What's new

Opinionated - China Chipping Away to Semiconductor Dominance

AMD Epyc 7742 is a 64-core CPU with a 2.25GHz base clock, 3.4GHz max clock, and 225W TDP at 7nm.

Compare that to the Kunpeng 920 from Huawei.


64-core CPU is for high-performance computing, enterprise workloads, artificial intelligence and cloud-native apps.

But Huawei also has 4/8 core CPU for desktop.

MdhdQug.jpg
Cores and thread, even systems, gives flat performance beyond a particular numbers. So 64 doesn't mean good as well.
 
.
Cores and thread, even systems, gives flat performance beyond a particular numbers. So 64 doesn't mean good as well.

It all depends on what you're using the CPU for. For example, each of the top 4 supercomputers on the TOP500 list have over a million cores each. Sunway Taihulight has over 10 million cores.

But what you're saying is correct for desktop purposes.

Anyway, let me show the difference between Intel's 14nm vs Huawei's 7nm.

Intel Xeon Platinum 9282 is a 56-core server microprocessor operating at 2.6 GHz base, 3.8 GHz turbo boost, with a TDP of 400W at 14nm. Launched in April 2019.

Huawei Kunpeng 920 is a 64-core CPU operating at 2.6 GHz(base clock?) with a TDP of 180W at 7nm. Launched in January 2019.

Mp1YOfX.jpg


You can literally combine together 2 Kunpeng 920s (128 cores!) and still have a lower TDP than Intel's 400W.

Also the price...

Intel’s Xeon 9200 lineup is expected to be priced somewhere between in the $25K to $50K range, while the 56-core 9282 won’t be any less than $40K. You are looking at a TDP of 400W and a boost clock of 3.8GHz. Neat, but for that much, you can buy four 64-Core AMD Rome processors plus some extras too.
https://www.techquila.co.in/7nm-amd-epyc-rome-cpu-64/

:rofl::lol:
 
.
It all depends on what you're using the CPU for. For example, each of the top 4 supercomputers on the TOP500 list have over a million cores each. Sunway Taihulight has over 10 million cores.

But what you're saying is correct for desktop purposes.

Anyway, let me show the difference between Intel's 14nm vs Huawei's 7nm.

Intel Xeon Platinum 9282 is a 56-core server microprocessor operating at 2.6 GHz base, 3.8 GHz turbo boost, with a TDP of 400W at 14nm. Launched in April 2019.

Huawei Kunpeng 920 is a 64-core CPU operating at 2.6 GHz(base clock?) with a TDP of 180W at 7nm. Launched in January 2019.

Mp1YOfX.jpg


You can literally combine together 2 Kunpeng 920s (128 cores!) and still have a lower TDP than Intel's 400W.

Also the price...

Intel’s Xeon 9200 lineup is expected to be priced somewhere between in the $25K to $50K range, while the 56-core 9282 won’t be any less than $40K. You are looking at a TDP of 400W and a boost clock of 3.8GHz. Neat, but for that much, you can buy four 64-Core AMD Rome processors plus some extras too.
https://www.techquila.co.in/7nm-amd-epyc-rome-cpu-64/

:rofl::lol:
Intel chips are power sucker highly inefficient, years of monopoly created this.
 
.
nasdaq.com
Baidu Unveils High-Performance AI Chip, Kunlun, at Baidu Create 2018
4 minutes
The search engine provider releases China's first cloud-to-edge AI chip for training and inference
BEIJING, July 03, 2018 (GLOBE NEWSWIRE) -- Baidu Inc. (NASDAQ:BIDU) today announced Kunlun, China's first cloud-to-edge AI chip, built to accommodate high performance requirements of a wide variety of AI scenarios. The announcement includes training chip "818-300"and inference chip "818-100". Kunlun can be applied to both cloud and edge scenarios, such as data centers, public clouds and autonomous vehicles.

Kunlun is a high-performance and cost-effective solution for the high processing demands of AI. It leverages Baidu's AI ecosystem, which includes AI scenarios like search ranking and deep learning frameworks like PaddlePaddle. Baidu's years of experience in optimizing the performance of these AI services and frameworks afforded the company the expertise required to build a world class AI chip.

