When the semiconductor process develops to the 28nm node, the perfect balance between chip performance and cost is achieved. However, 14nm, 10nm, 7nm, 5nm, and the 3nm process that can be mass-produced in 2022 are constantly refreshing the industry’s perception of advanced process technology. The iteration speed of advanced process technology after 28nm has exceeded the development speed of all previous process nodes. There are fewer and fewer manufacturers achieving mass production, but there are more and more requirements for their products.
The rapid iteration of portable Electronic products represented by smartphones has put forward higher requirements for the PPA of chips, breaking the balance between performance and cost. In order to obtain chips with smaller size and higher performance, the cost has been reduced by some manufacturers. Suppliers are placed in a secondary position (provided that there is a huge demand for chips, which can dilute the cost of a single chip).
With rising design costs, only a few customers can afford to move to advanced process nodes. Therefore, in terms of unit chip cost, 28nm has obvious advantages and can maintain a long life cycle. On the one hand, compared to 40nm and later processes, the 28nm process has significant advantages in frequency regulation, power consumption control, thermal management and size compression. On the other hand, due to the use of FinFET technology in 20nm and more advanced processes, it is more difficult to maintain high parameter yield and low defect density, and the cost of each logic gate is higher than that of 28nm process.
Although the high-end market will be occupied by 7nm, 10nm and 14nm/16nm processes, 40nm, 28nm, etc. will not withdraw. For example, the 28nm and 16nm processes are still the main revenue sources of TSMC, and SMIC is continuing to improve the 28nm yield.
At present, the 28nm process in the industry mainly competes among the five companies TSMC, GF (GF), UMC, Samsung and SMIC.
After TSMC’s 28nm process was put into mass production in 2011, the revenue share climbed from 2% to 22% in just one year. The rapidly expanding advanced production capacity has helped TSMC to seize customer resources at every advanced process node. Expanding the first-mover advantage and making its production capacity structure significantly better than its competitors has brought higher gross profit margins with higher product added value.
For UMC, which focuses on the development of special processes, 28nm is its key business segment. For this reason, the company has also abandoned the research and development of advanced processes below 14nm.
As the first wafer foundry company in mainland China that has a process technology of 28nm and provides 28nm PolySiON and 28nm HKMG processes at the same time, SMIC has many factories in China, and its 28nm process products are mainly located in two factories in Beijing. It is produced in SMIC North Factory.
In addition, in recent years, the rapid rise of SOI technology is largely due to the vigorous promotion of GF. Industry insiders generally believe that for the SOI process, the 28nm process is more advantageous and can last for a long time, and when the process evolves further, SOI will have more and more advantages. 28nm is a demarcation point. At this point, the process can be easily converted to SOI, and there are more and more EDA tools supporting this transition.
There are mainly 7 manufacturers with or about to have 14nm process capacity, namely: Intel (INTC.US), TSMC, Samsung, GF, UMC, SMIC (00981) and Hua Hong (01347).
At present, the 14nm process is mainly used in the manufacture of mid-to-high-end AP/SoC, GPU, mining machine ASIC, FPGA, and automotive semiconductors. For various manufacturers, the process is also the main source of income, especially Intel, 14nm is its current main process technology, and the revenue it brings can be imagined in terms of the company’s size. For local wafer foundries in mainland China, especially SMIC, the 14nm process technology has already achieved mass production this year. In this way, in two or three years, with the maturity of the new production capacity, the market structure of the 14nm process is worth looking forward to.
Since the official launch of the 14nm process in 2015, Intel has relied on it for 4 years, and the process has also brought very considerable revenue to the semiconductor giant. From Skylake (14nm), Kaby Lake (14nm+), Coffee Lake (14nm++), to 14nm+++ launched in 2018, the company has been keeping the 14nm process updated. Intel originally planned to launch 10nm in 2016, but after many delays, it was late in 2019. From this, we can also see the company’s reliance on the 14nm process.
TSMC (TSM.US) mass-produced the 16nm FinFET process in the second half of 2015. Compared with Samsung and Intel, although their node naming is different, Samsung and Intel are 14nm, TSMC is 16nm, but they are in the same generation at the actual process technology level.
At present, the 16nm process is still the main force of TSMC’s revenue, with a contribution rate of about 25%.
14nm is the most advanced mainstream process technology of GF. It is located in Malta, New York, USA. In addition to 14nm, there are 28nm ones. The maximum production capacity is 60,000 wafers/month, mainly using 12-inch wafers. Mainly used for foundry high-end processors. At present, 14nm production capacity accounts for a small proportion of its total revenue.
