Chinese semiconductor industry

The launch ceremony of Tianxin Micro's first advanced process silicon-based epitaxy equipment was successfully held​

The analysis hits the crux of most of China's industries....

Alibaba/T-head made a couple of announcement. The first one we already talked about earlier is the unveiling of C908 which is supposed to be extremely energy efficient for the computation it does. 20% better than comparable
It can be used in smart communication, multimedia terminal, AR/VR, wireless communication and other scenarios.
Looks like Alibaba/T-head is really betting on continued development of RISC-V

Alibaba sees itself as a global leader in RISC-V and will continue it's integration into Alibaba Cloud and PC also (remember they came out with TH1520 a month ago).

Just as interestingly is what they said about Yitian-710 and Cipu
It's already been used in Alibaba's own operations as well as in the cloud by other companies.
In the next 2 years, 20% of new computing power capacity on Alibaba cloud will be using Yitian-710. The cost performance of computing power has been improved by over 30%, and the power consumption per unit has been reduced by 60%. Quite impressive.

There was a story recently that Samsung is pausing production of this, but that doesn't seem to be the case here. Looks like Alibaba is confident about its development. Given that it is not in the banned list, I don't see any problem why it cannot continue to be produced as an alternative to Intel and ARM chips.

According to

Please, Log in or Register to view URLs content!

, some well-known Chinese enterprises in scientific computing, the smartphone industry and Internet have significantly improved their cost performance after migrating to the Yitian 710 cloud instance.

Click to expand...

Alibaba also came out with CIPU (cloud centered computing system) earlier this year. Apparently, CIPU is showing 20% improvement in computing power over comparable platforms.

I think the only reason processors like the server CPUs made by Alibaba have not been sanctioned yet is because they are not for sale on the market. They are only used internally by Alibaba itself. But I would not be 100% certain the US won't just block their production eventually.

gelgoog said:

I think the only reason processors like the server CPUs made by Alibaba have not been sanctioned yet is because they are not for sale on the market. They are only used internally by Alibaba itself. But I would not be 100% certain the US won't just block their production eventually.

Click to expand...

They're used internally but not really. Alibaba Cloud is a cloud service provider and their servers and CPUs are used by all kinds of customers. It's just customers pay for utilization but not ownership. Grounds for ban, yes, for the same logic by the US if applied consistently. On the other hand, Alibaba Cloud is also a big customer for Intel, so that is also something to consider for them too.

Yizhu Technology focuses on domestic storage and computing integrated AI chips with large computing power, and the 28nm process achieves 10 times the energy efficiency ratio

"The key resources in the digital age are data, computing power and algorithms. Data is the new means of production, computing power is the new productivity, and algorithms are the new production relations. The three constitute the most basic production cornerstone in the digital economy era."

In September 2021 , in the "China Computing Power Development Index White Paper" released by the China Academy of Information and Communications Technology, the above conclusions were used to emphasize the fundamental role of computing power in the digital economy era. In February this year, a number of ministries and commissions jointly issued a notice agreeing to start the construction of national computing power hub nodes in 8 places including Beijing-Tianjin-Hebei, Yangtze River Delta, Guangdong-Hong Kong-Macao Greater Bay Area, Chengdu-Chongqing, Inner Mongolia, Guizhou, Gansu, and Ningxia , and planned 10 A national data center cluster, which marks the official launch of the national integrated big data center system - the "East Number and West Calculation" project.

This major project, known as the "South-to-North Water Diversion", "West-East Electricity Transmission" and "West-East Gas Transmission" in the era of digital economy, has attracted great attention from the whole society. Behind the hot spots, when the ultra-large-scale deep learning model with hundreds of billions of parameters becomes a reality, guiding the entire artificial intelligence industry to find a new feasible way and shortening the distance between us and general intelligence, the huge computing power caused by the massive data The demand has led to an objective situation at present: the development of computing power has never been able to keep up with the development of algorithms, which poses new challenges to the field of chip semiconductors.

How to achieve " low power consumption, high precision, and large computing power " has become the foundation and core of the sustainable development of the global digital economy in the "post-Moore era".