In 2011, Baidu started developing an FPGA-based AI accelerator for deep learning and began using GPUs in datacenters. Kunlun, which is made up of thousands of small cores, has a computational capability which is nearly 30 times faster than the original FPGA-based accelerator. Other key specifications include: 14nm Samsung engineering, 512 GB/second memory bandwidth, as well as 260TOPS while consuming 100 Watts of power.

In addition to supporting the common open source deep learning algorithms, Kunlun chip can also support a wide variety of AI applications, including voice recognition, search ranking, natural language processing, autonomous driving and large-scale recommendations.

With the rapid emergence of AI applications, dramatically increasing requirements are being imposed on computational power. Traditional chips limit how much computing power is available and thus how far AI technologies can be accelerated. Baidu developed this chip, specifically designed for large-scale AI workloads, as an answer to this demand. Baidu believes that it will allow for significant advancements in the open AI ecosystem.

Baidu plans to continue to iterate upon this chip, developing it progressively to enable the expansion of an open AI ecosystem. As part of this, Baidu will continue to create "chip power" to meet the needs of various fields including intelligent vehicles, intelligent devices, voice recognition and image recognition.
Baidu and Samsung Electronics Ready for Production of Leading-Edge AI Chip for Early Next Year
Korea, China on December 18, 2019

Designed based on Samsung’s 14nm process and I-Cube TM package technology, Baidu KUNLUN chip to expand AI ecosystem and transform the user experience


Baidu, a leading Chinese-language Internet search provider, and Samsung Electronics, a world leader in advanced semiconductor technology, today announced that Baidu’s first cloud-to-edge AI accelerator, Baidu KUNLUN, has completed its development and will be mass-produced early next year.


....

Baidu and Samsung Electronics Ready for Production of Leading-Edge AI Chip for Early Next Year – Samsung Global Newsroom
 
.
DECEMBER 18, 2019 FEATURE
Study highlights the potential of nanotube digital electronics
by Ingrid Fadelli , Tech Xplore

10-studyhighlig.jpg
Device uniformity and medium-scale CNT integrated circuits. (a) Optical image showing a wafer covered with CNT FETs, with (b) corresponding transfer characteristics of 300 FETs. (c) SEM image showing a CNT 8-bit full adder CMOS circuit composed of 256 CNT FETs, with input (A and B) and output (Sum) waveform of the sum for a carry-in (d) Cin=0 and (e) Cin=1 from a previous addition. Credit: Dr. Haitao Xu.

Some experts in the field of electronics engineering have suggested that the use of silicon complementary metal-oxide semiconductors (CMOS) will start declining rapidly by the end of 2020. Despite their predictions, a class of alternative materials that can effectively sustain the computational power of new devices, while maintaining good energy efficiencies is yet to be clearly established.

Over the past few years, researchers have proposed several materials that could ultimately substitute current CMOS devices. Some of the most promising candidates are carbon nanotube (CNT)-based electronics, which can be fabricated using a variety of different techniques.

A team of researchers at Peking University and Xiangtan University in China has recently carried out a study investigating the potential of CNT materials for fabricating electronics. In their paper, published in Nature Electronics, the researchers discussed the development of nanotube-based CMOS field-effect transistors over time, while also highlighting some of the CNT materials that are currently available to electronics manufacturers.

"CNT is an ideal electronic material that offers solutions where other semiconductors fundamentally fail, particularly when scaled to the sub-10 nm dimensional scale," Lianmao Peng, one of the researchers who carried out the study, told TechXplore. "In this work, we demonstrated that CNT based electronics have the potential to outperform that of silicon technology by a large margin (experimentally demonstrated over ten times advantage) and that large-scale integrated circuits (ICs) can be constructed using carbon nanotubes."

The relevant physical parameters of CNTs, such as their structure and electronic properties, are now well-known in the field. To effectively explore the potential limitations of CNT materials, Peng and his colleagues Zhiyong Zhang and Chenguang Qiu thus analyzed the performance and qualities of individual CNTs, focusing on these specific parameters.

