Chinese CPU , IC and Operational Systems

Gowin, TSMC team up for 28nm FPGA chips

https://digitimes.com/news/a20171102PD209.html

Jean Chu, Taipei; Jessie Shen, DIGITIMES [Friday 3 November 2017]

China-based Gowin Semiconductor has announced the development of its first 28nm FPGA product, the GW3AT-100, fabricated by Taiwan Semiconductor Manufacturing Company (TSMC).

Gowin is already working with TSMC to produce mid-density FPGA chips using the foundry's 55nm SRAM process technology. Gowin's 55nm FPGA solutions are rivaling the company's larger international peers Xillinx' Spartan FPGA series and Altera's Cyclone solutions.

The recently developed solution is what Gowin claims China's first homegrown 28nm FPGA chip, which will compete with Xilinx' Kintex-7 series.

FPGA demand will be driven by emerging applications such as robots, drones, big data, IoT, autonomous driving and 5G communications, In particular, FPGAs are seen as an important element to deliver high-performance artificial intelligence (AI). According to Gowin CEO Jason Zhu, the company's FPGA chips will be targeting the Asia Pacific market for growth.

Gowin has added new distributors in South Korea and Taiwan, said Zhu. The company is also looking to build its presence in Japan, Australia, India, North America and Europe in 2018, Zhu indicated.

Zhu said Gowin expects to cumulatively ship over two million wafers before the 2018 Lunar New Year, which falls in February.

Zhu worked for Lattice Semiconductor in 1996-2011, and was involved in the development of Lattice's seven generations of FPGA series. Zhu was once the core technical leader in programmable logic of the China National 863 program.

Ning Song, president and CTO of Gowin, has more than 25 years of R&D experience in the EDA industry including Lattice and Cadence Design Systems.

Founded in 2014, Gowin is based in Guangdong. The company recently struck a deal with Guangzhou Economic and Technological Development Zone enabling it to obtain government subsidies of over CNY80 million (US12.1 million).

 

Report: TSMC's 3nm Fab Could Cost $20 Billion

EE Times

https://www.eetimes.com/document.asp?doc_id=1332419

10/9/2017 03:01 PM EDT

• 
  SAN FRANCISCO — A 3mn fab being planned by chip foundry giant TSMC is likely to cost more than $20 billion to build and equip, TMSC Chairman Morris Chang told the Bloomberg news service.
  
  TSMC announced last week it would locate what is the world's first announced 3nm fab in the Tianan Science Park in southern Taiwan, laying to rest speculation that TSMC might build the fab in the U.S. or elsewhere outside of Taiwan. TSMC did not give a timeframe for the fab's completetion, but has said in the past it would build a 5- or 3nm fab as early as 2022.
  
  
  
  
  Morris Chang
  
  
  Chang, who also last week announced plans to retire next June, also told Bloomberg that TSMC would increase capital spending to as much as $11 billion from $10 billion in recent years to keep up with the pace of technology migration.
  

 

That's Taiwan though, not China.

 

28nm is last gen AMD.

 

http://news.moore.ren/industry/9644.htm


In June 2014, the State Council promulgated the Outline for Promoting the Development of the National Integrated Circuit Industry. In September of the same year, the National Integrated Circuit Industry Investment Fund (also known as the "Grand Fund") was formally established with a fund-raising of 138.72 billion yuan in the first phase. At the beginning of 2017, at an industry summit held in Shanghai, Ding Wenwu, general manager of National IC Industry Investment Fund Co., Ltd., said that as of the end of 2016, the National Funds had decided to invest 43 projects and accumulate project commitments The amount of investment is 81.8 billion yuan, and the actual capital contribution is over 56 billion yuan. Has implemented the project covers the integrated circuit design, manufacturing, packaging testing, equipment, materials, ecological construction and other sectors.

Funds have also been set up in various parts of the country to support the development of integrated circuits. ICIFs have been set up in a number of provinces and cities such as Beijing, Shanghai, Sichuan and Hubei. In December 2013, Beijing announced the establishment of an integrated circuit industry development equity investment fund with a total fund size of 30 billion yuan. Shanghai IC Industry Fund, Nanjing IC Industry Development Fund, An Xin Industrial Investment Fund (Fujian) Fund size as high as 50 billion yuan. According to incomplete statistics, a large fund can mobilize about 500 billion yuan of local capital. In addition, some investment agencies are also actively looking for some of the hot frontier.



http://news.moore.ren/industry/9684.htm

The achievements of National science and technology major projects and industrial development planning:


First, with the support of major projects, the Chinese IC industry has formed a technical system and established an industrial chain. The industrial ecology and competitiveness have been improved.

