How to follow the edge calculation of artificial i

  • Detail

How is the process of "core making" in China? Artificial intelligence core making closely follows the edge computing tuyere news of CNR technology on March 28 (Jasman) last year, "core making" became a major artery of China's AI technology in 2018 with the investment of industry, capital, technology and other aspects in the field of AI chips. Chinese technology companies such as Baidu, Alibaba, Tencent, Huawei, Cambrian, horizon and yunzhisheng have experienced a wave of "core making fever"

after one year, how is the process of Chinese enterprises' independent research and development of chips? What are the opportunities for China core to break the game

"at present, only the application software of domestic chips is ideal compared with the world's advanced level, but the raw materials and equipment, manufacturing, design, system software and other aspects are different from the international advanced level." Wang Yu, a professor of the Department of electronic engineering of Tsinghua University, pointed out at the 2019 Xinzhiyuan AI technology summit that there is a large gap in CPU (central processing unit) in terms of chip design, and the market is occupied by old brand manufacturers such as Intel; With the rise of horizon, Cambrian and other chip companies, the gap in AI chip design is small; The field of new devices is in the same starting line. As a member of the academic community, Wang Yu also recognizes the importance and problems of cooperation between academic and industrial circles, and calls for building a semiconductor innovation ecosystem by reading the guide to the development of new material industry

in addition, in reality, solef PVDF series membrane materials applied to filter membranes have improved the reliable, high-performance and profitable use required by products. Data centers still need stronger and faster training capabilities, which are the work absorbed by specimen deformation and fracture, while "ai+" urgently needs reasoning capabilities from terminal to cloud, which is the basis for large-scale investment and competition among technology giants

the diversity of the location, size, cost, power consumption requirements of application deployment AI capabilities and the requirements of rapid integration with other computing capabilities put forward higher-level requirements for AI computing. As a chip giant, Intel, on the one hand, is exploring quantum computing and neural pseudo computing, on the other hand, it is also exploring super heterogeneous computing forms. In the future, whether it is cloud or terminal, it will be the world of AI super chips

China AI core making closely follows the edge computing tuyere

with the gradual increase of data generated by the network edge and the development of IOT, many new computing models are being put forward. For example, if data can be processed and analyzed at network edge nodes, this computing model will be more efficient than cloud computing

for IOT, the edge computing technology has made a breakthrough, which means that many controls will be realized through local devices without being handed over to the cloud, and the processing will be completed at the local edge computing layer. This will undoubtedly greatly improve the processing efficiency and reduce the load on the cloud. Because it is closer to users, it can also provide users with faster response and solve their needs at the edge

in March this year, the new actions of the technology giant attracted the attention of the industry. Google released a thousand yuan development board equipped with edge TPU chip, which can perform reasoning faster than any other processor architecture; NVIDIA also released an edge computing product, the Jetson nano artificial intelligence computer, on the occasion of the tenth anniversary of GTC, for only $99; Horizon, an AI chip company focusing on edge computing, raised US $600million with a valuation of more than US $3billion

from large technology companies to start-ups, it seems that they are on the edge of edge computing. This chip war has entered the end from the cloud, seizing the edge. What are the trends of AI core and intelligent cloud in the next decade

with Intel, NVIDIA and arm companies occupying the data center and chip market, Chinese companies try to break through from the edge and terminal, take advantage of China's huge manufacturing hardware industry chain and scene advantages, and try to establish their own AI chip ecosystem

at the same time, the architecture of aiot (artificial intelligence IOT) is also evolving, and the transformation from centralized and cloud to architecture is progressing smoothly. The neat layer associated with the edge architecture will evolve into a more unstructured architecture, including various devices and services connected in the dynamic grid. These structured structures will achieve a more flexible, intelligent and responsive IOT system

in addition, in the face of the two practical problems of AI chips, which are constantly evolving algorithms and closely related to algorithms and applications, the chip architecture that can take into account flexibility and high energy efficiency is under constant exploration. Dr. lixiaohan, vice president of yunzhisheng, told us that the new aiot chip (artificial intelligence IOT chip) is needed for the structure and operation method of IOT combined with artificial intelligence concrete pressure testing machine, and multimodal AI chip is the key step

he pointed out that the superposition of AI and IOT requires the transformation of traditional solutions in five directions: from general architecture to AI architecture, from hardware dependence to software hardware integration, from single-mode to multi-mode interaction, and from equipment independence to collaboration

lixiaohan said that the chip design for artificial intelligence is faced with four major challenges: fragmented scene, von Neumann memory wall, low power consumption requirements of edge side applications and security requirements. In terms of chip design, it is required to face specific scenarios, provide multimodal processing capability based on the idea of end cloud interaction, achieve an excellent balance in performance, power consumption and area, and take into account the needs of connection and security

"to solve the common problem of von Neumann memory wall in the industry, the key is to shorten the distance between computing units and storage units, and make them as close as possible. Heterogeneous computing system architecture design, accelerator proximity storage structure, and multiple solutions from general API (Application Programming Interface) functions to special instruction sets can further solve this problem." Li Xiaohan said so



Copyright © 2011 JIN SHI