go directly to the PK ecology of large companies. However, there are many applications on the terminal, and new players also have a foothold. Opportunity.
Kneron's customers include international manufacturers of mobile phone supply chain chips, first-tier manufacturers in the field of home appliances, etc. The ultra-low-power NPU is not only successful in tapeout, but also a series of NPU AI chip products have achieved large-scale production and created millions. Dollar income.