Second, China's high-end chip design capabilities have been significantly improved.

Third, China's manufacturing process has made great progress, with mass production of 65, 40, 28nm technology and breakthroughs in 14nm technology research and development. The competitiveness of the featured technology has also been enhanced.

Fourthly, China's IC packaging industry has entered a high-end market from mid-to-low end with a substantial increase in its competitiveness. Mass production of high-density packaging, into the high-middle package, copper bumps and other technologies to achieve independent innovation, 3D package made a breakthrough. Among them, the packaging industry chain innovation alliance played an important role in promoting collaborative innovation.

Fifth, the key equipment and materials to achieve from scratch, the overall level of 28nm, some products have entered 14nm, domestic and foreign production lines. Specifically, there are 16 kinds of 12-inch equipment through the production line assessment cognition into mass sales, the overall technical level of 28nm. 29 kinds of IC packaging and testing equipment to complete research and development, assessment and certification into mass sales, high-end packaging equipment has formed a whole line integration capabilities.

Sixthly, a batch of backbone enterprises with innovative vitality and certain international competitiveness have been nurtured.

At the same time, Ye Tianchun also stressed that the intellectual property rights of China's integrated circuits have also been improved. During the implementation of major national special projects, Chinese companies applied for a total of 43292 invention patents, of which 23,477 were major special patents, accounting for 53.23% of the total number of patent applications.

 

http://news.moore.ren/industry/9403.htm

Recently, the Nanjing Panda (code-named CLP 14) held its third exhibition of high-tech equipment and achievements in civil-military integration and development in Beijing. The exhibition featured the HuaRui DSP chips jointly developed by Tsinghua University and Godson.

DSP chip, scientific name of digital information processing chip, is a special chip, designed to deal with a variety of signals such as radar infrared radio signal weapon, the current advanced radar are used digital components, software design, which is a necessity, Such as cell phone chips.

Hua Rui 1 is a 4-core microprocessor independently developed in China. The processor integrates vector processing components internally and extends a vector instruction set including a fast Fourier transform (FFT) butterfly operation instruction with Strong FFT operation and digital pulse pressure processing capabilities. The real-time measurement under embedded real-time operating system shows that Huarui 1's single-core processing time of 1 024 single-precision floating-point complex FFTs and digital pulse pressure are 6.12μs and 15.36μs respectively. After using multithread FFT parallel processing technology, 4 nuclear speedup up to 3 times more than the full load power consumption of only 10 W, the overall performance is superior to similar processors at home and abroad, for real-time requirements of the radar signal processing and electronic warfare and other fields.

Hua Rui 1 uses 65nm CMOS technology, working frequency is 550MHz, processing capacity of 32GFMACS, power consumption is 10W. In terms of technical specifications, Hua Rui No. 1 DSP performance significantly better than Freescale's MPC8640D, ADI's TS201 and the United States Texas Instruments C6701. According to the research conducted by Nanjing Institute of Electronics Technology, HuaRui No. 1 overall performance is superior to similar processors at home and abroad, and is suitable for fields such as radar signal processing and electronic countermeasure demanding real-time.

It is precisely because of the excellent performance of Hua Rui 1, this DSP has been successfully applied to radar products, creating a multi-core domestic DSP chip product applications, "the three most": the largest radar equipment applications, A single set of applications and the largest number of applications (the news implies that the most advanced domestic J-20 radar should also use this chip).

 

http://news.moore.ren/industry/9405.htm

The Rishen semiconductor bear the country "02 special" one of the core projects, 40 ~ 28 nm integrated circuit manufacturing 300 mm silicon. With the scientific and technological progress step by step, the industrialization goal has been realized step by step. The monopoly of the IC on the IC is being thawed. The country started to form a complete semiconductor industry chain, which has long plagued China's semiconductor industry. Is expected to be cracked.

Semiconductor silicon manufacturing process, is appeared in the 1950s of last century, the United States, Germany, Japan, South Korea and other countries already applied for layers of patent protection. As a "latecomer," China wants to make its own products on the whole chain and must break the patent "encirclement" set up abroad. After many arduous efforts, the Rishen at the beginning of undertaking a number of breakthroughs in intellectual property. At present, the Rishen has been in the crystal growth, silicon wafer substrate technology, epitaxial material growth and advanced CMOS devices and processes, high-power device structure, materials and processes, and other aspects of the patent layout. A total of 425 applications for invention patents were completed, and 47 invention patents were granted.

