Advertisement
Advertisement
Semiconductors
Get more with myNEWS
A personalised news feed of stories that matter to you
Learn more
A visitor views exhibits during the 17th China International Semiconductor Expo in Shanghai, Sept. 3, 2019. Photo: Xinhua

Explainer | What semiconductors are and why China needs to make them itself

  • While TSMC is moving to 5nm production, SMIC’s most advanced process node is 14nm – which contains half as many transistors as the 7nm node
  • HiSilicon was important to Huawei because it enabled the company to design custom processors that were not off the shelf products available to its competitors

Semiconductors, also known as microchips, are silicon-based devices made up of hundreds of millions, and in some cases billions, of transistors, which act as tiny “switches” to control the movement of electrons.

Today’s most advanced chips are produced using what is called the 7-nanometre manufacturing process, where one nanometre equals about one billionth of a metre. Only two companies in the world - Samsung and TSMC - are able to produce 7nm chips in volume. Smaller nanometre process nodes are important because they boost circuit performance and reduce power consumption.

Although China imports US$300 billion worth of semiconductors annually – more than half of which are re-exported in finished electronics products – it is far behind the tech curve when it comes to making them.

The imposition of US sanctions that cut off Huawei’s access to advanced chips has accelerated Beijing’s desire to become self-sufficient in semiconductors, but the goal will not be easy and could take decades, according to experts.

Here is all you need to know about semiconductors and their critical role in the balance of power in the US-China tech war.

What is a semiconductor and how are they made?

A semiconductor is a substance that conducts electricity under some conditions but not others, making it a good medium to control electrical current. Most of the world’s semiconductors, also known as integrated circuits (ICs) or microchips, are made from pure elements such as silicon. The raw material for silicon is sand, which is purified and melted into solid cylindrical ingots, which are then cut into thin disks or wafers ready for processing into finished chips.

Silicon wafers undergo a series of highly complex and intricate manufacturing steps (such as etching and diffusion) in a factory called a wafer fab. The fabrication process, which can require hundreds of steps, lays down patterns of various materials on top of one another. The “recipe” used to make chips varies depending on their intended end use.

The individual chips (called die) are then cut up and packaged into finished semiconductors that are embedded into electronic devices like smartphones, TVs, computers and medical equipment.

There are three business models in the semiconductor industry. Companies that focus only on design (eg, Qualcomm, Nvidia) are referred to as “fabless” firms, those that focus only on manufacturing (eg, TSMC, SMIC) are called “wafer foundries”, while firms that do both are called Integrated Device Manufacturers, or IDMs (eg, Intel, Samsung).

There are two other types of companies involved in the semiconductor supply chain. Electronic design automation (EDA) firms (eg, Cadence, Synopsis) develop the complex software used to design integrated circuits, while the so-called “back end” companies (eg, Amkor, ASE) do the assembly and testing of the individual chips after they leave the wafer foundry.

China to curb ‘current chaos’ in semiconductor industry with more oversight

What are the most commonly used types of chips?

Microchips can contain billions of transistors that control how electricity flows in the device. The arrangement of transistors on the silicon will determine the function of the chip.

For example, the central processing unit (CPU) and graphics processing unit (GPU) are computing engines used for processing huge amounts of data. However, they each have a unique architecture and are built for different purposes.

CPUs, the “brains” of all computers, execute the commands and carry out the processing needed for computers and operating systems. They are suited to a wide variety of workloads and have the advantage of versatility, multitasking and ease of programming.

GPUs, designed with thousands of processor cores running simultaneously, can process data several orders of magnitude faster than a CPU, but they are not as versatile. Originally designed for graphics, GPUs are now used in a wide range of computationally intensive applications such as gaming, artificial intelligence and machine learning.

Random-access memory (RAM) chips are used to store data, with dynamic RAM (DRAM) being the most widely used in computers. They look like a series of temporary storage areas connected together.

Chip maker SMIC’s founder says ‘optimistic’ China can catch up with US in semiconductors

But unlike a hard drive, which stores data indefinitely, RAM resets every time the system is rebooted. Computers need quick access to temporary data to run programs and execute tasks. Thus, applications read and write data to the RAM, which is much faster than accessing data from a storage device.

Why is smaller more powerful when it comes to semiconductors?

In 1965 Gordon Moore, who would later co-found US chip maker Intel, wrote a paper that predicted the number of transistors on an integrated circuit would double roughly every two years. Moore‘s Law has held since despite many experts predicting it would hit a brick wall.

Today, the most advanced chips contain more than 7 billion transistors and are produced in factories called wafer fabs using what is called the 7-nanometre manufacturing process. One nanometre is equal to about one billionth of a metre. There are only two companies in the world producing 7nm chips in volume: South Korea’s Samsung Electronics and Taiwan-based TSMC. In July US chip giant Intel said it was facing a six month delay in its transition to the 7nm process.

