Compared with two years ago, Samsung’s 3nm attention this year has been significantly reduced. It has its own reasons, and it may be more likely to be changes in the technology market.
Recently, TSMC finally publicly acknowledged its 3nm plan, and said that specific technical details will be announced in April. Finally, the key battle in the semiconductor manufacturing industry that determines the future process direction – the 3nm technology battle is here.
So far, on the 3nm process, only two companies in the industry have this competitiveness, one is TSMC and the other is Samsung.
In fact, before TSMC, Samsung has already announced the details of its 3nm process, and it has been leading the way in 3nm. Recently, it has just announced that it has successfully developed the first 3nm process chip. The technology used is the GAAFET technology that was rumored before, and said that its technical completion has far exceeded expectations.
It can be said that judging from the announced news, Samsung is much more reliable than TSMC. But somewhat unexpectedly, after the news from Samsung came out, the response was mediocre, and it did not seem to have received much attention.
What’s wrong with Samsung’s 3nm plan?
Samsung adds code manufacturing and pushes 3nm mass production
First of all, let’s see how powerful it is to win 3nm?
If Samsung can successfully push the process to mass production and continue Moore’s Law, from the perspective of the existing technology pattern, it can completely PK the FinFET-based 3nm process in one fell swoop, and turn over and steal TSMC’s existing cutting-edge process market share and have the opportunity to grow into the world’s largest wafer foundry.
Therefore, for Samsung and TSMC, the 3nm battle is more like a race against time to see who will mass-produce first. And Samsung can be said to have taken the lead.
As early as 2018, Samsung announced its plans for 3nm process technology and said it would seize the future high-performance computing and IoT markets.
At that time, Samsung gave a clear technical plan, and Ryan Sanghyun Lee, vice president of Samsung’s foundry business market, also publicly stated that Samsung has been developing GAA technology since 2002. Later, at the “2019 Samsung Foundry Forum” held in the middle of last year, Samsung Electronics directly sent the 3nm engineering design kit to the on-site semiconductor design companies.
Figure ｜ Ryan Sanghyun Lee, Vice President of Marketing of Samsung Foundry Business
It can be said that it is precisely because of the long-term and huge investment in 3nm that Samsung can advance the results to the eve of today’s mass production. Just like the determination and perseverance to make storage at that time, Samsung’s persistence in the foundry business is actually somewhat unexpected.
During the period of 2015-2016, Samsung once took orders from many major customers of TSMC and achieved a substantial increase in revenue. Later, despite the decline in sales due to the decline of smartphones and TSMC’s advanced manufacturing process, it fell to the fourth place in the world. However, Samsung still has no momentum to give up. In early 2018, at the Samsung Foundry Forum held in Seoul, South Korea, Samsung stated that the goal is to surpass UMC and GF, and then surpass TSMC.
From the recent ranking in the fourth quarter of 2019, we can see that Samsung has continued its advantages in the first three quarters, occupying the second place after TSMC with a market share of 17.8%. GF of three is only 8%.
Some analysts said that in order to compete with TSMC for the right to speak in the advanced process technology market in recent years, Samsung has made great efforts – investment, independent foundry business, poaching, etc. To achieve the current results, Samsung’s foundry technical capabilities and customer recognition have contributed to the improvement.
In comparison, although the status of TSMC in the existing market is difficult to shake, in terms of 3nm technology, there is only news of the factory, and the official has not disclosed more, and its 3nm plan looks more “illusory”.
3nm GAA Tough Battle
So back to the issue of Samsung’s 3nm not getting attention, it’s hard not to take a second look at the technology.
Samsung uses 3nm GAA.
Previously, the Belgian Microelectronics Research Center published a research report and publicly stated that the gate-all-around (GAA) transistor will be the most likely “candidate” to break through the FinFET process below 7nm technology in the future.
GAA transistors have a gate on all sides of the channel to overcome the physical scaling and performance limitations of FinFETs, including supply voltage. Compared with the current FinFET Tri-Gate tri-gate design, GAA technology overcomes the physical and performance limits of the current technology due to the redesign of the underlying structure of the transistor, enhances gate control, and greatly improves performance.
In this technical direction, there are four main directions: nanowires, plate-like structure multi-channel bridge fins, hexagonal cross-section nanowires and nanoring technology. Samsung uses MBCFET (Multi-Bridge Channel FET), that is, plate Flake structure multi-way bridge fins.
Samsung believes that the mainstream nanowire GAA technology has a small channel width, so it can only be used for low-power designs, and the manufacturing difficulty is relatively high, so this solution is not adopted.
According to Samsung, it offers four different schemes in its PDK design, which can be used in different parts of a chip or directly used to manufacture the entire chip. In terms of performance, compared with the 5-nanometer manufacturing process, the 3-nanometer GAA technology has improved logic area efficiency by more than 35%, reduced power consumption by 50%, and improved performance by about 30%, with obvious advantages.
