In June 2016, Vimicro announced that it has successfully developed China's first embedded neural network processor (NPU) chip and became the world's first embedded video capture device with deep learning artificial intelligence Compression-coded system-on-chip, and named "Starlight Smart One." The depth of learning based on the chip used in face recognition, up to 98% accuracy, more than the human eye recognition rate. The depth of learning based on the chip used in face recognition, up to 98% accuracy, more than the human eye recognition rate. The chip mass production on March 6 this year, the current shipments of more than ten million pieces.
Zhang Yandong, chief executive officer of the lab and CTO of Vital Star Micro, said in an interview that the chip equipped with a neural network processor is used in a surveillance camera. The camera is upgraded from "eyes" to "eyes with a brain," which is For the first time in the world.
"Digital multimedia chip technology" State Key Laboratory was established in 2010, relying on the Beijing Star Microelectronics Co., Ltd., approved by the Ministry of Science and established. According to statistics, Vimicro Microelectronics Co., Ltd. was founded in 1999 by the former State Ministry of Information Industry directly invested in the company specializing in chip technology, "national team", its research and development "Star Series chips" have broken the international market without "China Core "situation.
"Starlight Smart One" is an embedded NPU. Neural Network Processor NPU (Neural Processing Unit) has not been well-known, but it is a hot technology in the field of chips. In contrast to CPU processors in the von Neumann architecture, it uses a disruptive new architecture called "Data Driven Parallel Computing." If the Von Neumann architecture handles data as a single lane, Data-Driven Parallel Computing is a 128-lane multi-lane parallel that can process up to 128 data at a time, making it easy to handle large amounts of multimedia data in video and image formats.
In the industry, the calculation of unit power consumption performance, which is the ratio of performance to power, is used to measure the pros and cons of the processor architecture. According to Zhang Yandong, chief executive officer of the laboratory and chief technology officer of Vimicro Microelectronics, the performance power consumption of "Starlight Smart One" is at least "two or three orders of magnitude" higher than the traditional Von Neumann architecture, which is hundreds of times.
High power consumption is criticized by many top AI technologies. IBM's 20th Century Deep Blue and Google 2016's AlphaGo were powered by massive data calculations, with the former using supercomputers and the latter using server clusters that were unable to separate from the constant temperature and humidity room. AlphaGo need $ 3000 for a game of chess. Zhang Yundong called them "a scientific experiment", far away from the technology and application.
This underscores the miniaturization, low power consumption and low cost benefits of embedded NPUs, accelerating the deployment of artificial intelligence technologies. For example, UAVs have high requirements on the weight and power consumption of the camera, otherwise it will affect take-off and endurance. The "Starlight Smart One" only ordinary stamp size, weighs only tens of grams, it was born so that many small devices such as surveillance cameras have the possibility of artificial intelligence, taken from the mysterious room, artificial intelligence, cross-life applications A step.
Zhang Yundong, respectively, the chip in the neural network processor and video capture, compression coding system, analogy to the brain and eyes. The chip equipped with a neural network processor is used in a surveillance camera, and the camera is upgraded from "eyes" to "eyes with a brain."
In addition to the neural network processor, Yang Xiaodong, director of the State Key Experimental Academic Committee, introduced "Starlight Intelligent One" integrated with the national standard (SVAC) audio and video codecs. This national standard in 2010, developed by Vimicro and the Ministry of Public Security, is the world's first video compilation standard specifically for security surveillance. Zhang Yundong explained that because the video file is too large, only through the coding can be transmitted on the network, while the original standard to borrow abroad, is for radio and television programs, poor security. SVAC solves the problem. After six years, the standard has been used in large-scale in China and is expected to go international.
The marriage of SVAC standards and artificial intelligence analysis technology has good results in security and investigation. In November 2013, a series of explosions occurred near the provincial party committee in Shanxi Province. According to the cameras near the provincial party committee, the police locked the vehicles of the suspects. Through artificial intelligence analysis and analysis, the police cracked the case within 48 hours.
In addition to security, the future, the brain and the eye of both Starlight Smart One will be applied to car driverless or intelligent assistance, UAV automatic shooting, tracking, obstacle avoidance, intelligent robot comprehension of synthetic language, entertainment and escort . "Deep learning chip has a learning process, like a growing child, its training, learning has a process, using a lot of samples as its learning data." Zhang Yun-dong said that as the sample continues to accumulate, the chip The "IQ" will be higher and higher.
The next generation of NPUs will bring surprises? Chip may be more close to the human brain computing mechanism, low power consumption and strong computing power to make a transformative leap, making artificial intelligence in the wearable hardware.
zhangjiagang sevenstars machinery co.,ltd , https://www.zjgsevenstarsmachinery.com