"Our results show that at sub-10 nm technology nodes, CNT transistors can be 3 times faster, and 4 times more energy efficient than their silicon counterparts," Peng explained. "We demonstrated that, even using the very limited university fabrication facility, we can fabricate transistors that outperform silicon transistors by many times, indicating that the chip industry could move ahead with the current speed for many more decades."

The study carried out by Peng and his colleagues provides further evidence suggesting that CNT transistors are a viable and desirable alternative to current silicon CMOS devices. In their analyses, the researchers also highlighted some of the advantages and disadvantages of the medium-scale integrated circuits that have been developed to date, as well as the challenges that are currently preventing their large-scale implementation.

According to Peng and his colleagues, developing integrated circuits (ICs) with new 3-D chip structures could enhance the performance of CNT materials further, making them up to hundreds of times more powerful. Their analyses and previous findings gathered by other research teams ultimately hint at the possibility of CNT technology being the solution to deliver more powerful and highly energy efficient chip technology in the post-Moore era.

"Right now, we can fabricate few extremely powerful transistors on individual CNTs, but not very complicated ICs," Peng said. "On the other hand, we can build CNT based ICs with over 10k transistors in three-dimensions using CNT thin film but with very limited performance. In the future, we need to combine the two directions of research, building high-performance large-scale ICs using CNT films with performance exceeding that of silicon chip technology."

More information: Lian-Mao Peng et al. Carbon nanotube digital electronics, Nature Electronics (2019). DOI: 10.1038/s41928-019-0330-2

Study highlights the potential of nanotube digital electronics | TechXplore
 
.
.
Baidu and Samsung Electronics Ready for Production of Leading-Edge AI Chip for Early Next Year
Korea, China on December 18, 2019

Designed based on Samsung’s 14nm process and I-Cube TM package technology, Baidu KUNLUN chip to expand AI ecosystem and transform the user experience


Baidu, a leading Chinese-language Internet search provider, and Samsung Electronics, a world leader in advanced semiconductor technology, today announced that Baidu’s first cloud-to-edge AI accelerator, Baidu KUNLUN, has completed its development and will be mass-produced early next year.


....

Baidu and Samsung Electronics Ready for Production of Leading-Edge AI Chip for Early Next Year – Samsung Global Newsroom

Just curious, just don't know why Baidu choose Samsung but not TSMC or other vendors? Qualcomm may drop Samsung because it doesn’t want Samsung to know its trade secrets and design drawings which it can then potentially use to improve its own Exynos chipsets. If Baidu go with Samsung, later Samsung may steal AI tech from Baidu.
----
For reference: https://wccftech.com/qualcomm-snapdragon-865-orders-samsung-no-deal/
Qualcomm May Think That Providing Snapdragon 865 Orders May Lead Samsung to Make a Proper Exynos Equivalent SoC
It was initially believed that the Snapdragon 865 would be made by Samsung, but a new report says that the order instead went to TSMC. That’s apparently not because Qualcomm thinks TSMC is any better, but because it reportedly doesn’t want Samsung to know its trade secrets and design drawings which it can then potentially use to improve its own Exynos chipsets.

On the other hand, the mid-tier Snapdragon 765 and Snapdragon 765G chips will apparently still be manufactured by Samsung, but they aren’t Qualcomm’s crown jewel by any means. In one way, getting Snapdragon 765 and Snapdragon 765G orders is good news for the South Korean giant, as its market share has apparently reduced in recent times, while TSMC still continues to dominate the industry.

With Samsung having shut down its custom core development facility due to being inferior in both performance and efficiency, and by extension, not competitive, it seems that Qualcomm has nothing to worry about. However, it turns out that the manufacturer still wants to be on the safe side and guard its intellectual property.

Apple and Huawei already make their own chipsets so they are least bothered about this. OPPO is apparently also planning to make its own SoC to power its phones. Not sharing its design with Samsung and weakening its business further by not giving its orders sounds like a pragmatic decision on Qualcomm’s end. Otherwise, in the worst-case scenario, Samsung might overtake it down the road, but that’s very unlikely for now.
 