However, the solution to the patent issue is only the "first step" in the innovation long march. The "02 special" program requirements, is the contractor not only to research and development success, but also to achieve the goal of mass production. In the eyes of international monopolists: Even if your products are made, if the cost is high, can not sell, still can not hold on!

In fact, mass production after key technological breakthroughs in the key areas of semiconductor chip design and manufacturing is the multi-scale comprehensive competition of quality, stability and cost control. Strongly supported by the state, the Rishen as the world's leading 12-inch large wafer manufacturers, is also the only Chinese largest 12-inch large wafer manufacturer.

 

"Moore's Law" is the number of transistors that can be accommodated on an integrated circuit, doubling about every 18 months. But now the time has changed. Each generation IC technology will last longer, 3 or 4 years. And 3 or 2 manometers IC may be the limit or bottleneck when no technological breakthrough.　It means that the leading IC manufacturers already close to the finish line, and had to wait for the competitors catch up from behind.

This trend can let Chinese IC manufacturers have more time for R&D and catch up the latest IC manufacturing technology. At present, China's manufacturing process has reached mass production of 28nm technology and breakthroughs in 14nm technology research and development. It is not far away for the Chinese to catch up the leaders which uses 10 nm mass production technology right now.

 

And if you half nm, the number quadruples (36 months). So now China is at 28nm vs 7nm. So 6 years behind.

They can build a fast supercomputer but the compute power per node is still way lower, which makes it less useful.

 

This also means that the power and energy issues are higher here. Which means that China (despite talking so much about it) is going to be behind in the exascale computing race. Either that or they will need to put a multi-TW nuclear plant next to their exascale datacenter.

 

It depends really. If you look at their current supercomputers, sure, they have the fastest two, but the compute power per node is way lower, they just have more nodes. This makes them less reliable, more power hungry and less versatile at computes that can't be broken down into the optimum number of parallel paths.

Technically, you could make an exascale supercomputer with x286 CPUs with sufficient nodes, but it would be terrible.

 

Exactly my point. You can make an exascale machine with 286s but the size, the power needed, the network fabric needed will be so massive that you can't use such a machine for any kind of serious work. The cost per TFLOP will be way beyond the costs of similar sized machines and this is why I said sometime earlier in this thread that their chances of winning the exascale race is pretty low. If you look at their current state-of-art processors, they are still very far behind the curve and their parallel processors are not up to scratch yet.

Another pointer that they aren't there is the point (I think it was @RMLOVER who made the point) that they use such machines for computing Govt applications. Govt applications are NOT supercomputing applications and are not meant for such systems - you might put them on clusters of servers but not on Supercomputers. Looks like despite the talk here (yes, they have made some progress, though why they are using the MIPS architecture is beyond me), they are not going to win the exascale race.

But yes, that does not mean that they can't put their processors in low end devices like phones, IoT etc., which is a pretty large market by itself.
All in all, a very interesting thread.

 

Tianhe-3 to offer faster, sharper data processing
By ZHANG MIN in Tianjin and MA SI in Beijing | China Daily | Updated: 2017-05-17 07:17



An engineer inspects the Tianhe-1 supercomputer at the National Supercomputer Center in Tianjin. YUE YUEWEI / XINHUA



China is stepping up research and development of the Tianhe-3 supercomputer, which aims to be 10 times faster than the current world leader, as it strives to meet the country's growing needs for more accurate and efficient computing capabilities.

The Tianhe-3 is designed as the world's first prototype exascale supercomputer, which means it can make a quintillion (1 followed by 18 zeros) calculations per second, and is intended to arm the country's manufacturers and government with greater speed, precision and scope for research.

The supercomputer will be applied in such fields as the analysis of smog distribution, airplane designs, oil surveying and the development of artificial intelligence.

These functions overlap with the nation's existing supercomputers, such as the Sunway TaihuLight, which is currently the world's fastest supercomputer, with a peak speed of 125 quadrillion (1 followed by 15 zeros) calculations per second, said Meng Xiangfei, director of application at the National Supercomputer Center in Tianjin.

