Chinese semiconductor industry

AF-1 said:

I think it should have been obvious for years, that relying on western technology to create something powerful, to rely technology/economic growth using western tech, is not possible anymore.
Who has ambitions to advance to leading levels, to compete with world top players, must posses its own tech, a complete supply chain...

Creating AI more advanced than western, using western tech to achieve that... come on, thats not reasonable thinking.

Click to expand...

Put your shoes in a private company that lives or dies by the quality of your equipment. Would you rather put your faith in established players, or would you try something entirely new that could end up good or bad for your company since it is unproven?

This is why an embargo is a boon for many domestic players. Other firms have no choice but to purchase their equipment. There will be blood and toil in the beginning, but eventually it will lead to a positive feedback loop of continual product enhancements.

siegecrossbow said:

Put your shoes in a private company that lives or dies by the quality of your equipment. Would you rather put your faith in established players, or would you try something entirely new that could end up good or bad for your company since it is unproven?

This is why an embargo is a boon for many domestic players. Other firms have no choice but to purchase their equipment. There will be blood and toil in the beginning, but eventually it will lead to a positive feedback loop of continual product enhancements.

Click to expand...

We have comrade Trump to thank for witnessing pages and pages of three digit revenue growth in Chinese SME companies in this thread.

Now we have comrade Biden to thank for witnessing what will be pages and pages of three digit revenue growth in Chinese GPU companies.

In fact, the fine folks at Cambricon have probably already started celebrating.

siegecrossbow said:

This is why an embargo is a boon for many domestic players. Other firms have no choice but to purchase their equipment.

Click to expand...

This is why Trump is now called Comrade Trump and is due for a promotion in this Party Congress and Biden is also increasingly becoming likely to become a CPC member

China's greatest friend has always been America

siegecrossbow said:

Put your shoes in a private company that lives or dies by the quality of your equipment. Would you rather put your faith in established players, or would you try something entirely new that could end up good or bad for your company since it is unproven?

This is why an embargo is a boon for many domestic players. Other firms have no choice but to purchase their equipment. There will be blood and toil in the beginning, but eventually it will lead to a positive feedback loop of continual product enhancements.

Click to expand...

Precisely, in the past enterprises believe that no one would be fired if they choose Nvidia. Choosing local components probably still work but some bean counters showed that it will reduce profit by X% and they hated that.

Cambrian: MLU370-X8 has achieved in-depth cooperation with leading Internet companies, and shipped over one million edge chips​

According to the micro-net news, at the beginning of this year, Cambrian released the new training accelerator card MLU370-X8, and it quickly made waves in the domestic Internet, server and other markets. MLU370-X8 is equipped with dual-chip four-core Siyuan 370, integrated Cambrian MLU-Link™ multi-core interconnection technology, and completed the adaptation work with domestic mainstream server manufacturers at the beginning of its release.

Recently, Cambricon disclosed in an investor survey that after the release of the MLU370-X8 accelerator card, with its excellent product competitiveness, it has achieved in-depth cooperation with some leading Internet customers in some scenarios. The company has formed a certain income scale. In addition, some customers have completed product introduction and are in business contact. In the financial field, the company has in-depth exchanges on OCR and other related businesses and product applications with leading banks and well-known enterprises, and has conducted in-depth technical exchanges on new business scenarios (such as natural language processing, etc.), and some enterprises are conducting business trials. In terms of server manufacturers, the company's products have also been recognized by head server manufacturers.

It is understood that the MLU370-X8 provides twice the memory bandwidth of the Siyuan 370, combined with the MLUarch03 architecture and MLU-Link multi-core interconnection technology, to give full play to the advantages of the Siyuan 370 chip in training tasks. MLU370-X8 is positioned at the mid-to-high end, and it is combined with high-end training products Siyuan 290 and Xuansi 1000, which further enriches Cambrian’s training computing power delivery methods; -S4 smart accelerator cards work together to form a complete cloud training and reasoning product portfolio.

At present, Cambrian has mastered the intelligent processor micro-architecture, intelligent processor instruction set, SoC chip design, processor chip functional verification, advanced process physical design, chip packaging design and mass production testing, hardware system design in the field of intelligent chips. In the field of basic system software technology, he has mastered programming framework adaptation and optimization, smart chip programming language, smart chip compiler, smart chip math library, smart chip virtualization software, smart chip core driver, cloud edge Seven categories of core technologies, including the terminal integrated development environment. Among them, the two major technologies of processor micro-architecture and instruction set belong to the lowest core technology. At present, the company's fifth-generation intelligent processor micro-architecture and fifth-generation intelligent processor instruction set are all under development.

