In-depth analysis of the semiconductor industry's fortunes in 2017

Author: Mo Dakang

As the new year begins, let’s talk about the fortunes of the global semiconductor industry.

Normally, data will not be available until March this year, but the IEDM at the end of last year , the CES at the beginning of each year, the annual Semiconductor Industry Strategy Conference ( ISS ) organized by SEMI in Half Moon Bay, San Francisco, USA ( 9-11 ), and TSMC's Q4 earnings conference have all ended, so the basic reference points may already be available.

According to published data, Gartner predicted that semiconductors will grow by 4.7% year-on-year in 2017 , reaching 340 billion US dollars. Deutsche Bank predicted in December last year that semiconductors will grow by 5% this year . WSTS predicted in the fall of November last year that semiconductors will decline by 0.1% year-on-year in 2016 and grow by 3.3% in 2017. SEMI predicts that the global semiconductor industry will recover from last year's weakness and resume strong growth in 2017 , with the continuous impact of 3D NAND Flash and wafer foundries on production capacity. The annual output value is expected to exceed 360 billion yuan, an annual increase of 5-7% . In addition , Morris Chang predicted that semiconductors will grow by 4% , foundry will grow by 7% , and TSMC will grow by 5%-10% in 2017 .

➤ What is the driving force?

Industry forecasts must be based on data, not guesswork.

TSMC said in its Q4 earnings conference that global smartphone shipments grew 6% year-on-year , with high-end phones growing 3% , mid-range phones growing 5% , and low-end phones growing 8% . Computer shipments fell 5% year-on-year , tablets fell 7% , and IoT product shipments grew 34% . Gartner predicts that global shipments of personal computers, tablets, ultra-mobile products, and mobile phones will total 2.3 billion units in 2017 , the same as in 2016 .

DRAMexchange predicts that the average price of server memory modules will increase by 25% in the first quarter of 2017 , and the high-capacity ones will increase by more than 30% . DDR4 32Gb has exceeded $ 200 , and 16GbRDIMM has also climbed to $ 100 . The memory market rebounded from the bottom in June 2016 , rising from the lowest $ 1.31 per unit to $ 1.94 at the end of the year . With limited supply and increased demand, memory is expected to continue its upward trend this year .

IC Insight predicts that the global memory market will have an average annual growth rate (CAGR) of 7.3% in the next five years (2016-2020) , increasing from USD 77.3 billion in 2016 to USD 109.9 billion in 2020 , indicating that the future of the memory market is bright.

However, Morris Chang made it clear in the Q4 earnings conference that the foundry industry in 2017 is flat , which means that global foundry business this year may not be as good as last year, which is worthy of our vigilance.

➤ How to interpret the " fortune " of the semiconductor industry in 2017

Different people have different opinions on the fortune of the industry. The general view is that the performance of the semiconductor industry in 2017 is neither outstanding nor pessimistic. It may be affected by the global political situation, and there will be more uncertainties, especially the progress of China's semiconductor industry.

In the terminal electronic product market , computers and tablets are still declining, and the growth of smartphones has been weak. New market highlights, such as AR/VR, autonomous driving, smart homes and cities, including the Internet of Things, are very attractive, but the market is fragmented, and none of them can be compared with the volume of computers and mobile phones, which can be held by individuals. In the final analysis, there is still a lack of large-scale and wide-ranging terminal electronic product markets in the near future. Either the volume is not large enough, or, like the Internet of Things, although their volume is large enough, the silicon content is relatively small.

In terms of process technology , although 10nm has been mass-produced and 7nm is on the way, the investment amount is too large and it takes a longer time to ramp up (some people think that the yield rate is not high for the time being). As for new directions, such as 3D NAND , new alternative memory, EUV lithography, TSV , 2.5D , 3D packaging, etc., they are all technical "hard bones" that require money, time and patience.

The continuous advancement of semiconductor manufacturing technology has not been smooth sailing. There have been many times when it was thought that it might not be possible, but then it "saw the light again". At this point, 7 nanometers is no longer a suspense, and people may still be discussing what to do with 5 nanometers? But some people have already said that it can reach 1 nanometer in the future ? It's incredible.

TSMC 's 7nm process has been put into trial production in the first quarter of this year. Its main customers include FPGA giant Xilinx and graphics chip giant NVIDIA . It is also widely rumored that Qualcomm's high-end mobile phone chips will return to TSMC and use 7nm wafer production. TSMC announced that it plans to complete capacity construction and certification in the fourth quarter of 2017 , and then enter the mass production stage, so it will enter 7nm mass production in 2018 .

Industry insiders said that Intel was already 6-9 months behind in developing 14nm , and the Moore's Law it has been advocating for many years has been abandoned. Due to the addition of optimization programs, Intel's current process update cycle has also been extended from the original two years to three years.