.
China unveils Brain-Computer Interface chip
Source: Xinhua| 2019-05-18 20:56:37|Editor: Liangyu

TIANJIN, May 18 (Xinhua) -- China has achieved a breakthrough in Brain-Computer Interface (BCI) chip research, with its first BCI chip "Brain Talker" making its debut on Friday at the ongoing World Intelligence Congress held in northern China's Tianjin Municipality.

BCI is a system allowing a person to control a computer or other electronic device using his or her brainwaves, without requiring any movement or verbal instruction.

Brain Talker, specially designed for decoding brainwave information, may replace traditional computer devices used in BCI due to its more portable size, precision in decoding, high efficiency in computing and faster communication ability.

Ming Dong, director of the Academy of Medical Engineering and Translational Medicine in Tianjin University, said the chip can identify minor neuron information sent by the brain wave from the cerebral cortex, efficiently decode the information and greatly quicken the communication speed between the brain and machine.

"Brain Talker makes BCI technology more promising for civil use since the chip is more portable, wearable and simpler," Ming added.

Brain Talker was co-developed by Tianjin University and China Electronics Corporation with fully independent intellectual property rights.

Cheng Longlong, a data scientist from China Electronics Corporation, said scientists would endeavor to enhance the performance of the chip for wider use in the fields of medical treatment, education, home life and gaming in the future.

6d1906aee46d462fb860abcfc14d1061.jpg
China-developed interface makes thought-controlled typing a reality
(People's Daily Online) 14:40, December 24, 2019

FOREIGN201912241441000512827476499.jpg
(Photo/Xinhua Daily Telegraph)

Researchers in northern China's Tianjin have developed a brain-computer interface (BCI) that allows users to type words with their minds, Xinhua reported on Dec. 23.

The device, independently developed by China, provides a direct communication channel between the human brain and the external world, according to a researcher with Tianjin University.

The cap can pick up electroencephalograph (EEC) signals and send them to the computer. Then, the brain patterns, which carry the user's intent, are decoded by an algorithm on the device. Eventually, the information is converted into instructions and triggers an action.

Using the BCI, a graduate student in Tianjin University typed 69 characters in one minute via mind control, winning the first prize in a BCI competition held earlier this year. The speed proved faster than hand typing.

The device is applicable in medicine, rehabilitation, brain cognition, neural feedback and signal processing, researchers said, adding that the BCI technology will be adapted for civilian use, and the device will become easier to carry and use, becoming wearable in the future.

 
.
China Focus: China's major chip maker unveils new generation of processors
Source: Xinhua| 2019-12-24 20:17:40|Editor: Xiang Bo

BEIJING, Dec. 24 (Xinhua) -- China's major chip maker Loongson on Tuesday unveiled its new generation of general-purpose CPUs with improved processing performance in Beijing.

The new generation CPUs, Loongson 3A4000 and 3B4000, adopt a 28 nm technique and their main frequency reaches 1.8GHz to 2.0GHz. Their performances are more than twice that of the previous generation, according to Hu Weiwu, president of Loongson.

All the source code of the chip is independently developed, and a security control mechanism is designed in the processor core. An open-source operating system has also been developed to support enterprises. The upgrade of Loongson CPUs and motherboard will not affect the compatibility of the operating system and applications.

Hu said Loongson is taking steps to compete with international chip giants, adding that Loongson will launch new CPUs that adopt a 12 nm technique in 2020 and 2021. Their main frequency is expected to reach 2.5GHz, with the processing performance reaching world-class levels.

Loongson was initiated by the Institute of Computing Technology under the Chinese Academy of Sciences (CAS). Loongson chips have been applied to China's space program, finance, energy and other fields.

Loongson is trying to establish a new information industry ecosystem and statistic shows that Loongson chip shipments have reached more than 500,000 in 2019.

Zhou Dejin, an official from the CAS said Loongson has mastered the core technology of CPUs through independent research and development after 20 years' effort, showing its determination in building an independent and innovative information industry system.

Loongson's users, including Lenovo and the China Academy of Launch Vehicle Technology, also released products based on Loongson's new generation processors, such as desktop computers, notebooks, servers, network security equipment and industrial control computers.