"What sets the Tianhe-3 apart from the others is, for example, that it can help design more reliable aircraft by giving a view of its detail at a magnification of up to 10 times," Meng added.

He explained that when forecasting weather on a national scale, the measuring range of the Tianhe-3 can be as accurate as 1 square kilometer. That would be five to 10 times more accurate than other supercomputers.

"I think some weather forecasting applications in smartphones will become more accurate by the time the Tianhe-3 is fully functioning," said Meng, who confirmed the prototype is expected to be finished early next year and become fully operational by 2020.

He said that the development of the Tianhe-3 differs from the Sunway TaihuLight in some quite significant ways.

"To be specific, we are trying to make a 'supercell' which combines calculation, communication, big data and AI, which used to work separately. That is why the National Supercomputer Center in Tianjin has brought in a new batch of AI experts," said Meng.

China is taking its computing power to the next level amid intensifying global competition in this field.

The United States is working on an exascale supercomputer, but its prototype will not appear until 2023. Japan is building a machine that can make 130 quadrillion calculations per second, which could surpass Sunway TaihuLight.

Wang Gang, an associate professor at the College of Computer Science and Technology of Jilin University, said China's supercomputers are chiefly used to assist scientific research, which need to process a huge amount of information within a short time span.

"But as the Tianhe-3 improves AI capabilities, it will not only calculate faster but also smarter. It will therefore have greater potential in terms of commercial application," Wang said.

Wang is working on a deep-learning project for autonomous driving. He has used the Sunway TaihuLight supercomputer to assist his research, and expects the Tianhe-3 to be more helpful.

"China has an edge in hardware, but more efforts are needed to develop tailor-made software for supercomputers, in order to widen its application," Wang said.

 

Chinese has start the other way to make supercomputer

China will open a $10 billion quantum computer center and others also investing in quantum computing
brian wang | October 10, 2017 |
RBP
https://www.nextbigfuture.com/2017/...hers-also-investing-in-quantum-computing.html



Save



On 37 hectares (nearly 4 million square feet) in Hefei, Anhui Province, China is building a $10 billion research center for quantum applications. This news comes on the heels of the world’s first video call made via quantum-encrypted communications and the completion of a quantum-encrypted fiber optic trunk cable.

The National Laboratory for Quantum Information Sciences, slated to open in 2020, has two major research goals: quantum metrology and building a quantum computer. Both efforts would support military and national defense efforts, as well civilian innovators.

Pan Jianwei, a leading Chinese quantum scientist, says that the first general-purpose Chinese quantum computer could have a million times the computing power of all other computers presently in the world. In the computers we use today, information is encoded in a series of bits set as either 1 or 0. In a quantum computer, bits would theoretically be able to hold one, both, or some combination of these states. They could be used to speedily crack encrypted messages or solve complicated research problems involving anything from weather modeling to fusion research and biomedicine, because quantum bits allow certain calculations that happen one by one on a standard computer to occur simultaneously.

China QUESS quantum satellite is projected to have only a two-year lifespan transmitting quantum keys between China and Europe, it’ll be succeeded in 2030 by a constellation of quantum satellites.

 

China speeds ahead of U.S. as quantum race escalates, worrying scientists
BY TIM JOHNSON

http://www.mcclatchydc.com/news/nation-world/national/national-security/article179971861.html
tjohnson@mcclatchydc.com

OCTOBER 23, 2017 5:00 AM

WASHINGTON
U.S. and other Western scientists voice awe, and even alarm, at China’s quickening advances and spending on quantum communications and computing, revolutionary technologies that could give a huge military and commercial advantage to the nation that conquers them.

The concerns echo — although to a lesser degree — the shock in the West six decades ago when the Soviets launched the Sputnik satellite, sparking a space race.

In quick succession, China in recent months has utilized a quantum satellite to transmit ultra-secure data, inaugurated a 1,243-mile quantum link between Shanghai and Beijing, and announced a $10 billion quantum computing center.



“It doesn’t necessarily mean that their scientists are better,” said Martin Laforest, a physicist and senior manager at the Institute for Quantum Computing at the University of Waterloo in Ontario, Canada. “It’s just that when they say, ‘We need a billion dollars to do this,’ bam, the money comes.”

The engineering hurdles that China has cleared for quantum communication means that the United States will lag in that area for years.

“The general feeling is that they’ll get there before us,” said Rene Copeland, a high-performance computer expert who is president of D-Wave (Government) Inc., a Vancouver-area company that uses aspects of quantum computing in its systems.