Despite this, Intel is still at least two quarters behind TSMC in the race to 10nm. According to previous media reports, TSMC has successfully trial-produced its 10nm process in the second quarter of 2016 and is expected to lead its peers in mass production in the fourth quarter. Intel's Cannon Lake Tick 10nm will be mass-produced by the end of 2017 .

Thanks to TSMC's annual investment of nearly US$10 billion and a large number of R&D personnel, it is expected to enter 5nm mass production in 2020 .

相较之下，台积电的 10 纳米、 7 纳米制程技术虽然可能落后于英特尔，但台积电比英特尔提前 1-2 年跨入 7 纳米制程，可藉此缩短两家公司的差距。台积电在独家封装技术 “整合型扇出型封装” (integrated fan-out ， InFO) 的协助下，有望在 2017 年、 2018 年霸占 10 纳米和 7 纳米的晶园代工市场。

Scotten Jones , president of IC Knowledge, believes that many logic chip manufacturers will enter the 5- nanometer process in the near future. Among them, TSMC may arrive first in the second half of 2019 , but soon everyone will be on par.

Gary Patton , CEO of GF, believes that although Moore's Law may end, we can always find a way to move forward in the future.

As devices push into the atomic-layer limit, Gary Patton predicts chipmakers will adopt a gate-all-around transistor architecture around 2020 .

In the second half of 2019 , the semiconductor industry may enter the 5nm era using EUV . EUV will be used in at least some processes , but the finFET transistor architecture will still be used . However, the next generation of 3.5 nanometers may use horizontal nanowires, which means the end of the era of traditional process size reduction.

Nevertheless, Coventor 's process model experts believe that the use of stacked n- nanowires and p -nanowires at the 2.5- nanometer generation can still provide a 60%-70% increase in transistor density, indicating that continued industry progress can continue until 2025 .

ASML CEO confirmed that 13 EUV units were shipped in 2017 , including one order in 2016, 24 in 2018 , and 40 in 2019 , with each unit priced at around 100 million euros. The CEO further confirmed that demand for both logic and memory fabs improved significantly in 2017 , so the company believes that sales in 2017 will have " a significant growth ")

The industry has high hopes for EUV lithography, but the technical progress of EUV has been repeatedly delayed. Although the prospects seem very bright, no one can predict whether it can really enter mass production in 2018 ? EUV needs to be at least 4th pass patterning (SAQP) to be cost competitive with 193nm immersion lithography . Therefore , when EUV lithography truly becomes a mass production device in 2020 , it is estimated that 7nm - 5nm will become the main application in the market, which will drive the global semiconductor industry to another new climax.

In 2017 , apart from the price increase of memory chips, there are not many bright spots in other areas for the time being. The growth of fabless has shown signs of decline, and the prosperity of the foundry industry may not be as good as before. Although TSMC is in the lead, Samsung, Intel, SMIC and others will join in, so the competition in global foundry will be more intense.

In the next three years: Semiconductors will shift from a "slight decline / modern prosperity" to a "steady growth" . By 2020 , the global semiconductor market size will likely exceed US$ 500 billion, and the progress of China's semiconductor industry will play an important role in promoting it.

➤ China's semiconductor industry requires patience , time and persistence

2017 is a critical year for China's semiconductor industry. In 2015 , the industry was led by mergers and acquisitions , which accumulated enough upward momentum. In 2016 , the chip production line construction peaked, and many memory and 12- inch production line projects started construction. Therefore, some projects will enter mass production or equipment installation this year, which means a large investment stage. More importantly, breakthroughs in process technology are needed to prepare for the mass production of these production lines in 2018 .

For the current high tide of line construction, it is important to create momentum, but calmness is even more indispensable, because from now on, real investment will begin and the probability of risk will increase. Any expansion of production capacity is relatively simple as long as there is money. The key is to turn the capacity into "effective capacity". Obviously, the more effective capacity, the more popular it is, so it must be able to win in the global competition.

For example, 28 nanometers, HKMG 's logic foundry, and 32- layer 3D NAND mass production are all key technology nodes. We must have an objective basic understanding. It is unlikely to be achieved overnight. It requires patience and time. It is determined by a combination of factors. There is no shortcut. However, there is no retreat for China's semiconductor industry. We can only face the challenges head-on, so backbone enterprises must take responsibility.

2017 is a critical year for China's semiconductor industry. There are more uncertainties because our opponents have many opportunities to play their cards. Any card they play will have a big impact on China's semiconductor industry. Obviously, our opponents will also get corresponding "rewards."

The Chinese semiconductor industry must first have a plan for what to do if the worst-case scenario occurs. In summary, there are two basic points. First, do not be afraid, because it has happened before, and fear cannot change the status quo. On the contrary, it may become a "good thing" and force us to insist on self-reliance and strive for strength. Second, do not waver, let alone retreat, because this is also one of China's national policies.

There is no doubt that China's semiconductor industry will develop rapidly in the future, but it requires sufficient patience and time. The real test is that after the production line's capacity is built, it can become "effective capacity" and survive continuously.

Author: Mo Dakang