In terms of UMC, the company’s Fab 12A in Tainan entered mass production in 2002, and has now used the 14nm process to manufacture products for customers. However, UMC’s 14nm process accounts for only about 3%, which is not its main production line. This is directly related to the company’s development strategy, with UMC focusing on developing special processes.
The mobile phone market in mainland China is huge, and the mid-range and low-end occupy the bulk of the shipments, which gives an excellent opportunity for the development of mid-range mobile phone processor chips, and the market share of the corresponding 12nm process technology has also risen. .
Judging from the current wafer foundry market, there are not many manufacturers with 12nm process technology capabilities, mainly TSMC, GF, Samsung Electronics and UMC. UMC announced in 2018 that it would stop the research and development of 12nm and more advanced process technology. Therefore, at present, in the global foundry market, the main players of 12nm are TSMC, GF and Samsung. This is also evident from the various chips launched in the market in the past two years.
In August 2018, Huawei released the mid-range chip Kirin 710, which uses a 12nm process and was used in the mid-range model at the time, known as Mate 20 Lite overseas.
MediaTek is the main force of 12nm chips, and its representative products are mostly mid-range chips, including: Helio P22, manufactured using TSMC’s 12nm FinFET process; Helio P60; Huge mass market. The first product in the series is the Helio A22.
In mainland China, Ziguang Zhanrui has also made great efforts to expand in the mid-end and low-end markets in recent years, and has gradually expanded its market share. In the upcoming 5G market, the company has launched the Ivy 510, which adopts the TSMC 12nm process technology.
In addition to mobile phone processors, AMD’s graphics card RX 590 is also manufactured using a 12nm process, and the foundries of this product are GF and Samsung. Due to the deep relationship between AMD and GF, many of its chips are found by GF, but with GF’s announcement to withdraw from the research and development and investment of 10nm and more advanced processes, AMD had to distribute advanced products to Samsung and GF. TSMC foundry, thereby dispersing GF’s orders.
So far, only TSMC, Samsung and Intel have publicly announced plans for 10nm and have mass-produced or are about to mass-produce chips at this process node. In addition, SMIC is probably also carrying out research and development work on 10nm and 7nm processes, but the details have not been disclosed.
TSMC started the research and development of the 10nm process as early as 2013. According to the early plan, TSMC’s plan is to mass-produce the 10nm process in the fourth quarter of 2016. The actual mass production time is basically consistent with its plan. Mass production was achieved in early 2017. The iconic application is Apple’s A11 processor, which has brought huge benefits to TSMC.
However, after mass production, the proportion of TSMC’s 10nm revenue is basically the same, and the relative share is not high. 28nm and 16nm have always been the main sources of the company’s revenue.
Intel has also started 10nm research and development long ago. The original plan was to mass-produce it in 2016. At that time, EUV was immature. Therefore, Intel chose the multiple quad pattern exposure (SAQP) technology, but encountered difficulties in the research and development process, resulting in 10nm production. The delivery time has been delayed again and again. Judging from the situation at the time, the low yield caused by the use of SAQP technology may be the main reason for the delay in mass production.
Since then, Intel has not announced the progress of 10nm mass production. At the beginning of 2017, then Intel CEO Krzanich announced that the first 10nm processor Cannon Lake was ready to face TSMC and Samsung before the CES exhibition in the United States. However, it didn’t take long for Intel to officially release the details of the eighth-generation Core processor that inherited the seventh-generation core processor Kaby Lake, which was the mainstay of the year, saying that it will still be produced using the 14nm process, and the 10nm mass production time has been delayed again. After years of setbacks and delays, Intel’s 10nm is finally in mass production by the end of 2019.
In July 2015, Kelvin Low of Samsung Foundry, a manufacturing arm of Samsung Electronics, posted a video online confirming that Samsung has officially added the 10nm FinFET process to its roadmap.
In 2017, after mass-producing the 10nm process almost simultaneously with TSMC, Samsung received most of the orders for Qualcomm Snapdragon processors. However, in order to catch up with the mass production of TSMC’s 7nm process, Samsung’s energy and time on 10nm is relatively limited. In 2019, Samsung’s 7nm process received an order for Qualcomm’s Snapdragon 765. Similar to TSMC, the focus of planning is on the future 5nm and 3nm, and 10nm is also a short-lived.
At present, only TSMC and Samsung can mass produce 7nm chips.
In terms of TSMC, 7nm has trial production at the end of 2017, small batch production in 2018, mass production in 2019, and mass production of N7+ (EUV) in the second quarter of 2019. Compared with N7, N7+ Logic density is 15% to 20% higher than N7 while reducing power consumption.