Intel and Samsung are known as integrated device makers (IDMs) because they design and manufacture their own chips. (Samsung is vertically integrated because it also makes end-products that use its own chips). TSMC is an independent wafer foundry that produces chips for companies that do not have their own wafer fab. Known as fabless chip makers, these design-only companies include Qualcomm and Nvidia in the US and MediaTek from Taiwan.

Smaller nanometre process nodes are important because they boost circuit performance and reduce power consumption, meaning fewer battery charges for products like smartphones. Each new process node usually boosts speed by around 20 per cent and cuts energy use by 40 per cent, according to TSMC.

The next technology node is 5nm which began volume production at TSMC in the first half of 2020, while TSMC’s road map has it moving to 3nm in 2022.

A new wafer fab capable of 5nm production is expected to cost US$15 billion, twice as much as a 14nm fab, according to data from International Business Strategies, cited by SMIC in its recent prospectus.

As for Moore’s Law, TSMC general manager Luo Zhenqiu said in August that he believes it will still hold true for 3nm, 2nm, even down to 1nm process technology.

How far behind is China in semiconductors?

Although China imports US$300 billion worth of chips annually – about US$160 billion of which are re-exported in finished electronics products – it is a laggard when it comes to making them.

China’s most advanced chip maker is Semiconductor Manufacturing International Corp (SMIC), established in 2000 by Richard Chang, a former wafer fab specialist at Texas Instruments.

SMIC operates on the same business model as TSMC – an independent wafer foundry producing chips for fabless chip makers. Shanghai-based SMIC is heavily dependent on foreign technology to make its chips and was recently put on a US watch list that required American equipment suppliers to apply for a license before they could ship to the Chinese company.

China sees a chance to close US core tech gap with AI chips in 5G era

SMIC invests about 70 per cent of its revenue on equipment and materials imports from US, European and Japanese suppliers, according to industry analyst Andrew Wang.

However, while TSMC is moving to 5nm production, SMIC’s most advanced process node is 14nm – which contains half as many transistors as the 7nm node.

Under the Wassenaar Arrangement, the US and its allies restrict the export of dual use technologies to communist countries, meaning SMIC and other mainland Chinese chip makers cannot access the very latest chipmaking technology. For example, the most advanced lithography machines, called EUV scanners (because they use lasers operating at extreme ultraviolet wavelengths) have been blocked for shipment to China. EUV technology is critical for the 7nm and 5nm process nodes.

SMIC’s logo is seen on its factory building in Shanghai. Photo: EPA-EFE

That means SMIC can only use the earlier generation machines, called DUV (operating at deep ultraviolet wavelengths), which cannot produce chips more advanced than 14nm.

Although the competitiveness of China’s semiconductor vendors have improved greatly in recent years, the industry still relies heavily on key components from the west, resulting in a self-sufficiency rate of less than 20 per cent, according to a report from Deloitte.

China’s domestic production of integrated circuits accounted for 15.7 per cent of the chips that the country needed in 2019, according to research firm IC Insights.

In 2018, the world’s top 15 semiconductor companies by revenues originated from the US, Taiwan, Europe, Japan and South Korea respectively. No mainland Chinese company was on the list.

Why is European company ASML so important in the semiconductor industry?

The most critical stage in the chipmaking process is lithography, where light is used to print the tiny circuit patterns onto the silicon wafer. The light is projected through a blueprint, called a mask, and optics in the system shrink and focus the pattern onto a photo sensitive silicon wafer.

ASML of the Netherlands is currently the only company that can supply the most advanced lithography systems needed to produce chips at the 7nm and 5nm nodes.

Called EUV scanners, these machines perform imaging in a high vacuum instead of air, work with ultra-flat multilayer mirrors instead of lenses, and generate the light by vaporising droplets of tin with a high-power laser.

EUV lithography uses light with a wavelength of just 13.5 nanometres (nearly X-ray level), a reduction of almost 14 times from DUV (deep ultraviolet) lithography, which uses 193-nanometer light. DUV machines are available from Japanese suppliers Nikon and Canon but ASML is the only supplier for EUV, which is not available to Chinese chip makers due export restrictions imposed by western governments.

While TSMC is moving to 5nm production, SMIC’s most advanced process node is 14nm – which contains half as many transistors as the 7nm node. Source: SCMP

How difficult will it be for China to catch up to the west in semiconductors?

In the 1990s, China built two wafer production lines with help from Japan. Shougang Group, a Chinese state-owned steel conglomerate, emerged as the nation’s first six-inch wafer fab producer in 1995 after establishing a joint venture with NEC Corp. Later, Shanghai Huahong Semiconductor partnered with NEC to build China’s first eight-inch wafer production plant in 1999.

Both ventures failed to make a dent in the global semiconductor market, largely due to production inefficiencies and a rigid top-down bureaucracy that did not work well in the fast moving semiconductor industry. In the early 2000s, China’s most advanced wafer fab, SMIC, was established as a private venture with Chinese and international funding.