2019122418_4250f6409f38461bbb1b89c07b8fe055_7363_wmk.jpg
 
.
China IC capacity on track to expand further in 2020, says Digitimes Research | DIGITIMES
Eric Chen, DIGITIMES Reseaerch, Taipei
Friday 27 December 2019

China's overall IC production capacity will continue to expand in 2020 as both IDMs and foundry houses there have readied expansion plans covering 4-, 6-, 8- and 12-inch wafer fabs and will enter 14nm node next year, according to Digitimes Research.

The capacity expansion plans are driven by the mounting demand for logic ICs, memory chips, power chips and analog ICs for 5G and IoT applications in China, as well as the government's aggressive push for pursuing IC self-sufficiency, Digitimes Research noted.

Both IDMs and foundry firms have seen some of their new fabs start commercial runs in 2019, with some more to kick off volume production of new products or technologies in 2020, when new capacity expansion plans will also be implemented.

Meanwhile, the adoption of 14nm node, mass production of 3D NAND flash chips, and growing demand for power devices and analog ICs for IoT applications all serve to prompt Chinese IDMs and foundry houses to strengthen deployments in 2020 in both mature and special process nodes, whether at 12-inch, 8-inch or lower-inch wafer fabs.
 
.
Industry aims to wean itself off US technology amid trade war
By Li Xuanmin Source:Global Times Published: 2019/12/30 20:36:22

fab8c4b1-28a8-418e-b04b-2bbfdc3c9b84.jpeg
A view of Alibaba's AI chip Hanguang 800 Photo: Courtesy of Alibaba

China has made up its mind to become self-sufficient in chip technology. Amid a boiling trade war with the US that disrupts the global supply chain, China's chip industry is witnessing a sweeping change, with investment plowing in apace and breakthroughs being made in high-end chips that will significantly reduce reliance on imports.

In the latest move, China's government-funded "starlight chip project" announced on Monday that it plans to invest 10 billion yuan ($1.43 billion) in the next decade on chip technology research, standard-setting study, application development and large-scale industrialization.

Launched in 1999, the project has applied for more than 3,000 patents and formed several chip technological systems including digital media, intelligent security and artificial intelligence.

The project is a vivid example of how investment is shaping China's semiconductor industry this year, in particular after Washington's brutal crackdown on Chinese tech companies like Huawei and ZTE that could potentially cut off key US component supplies.

In October, China set up a second national semiconductor fund of 204.2 billion yuan in a bid to nurture the domestic chip industry, a 47-percent increase of the scale of investment compared with the first fund of 138.7 billion yuan, according to media reports.

"Chinese industry insiders and authorities are giving the biggest-ever incentives to the homegrown chip industry. We all feel a sense of urgency to wean ourselves off foreign technology, spurred by a spiraling trade war," a manager of a Beijing-based chip start-up who spoke on condition of anonymity told the Global Times on Monday.

The whole industrial chain has been shifting its attitude on chips made by Chinese suppliers, according to the manager.

"In the past, downstream vendors tended to prefer foreign chips over homegrown ones. Now, they gravitate toward ours and are willing to help us in accommodating, testing and even in improving functions," he explained.

The industry-wide effort has helped to fuel a boom in the design of advanced computer and smartphone chips. It has also led to a rapid expansion of the market share of homegrown memory chips.

In September, Huawei's HiSilicon unveiled its latest mobile application processor - the Kirin 990. The chipset series is widely believed to be the world's most powerful mobile system-on-chip, with a performance surpassing its foreign competitors such as Qualcomm.

"Huawei's Kirin series represents a major breakthrough in the chip industry. It shows that Chinese players have the ability to design all ranges of chips and their gap with leading foreign players is closing," Xiang Ligang, an expert in the telecoms industry, told the Global Times on Monday. "We just need some time to forge industrial chain ability."

China is on track to achieve its goal of being able to produce 40 percent of the semiconductors it uses by 2020 and 70 percent by 2025. Chinese firms currently supply more than 15 percent of the semiconductors used in the nation, industry insiders estimated.