After the mass production of TSMC’s 7nm process, more than 100 chips will be taped out in 2019, including CPU, GPU, AI, cryptocurrency chips, network communication, 5G, autonomous driving and other chips. Customers include Apple, Huawei HiSilicon, MediaTek, Qualcomm, Nvidia, AMD, Xilinx, Bitmain, etc. Apple’s A13 processor, HiSilicon’s 5G base station chip, and AMD’s GPU, CPU and server chips will all be shipped in the second half of 2019, and they are all striving for TSMC’s 7nm production capacity to keep its related production lines running at full capacity. The delivery time has been extended from 2 months to half a year, and customers have been queuing for production capacity.
At present, TSMC’s 7nm production capacity is still very popular. It is reported that in the second half of this year, TSMC’s 7nm production capacity will increase to 140,000 pieces per month. Among them, AMD occupies a large share. Due to the rapid rise of the company in the past two years, its orders in TSMC have also increased significantly. It is reported that this year, AMD’s 7nm orders will double, requiring 30,000 wafers per month. The round production capacity accounts for 21% of TSMC’s total 7nm wafer production capacity. In addition, Qualcomm accounts for about 18% of TSMC’s 7nm production capacity, and MediaTek will account for 14%.
Compared with TSMC, Samsung’s 7nm capacity utilization rate is much inferior. In the early stage of mass production, it started with a small amount of mass production of about 10K. As the number of orders placed by customers increases, the production capacity continues to increase.
In the early days, in addition to Samsung itself, the only customers of Samsung’s 7nm EUV were IBM. The other two big customers are Nvidia and Qualcomm.
In February this year, Samsung announced that it will open a new wafer foundry production line V1 specializing in EUV technology in Hwaseong Industrial Park, South Korea, mainly for mass production of 7nm. At present, V1 has been put into the production of 7nm and 6nm EUV mobile chips, and it can be foundry to the highest 3nm level in the future. According to Samsung’s plan, by the end of 2020, the total investment of the V1 production line will reach 6 billion US dollars, and the total production capacity of 7nm and more advanced processes will be three times that of 2019.
6nm is a transitional process technology between 7nm and 5nm.
TSMC launched the 6nm process (N6) in April 2019, and the design method is fully compatible with the 7nm process. With the further application of EUV technology, the logic density of N6 will be 18% higher than that of N7. According to the plan, TSMC’s 6nm process will be trial-produced in the first quarter of 2020 and will enter mass production before the end of the year.
On the Samsung side, the company announced in early 2019 that the first 6nm customer based on EUV technology began tape-out. In addition, Samsung originally planned to launch a 6nm LPE version in 2020. On the basis of 7nm EUV, Samsung’s 6nm uses its Smart Scaling technology to reduce the chip area and power consumption.
At present, only TSMC has achieved mass production of the 5nm process. Therefore, TSMC’s 5nm has become the industry’s favorite after its 7nm, and the production capacity is in short supply.
First of all, the Apple A14 processor and Huawei HiSilicon’s new 5G specification Kirin mobile phone chip are the first two major customers of TSMC’s 5nm process. In addition, Qualcomm’s 5G chip X60 and the new generation of Snapdragon 875 mobile phone chips will also use 5nm. Supply chain sources said that this year, Apple has contracted two-thirds of TSMC’s 5nm production capacity.
In addition, AMD, which is a big hit in the industry, is also fighting for TSMC’s 5nm orders, which are expected to ship next year. As for important customers such as MediaTek, Nvidia, Xilinx, and Bitmain, they are also queuing up behind for TSMC’s 5nm production capacity. In this case, TSMC plans to increase the 5nm monthly production capacity from the original 50,000 pieces to 80,000 pieces.
Samsung is also increasing the development of the 5nm process. According to information previously disclosed by Samsung, the 5nm LPE (5nm Low Power Early) process was originally planned to be put into mass production in the first half of this year, but from the current situation, mass production will wait until 2021. The main customer is Qualcomm.
TSMC’s 3nm is expected to be mass-produced in 2022. At that time, the first major customers will likely be Apple. In addition, it is reported that Intel’s latest GPU may also be produced by TSMC’s 3nm production line.
Samsung’s process that is expected to surpass TSMC may be 3nm, because Samsung is the first official company to announce the use of new GAA transistors. At the 3nm node, GAA surround gate transistors will be used to replace FinFET transistors, while TSMC will still use FinFET technology. Samsung hopes to mass-produce the 3nm process in 2021, and complete the development of the 3nm process in the first half of this year.