However, wafer fabs in China still rely heavily on foreign manufacturing equipment and materials from US suppliers like Applied Materials and Lam Research.

During celebrations this month to mark the 40th anniversary of Shenzhen’s designation as China’s first special economic zone, Chinese president Xi Jinping said the country “must take a path towards self-reliance”, which means “becoming independent in our innovation drive.”

A Huawei Kunpeng 920 chip is displayed during an unveiling ceremony in Shenzhen, Jan. 7, 2019. Photo: AP

Wei Shaojun, a Tsinghua University professor in the department of microelectronics and nano-electronics, said the idea of “starting from scratch” to achieve self-reliance in chips was “unrealistic” because the global technology supply chain was already highly integrated.

China will not be able to come up with an alternative to equipment from Applied Materials and Lam Research anytime in the next 10 years, according to an analyst at the Hong Kong office of a global investment bank, who requested anonymity due to the sensitive nature of the topic.

“I think fabless semiconductor companies in China have a much higher chance of success than manufacturers or equipment vendors in most cases. The main challenge though is if you lose access to global foundries such as TSMC,” the analyst said.

Officially, the Chinese government has never stipulated that chip making technologies should be of Chinese-origin. Rather, it has emphasised the need to attract foreign capital, technology and talent, encouraging domestic enterprises to expand their international cooperation, especially with companies from Taiwan.

“It’s very difficult for China to leapfrog in semiconductor technologies or innovate business models if it were forced into isolation,” said Li Pengfei, a researcher at the Institute of Industrial Economics under the Chinese Academy of Social Sciences. “The US and other developed countries have an edge over China in EDA [electronic design automation] tools, manufacturing equipment and materials.”

Why are semiconductors important for Huawei (and what is HiSilicon)?

As a manufacturer of electronic systems, Huawei needs the most advanced chips to make its product competitive. For example, the 7nm process enabled Huawei to achieve a 20 per cent improvement in performance and 40 per cent greater power efficiency for its Kirin 980 smartphone processor compared with the earlier Kirin 970 chip, according to Richard Yu, president of Huawei’s consumer business unit. The Kirin 980 contained 6.9 billion transistors, about 1.6 times more than those in the 970. Both chips were fabricated by TSMC.

With US sanctions cutting off Huawei’s access to TSMC, Huawei consumer business group CEO Richard Yu said in August that the Kirin 900 series may be the end of the line. “This year may be the last generation of Huawei Kirin high-end chips … This is a big loss for us,” he said.

The Kirin 900 series is not only used to power Huawei’s flagship phones but is also found in the company’s mid-to-high level Honor brand. HiSilicon, which is a fabless company using Arm IP, also designed the Kirin processors used in Huawei’s smartphones.

In January 2019 Huawei announced the HiSilicon-designed Balong 5000 chipset which supports 2G, 3G, 4G, and 5G phones. Used by many local modem makers, the Balong 5000 was fabricated by TSMC using its 7nm process node, according to the Huawei Report website, which is not affiliated with the company.

A chip designed by Huawei's HiSilicon subsidiary is seen on display at Huawei headquarters in Shenzhen. Photo: Reuters

As for its 5G base stations, Huawei previously used chips from US fabless companies like Xilinx, but that source was cut off due to US sanctions. While Huawei stockpiled US chips in anticipation of the ban it also directed its wholly-owned subsidiary company HiSilicon to design new chips for the base stations.

In the first quarter of 2020 HiSilicon’s sales surged 54 per cent year on year to about US$2.67 billion, making it the No 10 fabless chip maker, even though 90 per cent of its sales went to its parent Huawei, according to US research firm IC Insights.

HiSilicon was important to Huawei because it enabled the company to design custom processors that were not off the shelf products available to its competitors. The latest US sanctions on Huawei and its affiliates like HiSilicon came into effect on September 15, meaning TSMC has stopped making chips for Huawei.

For the foreseeable future, Huawei is relying on stockpiled chips to keep its product line going.

What is China’s chip industry likely to look like in the future?

Despite the industry’s heavy reliance on foreign technology, some analysts think China has a chance to catch up because of its huge market potential.

In a report last year Morgan Stanley said the fundamental questions to ask were: Should China be investing in chip design or increasing its production ability; and is localisation or globalisation the most effective way to achieve self-sufficiency?

How China’s semiconductor insiders rate the nation’s self-reliance drive

The bank concluded that the key to advancing China’s semiconductor industry lies in chip design rather than pursuing large-scale manufacturing. On the second question, it said the current tendency was towards localisation over globalisation.

In terms of specific chip designs, China is making progress in networking chips and power semiconductors for smartphones and wireless base stations, according to an analyst at the Beijing office of a global investment bank, who requested anonymity due to the sensitive nature of the topic.

“In microprocessors, it’s definitely a lot tougher for China to be independent. It depends if you still have access to ARM IP or not,” said the analyst.

Additional reporting by Celia Chen

4