The nation is also one step closer to producing about 5 percent of the world's memory chips by the end of 2020 from virtually none in 2018, the Nikkei Asian Review reported, quoting sources close to the matter.

But observers admitted that Chinese firms' chip manufacturing abilities are years behind their rivals due to their late start. China's largest chip manufacturer, SMIC, has reportedly begun mass production of chips using its 14-nanometer FinFET manufacturing technology, while top foreign players such as Samsung and Intel already are in a race to supply 7-nanometer chips to the market.
 
. .
China's first 14nm production line put into operation - cnTechPost
Jan 12, 2020

071.jpg

The Shanghai plant of Semiconductor Manufacturing International Corporation (SMIC), the first 14-nanometer production line in China, has been put into operation, the company said on its website.

SMIC has successfully achieved mass production of the first-generation 14-nanometer FinFET process last year. The SMIC Shanghai plant will build two integrated circuit advanced production lines with a monthly capacity of 35,000 pieces.

According to the Pudong Times, SMIC's Shanghai plant has achieved its goal one year earlier than expected.

At the end of last year, SMIC released its third quarter financial report for 2019, with revenue of US $816.5 million, an increase of 3.2% compared with $790.9 million in the previous quarter, compared with $850.7 million in the same period last year.

The profit was $84.662 million, an increase of 1014.8% compared with $7.591 million in the same period last year.
 
.
Huawei reportedly shifts 14nm chip orders from TSMC to China's SMIC - Huawei Central
img-1-dengli-1-80x80.jpg
By Deng Li
Posted on January 13, 2020

hisilicon-chips-img-1.jpg

Huawei’s semiconductor making arm HiSilicon has placed 14nm chip orders to China’s Semiconductor Manufacturing International Corporation (SMIC), says a report regarding this subject.

SMIC has recently received orders for 14nm FineFET process manufacturing from HiSilicon, which has been among major clients of TSMC’s Nanjing fab that was opened in late 2018 and entered directly 16nm FinFET chip production, reported Digitimes.

This could also be a result of recent US decisions to limit US technology supplies from foreign firms including TSMC from 25% to 10%, which affects TSMC’s supplies of 14nm, as it contains over 10% US tech for the manufacturing process.

However, The 7nm chip manufactured by TSMC contains only 9 percent of the US tech, so it can still supply the 7nm products.

14nm chips are mainly used for network equipment and as announced by Huawei, the Chinese company now producing its own network equipment chips without US technologies.
 
.
Huawei reportedly shifts 14nm chip orders from TSMC to China's SMIC - Huawei Central
img-1-dengli-1-80x80.jpg
By Deng Li
Posted on January 13, 2020

hisilicon-chips-img-1.jpg

Huawei’s semiconductor making arm HiSilicon has placed 14nm chip orders to China’s Semiconductor Manufacturing International Corporation (SMIC), says a report regarding this subject.

SMIC has recently received orders for 14nm FineFET process manufacturing from HiSilicon, which has been among major clients of TSMC’s Nanjing fab that was opened in late 2018 and entered directly 16nm FinFET chip production, reported Digitimes.

This could also be a result of recent US decisions to limit US technology supplies from foreign firms including TSMC from 25% to 10%, which affects TSMC’s supplies of 14nm, as it contains over 10% US tech for the manufacturing process.

However, The 7nm chip manufactured by TSMC contains only 9 percent of the US tech, so it can still supply the 7nm products.

14nm chips are mainly used for network equipment and as announced by Huawei, the Chinese company now producing its own network equipment chips without US technologies.
This is natural consequence of TW flirting with Americans. See, they haven't actually did anything yet, but the sole fact of this happening did more damage to them than then doing that deliberately. Now Taiwanese tech community will take a second though when taking anything from the government.

Only if TW will give an official guarantee that their industry is off limits to any political advances from their "allies," will they get our trust back.

If SMIC will finally learn how to get their conduct straight, I have no doubt that they will be able to undercut TSMC on cost in anything, but leading edge nodes.
 
Last edited:
.

Country Latest Posts

Back
Top Bottom