In fact, in addition to Samsung, Intel, TSMC and other manufacturers have not clearly announced the research on GAA technology, but they have also researched and accumulated it. However, judging from the current news, the information released by Samsung’s 3nm GAA process is the most detailed.
Different from the existing 7nm and 5nm processes, because the 3nm node will use a MOS tube with a new structure, it can be said to have a milestone significance for the development of the semiconductor process, very similar to the 22nm used by Intel to overthrow IBM’s leadership at that time. .
Previously, an article on Semiengineering said that, based on the current technical conditions, 2nm can be regarded as 3nm with a die-shrink process. In other words, in the battle of the process below 3nm, 3nm is the key node. Therefore, we can also understand why everyone thinks that once Samsung 3nm GAA is developed, its market influence will be unprecedented.
However, as an industry leader, TSMC’s 2nm and even 1nm news will inevitably leave more thoughts on the market, which also invisibly weakens the influence of 3nm GAA. Some people even think that 3nm will continue to use FinFET, and the so-called key Nodes will be 2nm or even 1nm.
Controversial Samsung and Technology
Of course, it is undeniable that Samsung’s plan has its own hidden dangers.
First of all, the technology is good, but if it is put into commercial use, the cost has to be considered. Strictly speaking, if the cost is too high, it is not impossible that the technology is stillborn.
According to IBS statistics, the research and development cost of 3nm devices is about 500 million to 1.5 billion US dollars; the research and development cost of the process requires 4 to 5 billion US dollars; the construction and operation of a FAB requires 15 billion to 20 billion US dollars, and there is no cap. The cost from research and development to commercial use, including the consideration of cost recovery in the later stage, is a “bottomless pit” for the head foundry, and the risk is extremely high.
Second, in addition to the turmoil in the technology market, the recent acceptance of Samsung’s foundry in the market is quite controversial.
Since Samsung split the foundry department, due to the change of calculation method, Samsung’s own Exynos mobile phone chip production is also included in the foundry revenue, so the market share of Samsung foundry has increased significantly from single digits. to ten digits. It is precisely because the data dimension has changed, which makes the industry doubt whether its performance has really grown. Therefore, whether Samsung’s second place is really stable has become a “confused account” that everyone has no way to figure out.
In November, Samsung’s Giheung plant in South Korea had defective products due to contaminated equipment used in its 8-inch wafer production line. A Samsung executive later admitted that the damage was estimated to be in the billions of won.
On December 31, there was another power outage in Hwaseong Park, South Korea. Some DRAM and NAND chip production lines of Samsung Electronics were forced to suspend, saying that it would take about two or three days to fully recover, and the losses were also heavy.
. . .
One after another, Samsung’s foundry has repeatedly suffered accidents, which has damaged its reputation in the industry.
At the same time, at present, GAA is not the only technical direction. The advanced packaging proposed by TSMC is a conservative approach to promote the development of the process, and the exploration of new materials other than silicon is considered by the industry to fundamentally change the existing process limitations. direction, which includes GAA transistors using new materials such as InGaAs and germanium nanowires.
And TSMC directly discloses the 2nm process after 5nm, which is more likely to shake the status of the key node of 3nm that is generally considered in the market.
Undoubtedly, these have blocked Samsung’s road to breaking through TSMC’s monopoly.
With the blessing of new materials, there is still more room for imagination in the 3nm process
According to market research firm IC Insights, from 2017 to the present, Samsung’s capital expenditure in the semiconductor field is estimated to be US$65.8 billion, which is about 53% higher than Intel’s and more than double the capital expenditure of all semiconductor companies in China combined.
In terms of foundry, Samsung also recently announced that most of its investment in the fourth quarter of this year will be used in infrastructure in the storage field, and EUV 7nm production will continue to increase to strengthen its own foundry competitiveness. In mid-October, a letter of intent from Samsung was also reported, indicating that it would order 15 EUV equipment from ASML, with a total value of 18 billion yuan.
Samsung’s determination can be seen here.
However, with the changes in technology and the development of the market, the variables of the future process technology in the market are also becoming larger and larger, so it is becoming more and more difficult for Samsung 3nm GAA to obtain the expected effect of defeating the enemy. TSMC once stated that in terms of materials, III-V materials may also replace traditional silicon as the channel material of transistors to increase the speed of transistors. Because studies have shown that the integration of indium gallium arsenide (InGaAs), gallium arsenide (GaAs) and indium arsenide (InAs) with FinFET and GAAFET shows excellent performance at smaller nodes; while dielectric materials such as ferroelectric The introduction of tungsten may enable ultra-steep subthreshold slopes to reduce transistor power consumption; cobalt may also replace tungsten and copper as interconnect wires to enhance stability and slow down signal delays.
At present, “new materials + new process technology” walking on two legs is expected to break the existing technical bottleneck and has become the mainstream argument. Judging from the published information, Samsung’s 3nm GAA only has progress in the process. Compared with TSMC’s current layout, it lacks the accumulation and exploration of materials. Therefore, the later the release, the smaller the first-mover advantage effect may be.