In fact, when the United States on the other side of the ocean blocked my country's chip semiconductor industry step by step, and hindered China's development of advanced semiconductor industry by continuously intervening in China's purchase of lithography machines and forming a "chip alliance", my country has been vigorously promoting the situation to get rid of the situation of chips being controlled by others. Industrial innovation. First , on August 9 , the United States passed the " Chip and Science Act of 2022 ", and then on August 31 , chip giants Nvidia and AMD received requests from the U.S. government and were restricted from exporting the latest flagship GPU computing chips and boards to China. A group of semiconductor companies with domestic independent research and development capabilities have actively or passively entered the public eye.

Founded in 2020 , Yizhu Technology is one of them, which takes the integrated AI computing power chip technology of storage and computing as a breakthrough . With Moore's Law approaching the physical limit, Yizhu Technology tried to break through the von Neumann bottleneck through architectural innovation, becoming an innovator in China's AI chip entrepreneurial army, and also solving the problem of domestic AI computing power, especially large computing power. Bureau offers a new direction.

Beyond the "Moore Era", Yizhu focuses on the AI computing power chip with integrated storage and computing architecture

As early as 1992 , Xu Juyan, an academician of the Chinese Academy of Engineering, predicted that after 2014 , the semiconductor industry will encounter an inflection point and enter the "post-Moore era".

This prediction is based on the fact that under the traditional von Neumann computing system using a separate architecture for storage and computing, 80%-90% of the power consumption occurs in data transmission, and 99% of the time is consumed in the memory read and write process. The proportion of energy consumption and time actually used for computing is very low, resulting in the emergence of "energy consumption wall" and "storage wall".

Dr. Xiong Dapeng, chairman and CEO , said: "Ezhu Technology has applied ReRAM to the integrated storage-computing AI high-power chip in a fully digital way. The advantages of doing so are: First, the integrated storage-computing architecture can break the traditional Von Neumann architecture. The storage and calculation wall, energy consumption wall and compilation wall under the architecture; Second, the storage medium ReRAM has an order of magnitude advantage in the main indicators such as computing power potential, computing power accuracy and computing power efficiency, and is currently the most suitable for storage and computing integrated AI computing . The memristor of the force chip; third, the fully digital technology path can support high precision while satisfying large computing power, so that the integrated storage and computing architecture can truly be implemented in the direction of AI large computing power. Therefore, Yizhu Technology can provide The industry brings AI inference computing solutions with large computing power, super energy efficiency ratio, low power consumption, and easy deployment ."

He emphasized that ReRAM , a new type of memristor, has a series of advantages such as non-volatility, small area, high density, low cost, low power consumption, and fast read and write speed . Technology has undoubtedly reached the "eve of the outbreak of commercial applications." In addition, ReRAM is compatible with the CMOS process. Whether it is the development of its own density, or through the evolution of the process and the use of 3D stacking technology, it can continue to improve the density and energy efficiency ratio of ReRAM from multiple dimensions. Yizhu believes that Yizhu's improvement in product computing power and energy efficiency ratio is a first-class level in China.

"The first generation of Yizhu's products can achieve the same computing power and 10 times the energy efficiency ratio based on the 28nm process , and at the same time have lower software ecological compatibility and construction costs, and the selected memristor has a lot of room for growth in the future, these are the support Yizhu's future development momentum and an important factor of certainty."

As AI scenarios become more and more popular and diversified, algorithm models become more and more complex, and society's demand for computing power is far beyond our imagination. According to the "China Computing Power Development Index White Paper", in 2020 , the total global computing power will reach 429EFlops , with a growth rate of 39% . According to IDC 's forecast, the number of global IoT devices will exceed 40 billion in 2025 , and the amount of data generated will be close to 80ZB . It is estimated that the global computing power scale will grow by more than 50% in the next five years, and the overall scale will reach 3300EFlops by 2025 . Among them, China 's AI chip market in 2021 will exceed 40 billion yuan, of which AI inference computing will account for more than 50% , and the CAGR is expected to exceed 40% in the next five years .

However, the country's strategic goals for carbon peaking and carbon neutrality also require that AI computing power is constantly rising, and at the same time, it must also achieve a rapid increase in the energy efficiency ratio . National requirements for energy structural optimization. Dr. Xiong Dapeng believes that under the background that traditional heterogeneous AI chips are facing the technical bottleneck of the process Moor wall, the algorithm model is increasingly complex and huge, and the energy consumption usage supervision is becoming more and more strict, the storage-computing integrated AI chip has a rare historical development opportunity. This also makes the advanced computing architecture of integrated storage and computing a cutting-edge research hotspot, and integrated storage and computing and AI large computing power chips are about to usher in an opportunity for the industry to explode.