At the same time, Cambrian continues to promote the optimization and iteration of the basic system software platform. For the reasoning software platform, Cambrian continues to improve the reasoning acceleration engine MagicMind and its surrounding ecology. Functionally, MagicMind continues to add basic features such as sub-channel quantization and weight hot update. The number of operators supported has been expanded to more than 200. The number of public Benchmark models and non-public private models has continued to increase. The latest version has fully supported image classification and video understanding. , semantic segmentation, similarity detection, text detection, OCR, speech processing, natural language processing, search, recommendation and other fields of cloud-side-end reasoning business.


As Coreteks said in the video I posted earlier, GPUs are one method to execute AI workloads but not the only method, what Nvidia offers is really good software ecosystem that makes easy to develop for their GPUs, like CUDA for example. But what is going to happen when others developing frameworks for AI workloads in different more efficient AI hardware architectures start to catch up with Nvidia? That could spell doom for Nvidia in AI.

Shengmei Shanghai’s net profit in the first half of the year increased by 164% year-on-year, and it plans to invest in high-end semiconductor equipment to expand research and development projects​

In the evening of August 7th, Shengmei Shanghai released its 2022 semi-annual report. From January to June 2022, the company achieved operating income of 1.096 billion yuan, a year-on-year increase of 75.21%, and the net profit attributable to shareholders of the listed company was 237 million. Yuan, an increase of 163.83% year-on-year, and earnings per share of 0.5500 yuan.

Shengmei Shanghai said that this is mainly due to the global semiconductor industry boom and the strong market demand for the company's semiconductor equipment. The company's sales orders and production capacity have continued to grow, and operating income has further increased; Growth has diversified the revenue structure and expanded the market size. From January to June 2022, Shengmei Shanghai's operating income of semiconductor cleaning equipment, front-end semiconductor electroplating equipment and advanced packaging wet-process equipment (including back-end electroplating equipment) have increased significantly.

The main products of Shengmei Shanghai include semiconductor cleaning equipment, semiconductor electroplating equipment, vertical furnace tube equipment and advanced packaging wet equipment. Continuous innovation in performance, production capacity, improving customer product yield and reducing customer costs. At present, through differentiated innovation and competition, it has successfully developed the world's first SAPS/TEBO megasonic cleaning technology and single-chip combined tank cleaning technology. At present, the company's semiconductor cleaning equipment is mainly used in the cleaning process in the field of 12-inch wafer manufacturing, and there is no competitive gap with similar products from international giant companies in terms of the applicable size of semiconductor cleaning equipment.

Shengmei Shanghai expects that, benefiting from the mainland's policy support for the integrated circuit industry and the increase in local demand, the company's main customers will still be in the stage of continuous production expansion in the next few years, which will be able to drive semiconductor manufacturing equipment including cleaning equipment and copper plating equipment. Demand maintained steady growth.

On the same day, Shengmei Shanghai also announced that it plans to invest in the construction of high-end semiconductor equipment expansion research and development projects. The new project name is the high-end semiconductor equipment expansion research and development project. According to the disclosure, the investment amount and source of funds: the total planned investment of the project is 748 million yuan, of which the site investment is 468 million yuan, the purchase and installation cost of research and development equipment is 20.703 million yuan, and the research and development cost is 260 million yuan. Among the investment funds of the project, Shengmei Shanghai used its own funds to invest 16.8591 million yuan, and the remaining 730 million yuan is planned to be invested in the construction with the super-raised funds from the company's initial listing.

"This project will expand the research and development of high-end semiconductor equipment on the basis of the company's original products, technologies and customer resources, create a multi-product matrix, consolidate the company's market position, and enhance profitability." Shengmei Shanghai disclosed that the products to be developed in this project include: The products in the expansion field of dry process equipment and supercritical CO2 cleaning and drying equipment have high standards, and some performance indicators are expected to reach the international leading level. The company's strong technical reserves can effectively reduce the difficulty of project research and development and ensure the smooth implementation of the project.

VeriSilicon's net profit in the first half of the year was 14.82 million yuan, and the amount of newly signed orders was 1.499 billion yuan.​

On the evening of August 7, VeriSilicon disclosed its 2022 semi-annual report. During the reporting period, the company achieved revenue of 1.212 billion yuan, a year-on-year increase of 38.87%; net profit attributable to the parent was 14.82 million yuan, and a loss of 45.65 million yuan in the same period last year. Yuan; basic earnings per share was 0.03 yuan.