But building a functioning quantum computer sets forth different kinds of challenges than mastering quantum communication, and may involve creating materials and processes that do not yet exist. Once thought to be decades off, scientists now presume a quantum computer may be built in a decade or less. The stakes are so high that advances by the U.S. government remain secret.

“We don’t know exactly where the United States is. I fervently hope that a lot of this work is taking place in a classified setting,” said R. Paul Stimers, a lawyer at K&L Gates, a Washington law firm, who specializes in emerging technologies. “It is a race.”



Pure quantum computers remain largely theoretical although simple prototypes exist. Many designs call for them to operate in super cold conditions, bordering on absolute zero, or around minus 458 degrees Fahrenheit, colder than outer space, without any noise or micro movements that can cause malfunction.

What has made them the Holy Grail for nations and private industry is that quantum computers, in theory, are magnitudes better at sifting huge amounts of data than the binary processors that power mainframes, desktops and even smart phones today. They also can process algorithms that break all widely used encryption.

Rather than doing a series of millions of computations, based on binary options of ones and zeros, quantum computers employ particles that exist in an infinite number of “superpositions” of the two states simultaneously, a condition that towering physicist Albert Einstein once labeled as “spooky.”

A quantum computer “can feel all the possibilities at once,” said Warner A. Miller, a physicist at Florida Atlantic University, who, like the others, spoke last week at a forum on quantum computing at the Hudson Institute, a think tank in Washington.

China splashed into the news in June when it announced that a satellite and a ground station had communicated through “entangled” quantum particles. Entangled particles, even if separated by thousands of miles, act in unison. Any change in one particle will induce a change in the other, almost as if a single particle existed in multiple places at once.

Such long-distance quantum communication smashed records, occurring over 745 miles, far beyond the mile or so scientists had tested previously, and signaled Chinese mastery over a form of communication deemed ultra-secure and unhackable.

“I read that on a Sunday and went, ‘oh sh-t,’” said Gregory S. Clark, an Australian-born mathematician who is chief executive of Symantec Corp., a global cybersecurity company with headquarters in Mountain View, California.

Neither the U.S. military nor private industry is known to have such a capability.



If the technology is refined, Clark said, it could make land-based communications infrastructure obsolete. “The whole world changes,” he said at a forum Sept. 19.

In early September, China chalked up another milestone, completing a quantum communication link between its capital and Shanghai, by far the biggest such link in the world, surpassing anything in the United States or Europe.

In such a link, if an encryption key used by either of two parties faces interference by a third party, the two parties know not to use it.

China again demonstrated the prowess of its space-based quantum satellite, dubbed Micius, on Sept. 29 when the head of the Chinese Academy of Sciences held a video conference with an Austrian scientist over a distance of 4,630 miles.

Also last month, China announced that it would build the world’s biggest quantum research facility, a $10 billion center in Hefei, capital of Anhui province, with the aim of building a working quantum computer that could break most any encryption within seconds.

China already has the world’s fastest supercomputer, the Sunway TaihuLight, which captured the title in the 2016 and 2017 at a competition in Frankfurt, Germany.

Monroe, the Maryland physicist, said China had set a goal of fully constructing the quantum research center within two years.

“If it costs $10 billion, China will just do it without asking, and they’ll put an army together to do it,” Monroe said. “I don’t think any other government in the world is able to throw together something (so) fast.”

Google, IBM and Microsoft all see huge opportunity in quantum computing and fund research labs. Commercial applications may include determining how polymers go together, mapping the genome, finding oil in complex geology, detecting cancer and handling air traffic.

Quantum computers can sift through vast amounts of data. One that handles 60 quantum bits, or units of quantum information, could hold 64 exabytes of data – 2,560 times more than all the material managed by the Library of Congress, which has 838 miles of bookshelves.

Military applications are vast and range beyond breaking enemy encryption to include quantum-enabled weaponry, navigation systems that can’t be jammed, and the use of quantum-powered artificial intelligence in war fighting.

In those areas, China is not believed to have an advantage.

“The point is, they are some distance from that quantum supremacy threshold,” said Arthur Herman, who leads the technology and defense program at the Hudson Institute.

Still, Herman called for U.S. policymakers to focus hard on the quantum challenge.

“We need a Manhattan Project style funding focus in order for a national quantum initiative to succeed,” Herman said, referring to the World War II era program to produce the first nuclear weapon.