Yizhu's fully digital storage and computing integration technology can practically apply and implement the storage and computing integration architecture on chip platforms with large computing power and high energy efficiency ratio . In part, the area and energy consumption of the chip are used for data calculation itself, so as to achieve multi-dimensional satisfaction of large computing power and high precision

cont.....

With the support of a strong technical team and the core nodes of the domestic industry chain, Yizhu will launch its first-generation chips in the first half of 2023 , and cast the second-generation chips in the same year .

As a technology track with many technical fields, high integration and strong innovation, the AI large computing power chip with integrated storage and computing architecture requires senior practitioners in the industry and excellent cutting-edge scientists.

The R&D capabilities of the Yizhu team cover the entire chain of storage devices, storage and computing arrays, chip architecture, chip design, software ecology, AI algorithms and engineering implementation. The R&D team has published more than 40 papers at the top conference. Experience in the field of high-end integrated circuit design; also has experience in the design, mass production and sales of 20+ SoC chips . Therefore, the team has foresight in technical positioning and technical strategy, and also has the confidence to explore the integrated technology of storage and computing.

As the basic means of production for the informatization, digitization and intelligent development of various industries in human society, the development of chips is rooted in the requirements of the informatization, digitization and intelligent development of human society. The reason for competing for the support of head chip companies. In the field of AI chips, the demand for computing power will continue to develop at a relatively high speed for a long period of time in the future, which has become the consensus of the entire industry.

But from another perspective, challenges and opportunities coexist in the current field of storage and computing integration. In 2021 , China's chip imports will exceed 430 billion US dollars. As China's local factories expand production capacity, China's purchase orders for overseas chip manufacturing equipment increased by 58% last year . If breakthroughs cannot be achieved through advanced technology, China's dependence on overseas chip semiconductor industry Will continue to increase, the current situation is urgent.

In Yizhu's view, under the premise that advanced technology cannot fully realize the localization of the entire chain in the short term, domestic semiconductor companies must have the technology to decouple from advanced technology, and at the same time realize chip products based on advanced technology. Equivalent or even better performance . In this regard, Yizhu’s AI high computing power chip based on the traditional 28nm CMOS process and the existing domestic industry supporting design can achieve the same computing power as the AI chip with advanced process technology but 10 energy efficiency ratio.

In addition, Yizhu's goal is not only to put the storage-computing integrated architecture into commercial use in the field of AI large computing power, but also hope to build the storage-computing integration together with other entrepreneurial partners who are working together on the storage-computing integration track.

Since the development of chip semiconductor field, it has become a highly coupled industry in the whole chain, and all links are interdependent and indispensable. However, through architectural innovation and based on localized industrial facilities, Yizhu will provide the industry with large computing power, ultra-high energy efficiency ratio, and easy-to-deploy AI large computing power chip products, without changing the physical space of the data center and existing infrastructure . Under the premise of greatly improving the computing power density and energy efficiency ratio, it contributes its own strength to the intelligent development of the domestic industry.

In the next two years, 20% of Alibaba Cloud's new computing power will use self-developed CPUs

gelgoog said:

GPU cores are much simpler to design than CPU cores. The main problem is making the drivers and custom hardware I think. The software stack you need to have for a GPU today is just huge.

China has better CPU designers than the ones at Zhaoxin. Unfortunately Zhaoxin mostly rehash already poor CPU designs made by US Centaur. It is a bit of a waste of their X86 license they got from VIA which will likely expire soon if it did not already.

Click to expand...

Moore Threads basically has released the same number of discrete GPUs as Intel in 2022
This is very impressive to me, not just technologically, but operationally as well. Even if the number they sell are only in the dozens, it is not a small achievement.

weig2000 said:

They're used internally but not really. Alibaba Cloud is a cloud service provider and their servers and CPUs are used by all kinds of customers. It's just customers pay for utilization but not ownership. Grounds for ban, yes, for the same logic by the US if applied consistently. On the other hand, Alibaba Cloud is also a big customer for Intel, so that is also something to consider for them too.

Click to expand...

The stock ownership of many of these early Chinese companies like alibaba, baidu, tencent are mostly compose of western or global funds/banks/investors, not Chinese. I am guessing that Xi prob doesn't care about wacking them up a bit with some of these with various "common prosperity" laws and fines as it would other local startups a opportunity to enter the same space which creates more jobs.