Specifically, in the first half of 2022, VeriSilicon achieved an operating income of 1.212 billion yuan, a year-on-year increase of 38.87%, of which the semiconductor IP licensing business (including intellectual property licensing fee income and royalty income) increased by 70.61% year-on-year. The stand-alone chip customization business (including chip design business income and mass production business income) increased by 25.27% year-on-year; the company achieved operating income of 652 million yuan in the second quarter of 2022, a year-on-year increase of 20.54%.

VeriSilicon has GPU IP, NPU IP, VPU IP, DSP IP, ISP IP, Display Processor IP six types of processor IP, more than 1,400 digital-analog hybrid IP and RF IP for integrated circuit design. During the reporting period, VeriSilicon has gradually formed a series of excellent IP and IP in the four fields of AIoT, wearable devices, automotive electronics and data centers based on its own technology, resources, customer accumulation, and market development trends.

During the reporting period, VeriSilicon joined the UCIe industry alliance to strengthen the company's layout in the fields of tablet computers, data centers and autonomous driving, and further lay a solid foundation for the company's chiplet technology development.

At the same time, it continuously enhances its chip design service capabilities and expands its first-mover advantage in FD-SOI technology. During the reporting period, VeriSilicon has successfully taped out chips at 14nm/10nm/7nm/5nm FinFET and 28nm/22nm FD-SOI process nodes in advanced semiconductor process nodes.

In addition, VeriSilicon's chip design process has obtained ISO 26262 automotive functional safety management system certification, and automotive electronics is one of the company's fast-growing application areas. VeriSilicon disclosed that with the important milestone of the company's chip design process obtaining the ISO 26262 automotive functional safety management system certification, it will further expand the company's competitive advantage in this application field in the future.

The advanced synthetic quartz project of Inner Mongolia Hengxin Quartz Co., Ltd. started construction​

On August 27, Inner Mongolia Hengxin Quartz Co., Ltd. held a groundbreaking ceremony for the advanced synthetic quartz project in Dalat Economic Development Zone. The start of construction of the project will inject new vitality and new kinetic energy into the clustering and intensification of the whole flag's new material industry.

It is understood that Hengtong Group is a national high-tech, innovative enterprise and an international industrial group in the fields of China's communication optical network, smart grid, new energy and new materials. After 30 years of development, its industrial bases are located in 13 domestic provinces and 11 overseas. 192 of the top 500 Chinese companies in 2021, with a cumulative sales of 138 billion yuan in 2021. The green low-carbon precision quartz material project invested for the first time in Dalat Banner has a total investment of 500 million yuan. The company will rely on Hengtong's global industrial layout, combine the industrial characteristics of Dalat Banner, practice the concept of circular development of industry complementarity, and focus on semiconductors. The R&D and manufacturing of high-purity quartz materials in the fields of chips, high-end optics, high-energy lasers, etc., while realizing the development of its own enterprise, break the international monopoly, realize domestic substitution, and strive to become the mainstream supplier of high-purity quartz materials in the world.

okay, now I've gone through the Biren Technology website. There are some interesting stuff there as I read through their entire list of news.

目前，壁砺104已经对部分用户开放了邀测，即将量产出货；创全球算力纪录的海玄OAM服务器目前正在进行紧锣密鼓的内部测试，预计今年第四季度开放邀测

Click to expand...

Says here that BR104 has already sent out invite to some customers for sampling. Will start mass producing very soon. The much larger 海玄OAM (which is an 8-way AI server developed with Inspur) will be sampling in Q4. Also said in there that they hope in 10 years, China will be able to take over half of the leading edge GPU market.

According to Tom's hardware, China Mobile and Baidu are among the initial customers. By all measures, it looks like BR100 is cutting edge and competitive with even H100.

I looked at who else they might partner up with. Looks like they are very involved with Shanghai government, getting frequent visit from them. They are doing a lot of research with Qinghua university, as well as Fudan and Zhejiang.

Looks like they also signed an agreement with UCloud back a year ago (about the same time they signed the partnership with China Mobile), so I assume they are another initial client. UCloud as you can imagine is a cloud service provider in China.
They also partnered with Ping An Technology a few weeks ago. This is the technology arm of Ping An Group. They are heavily invested in AI and probably have their own large data center. This seems like another huge partner.
They also signed a deal with Inspur electronic information a month to build the 8-way AI server.

people that showed up at BR100 announcement. A lot of politicians there both from Shanghai and national level. Also included in there are Ping An technology CEO, Inspur Group vice chair + their AI&HPC line CEO, Baidu PaddlePaddle (which is the Baidu cloud's AI framework) training chip technology head.

