This article is produced by NetEase intelligent studio (public number smartman 163). Focus on AI, read the next big time!
Wen / Yi
On the afternoon of May 16, a six-year-old AI company, Yunzhi Sound, held a press conference in Beijing to officially launch the first AI series chip for the Internet of things, UniOne. The following intelligent bacteria take you to take a look at the first generation of UniOne chips
UniOne is the chip brand of Yunzhi Sound, which is the general name of a series of chips. The first generation of chips released today is
Released this time
In the aspect of architecture flexibility, the main control CPU is connected with the AI accelerator internal RAM through Scratch-Pad, providing an efficient data channel between the CPU and the AI accelerator, so that CPU handles the results of the AI accelerator operation two times.
That is to say, UniOne provides product commonalities, such as far-field speech recognition and standardized components. However, the source code is open to personalized parts related to scenarios, product usage, and so on. This ensures high performance, high energy efficiency ratio and flexibility and versatility.
From the software level, swift provides a AI extension instruction set for deep neural networks. Li Xiaohan said that the time-consuming AI algorithm was abstracted and implemented in the new parallel architecture.
In power mode design, Yu Yan provides a multi-stage power mode from the static and static testing (2% of the chip energy consumption) to the live detection (10% of the energy consumption of the chip), to the target detection (20% of the energy consumption of the chip), and then to the normal mode. The power consumption is greatly reduced.
In addition, the cloud will not only provide chips, but also provide integrated cloud core product level solutions, docking AI cloud services, AI software solutions, and chip factories. At this conference, Mr Yun also announced strategic cooperation with the two companies. At the meeting, Yun Zhi announced that it was working with the Jingdong Alpha platform to build a customized intelligent benchmarking product, and jointly developed the AI chip of the car pre loading regulation level jointly with the technology of hundreds of millions of caffeine, and constructed a cockpit interactive platform with voice as the core.
AI chip R & D Road: grinding four years, first verify the rationality of the market.
On the other hand, the access of AI algorithm puts forward higher requirements for the parallel computing power and storage bandwidth of the device end chip, and although the traditional chip based on GPU can implement the reasoning algorithm at the terminal, it is not designed for the depth learning design, and the energy efficiency is far lower than the AI core chip. Therefore, the AI chip is more specialized in design, and has absolute advantages over the general chip in computing density and power consumption.
Huang Wei said that a AI chip must have three prerequisites, the full AI technology capability, the proven application scene, the chip design ability, and the cloud knowledge has been laid out for four years on the AI chip. In these four years, Yun Zhi sound has been based on the product form of IVM (general chip program) in the market of home, smart sound box and child robot, which verifies the rationality of the market, product and user scene.
Huang Wei said that Yun Zhi had set up a chip team as early as 2015, and in 2016 the company began to assess the market, products, technical routes and downstream partners. In 2017, the chip product definition, IP selection, algorithm optimization, tool preparation, and detailed product definition and technology module evaluation were launched.
Chinese power protruding in the field of AI chip
In the field of AI chips, the latest research report has been published by market research company Compass Intelligence, and the industry's demand for AI chips is also increasing, and more than 1700 start-ups are interested in AI chips. In the past three years, the top 24 AI chip companies have invested up to $60 billion in the field of AI.
Nvidia, Intel (Intel) and IBM (NPU) ranked the top 24 in the top 24 list of the world's top three, with Chinese companies holding seven seats and the highest ranking of twelfth HUAWEI, and the other six companies HiSilicon, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, MediaTe, and MediaTe. K), Imagination, Ruixin (Rockchip), core (Verisilcon), Cambrian (Cambricon) and horizon (Horizon).
Cloud knows how to get into the AI chip, and focuses more on application scenarios and solutions than other vendors. Coupled with the recent completion of the $100 million C round of cloud financing, it can be said that the market prospect is good and the supplies are abundant. In the future, whether cloud awareness can occupy a place in the field of AI chips remains to be tested by the market.