上海市闵行区委书记陈宇剑，闵行区委副书记、区长陈华文，上海市经济和信息化委员会一级巡视员傅新华，中国电子工业标准化技术协会理事长、中国工业和信息化部科技司原司长胡燕，上海临港经济发展（集团）有限公司党委书记、董事长袁国华，闵行区副区长李锐，临港浦江国际科技城党委书记、董事长施决兵以及闵行区、浦江镇相关部门领导；平安科技董事长兼CEO黄宇翔，浪潮信息副总裁、AI&HPC产品线总经理刘军，百度飞桨训练芯片适配技术负责人李琦等产业界嘉宾；壁仞科技的投资机构代表和新闻媒体代表，共同见证壁仞科技首款通用GPU芯片正式发布。

Click to expand...

So big player here seem to be Baidu PaddlePaddle, Ping An Technology, China Mobile and Inspur electronic information and UCloud. They have quite a few large players in the AI space that are ready to work with them. I don't think it's fair to say that Chinese companies are not buying local. They've started investing in local AI chip producers since 2018/2019. It's just taken a few years for the product to arrive on the market.

I would venture to say that with the A100 sanction, more Chinese tech companies are going to now reach out to Biren Technology and other domestic players. They don't know which one will necessarily succeed. I think they are getting involved with multiple domestic firms to make sure they are covered.

tokenanalyst said:

Cambrian: MLU370-X8 has achieved in-depth cooperation with leading Internet companies, and shipped over one million edge chips​

According to the micro-net news, at the beginning of this year, Cambrian released the new training accelerator card MLU370-X8, and it quickly made waves in the domestic Internet, server and other markets. MLU370-X8 is equipped with dual-chip four-core Siyuan 370, integrated Cambrian MLU-Link™ multi-core interconnection technology, and completed the adaptation work with domestic mainstream server manufacturers at the beginning of its release.

Recently, Cambricon disclosed in an investor survey that after the release of the MLU370-X8 accelerator card, with its excellent product competitiveness, it has achieved in-depth cooperation with some leading Internet customers in some scenarios. The company has formed a certain income scale. In addition, some customers have completed product introduction and are in business contact. In the financial field, the company has in-depth exchanges on OCR and other related businesses and product applications with leading banks and well-known enterprises, and has conducted in-depth technical exchanges on new business scenarios (such as natural language processing, etc.), and some enterprises are conducting business trials. In terms of server manufacturers, the company's products have also been recognized by head server manufacturers.

It is understood that the MLU370-X8 provides twice the memory bandwidth of the Siyuan 370, combined with the MLUarch03 architecture and MLU-Link multi-core interconnection technology, to give full play to the advantages of the Siyuan 370 chip in training tasks. MLU370-X8 is positioned at the mid-to-high end, and it is combined with high-end training products Siyuan 290 and Xuansi 1000, which further enriches Cambrian’s training computing power delivery methods; -S4 smart accelerator cards work together to form a complete cloud training and reasoning product portfolio.

At present, Cambrian has mastered the intelligent processor micro-architecture, intelligent processor instruction set, SoC chip design, processor chip functional verification, advanced process physical design, chip packaging design and mass production testing, hardware system design in the field of intelligent chips. In the field of basic system software technology, he has mastered programming framework adaptation and optimization, smart chip programming language, smart chip compiler, smart chip math library, smart chip virtualization software, smart chip core driver, cloud edge Seven categories of core technologies, including the terminal integrated development environment. Among them, the two major technologies of processor micro-architecture and instruction set belong to the lowest core technology. At present, the company's fifth-generation intelligent processor micro-architecture and fifth-generation intelligent processor instruction set are all under development.

At the same time, Cambrian continues to promote the optimization and iteration of the basic system software platform. For the reasoning software platform, Cambrian continues to improve the reasoning acceleration engine MagicMind and its surrounding ecology. Functionally, MagicMind continues to add basic features such as sub-channel quantization and weight hot update. The number of operators supported has been expanded to more than 200. The number of public Benchmark models and non-public private models has continued to increase. The latest version has fully supported image classification and video understanding. , semantic segmentation, similarity detection, text detection, OCR, speech processing, natural language processing, search, recommendation and other fields of cloud-side-end reasoning business.

Please, Log in or Register to view URLs content!

As Coreteks said in the video I posted earlier, GPUs are one method to execute AI workloads but not the only method, what Nvidia offers is really good software ecosystem that makes easy to develop for their GPUs, like CUDA for example. But what is going to happen when others developing frameworks for AI workloads in different more efficient AI hardware architectures start to catch up with Nvidia? That could spell doom for Nvidia in AI.

Click to expand...

Is it Cambricon or Cambrian? Is it an error due to the translation software or did the write get it wrong?