If Samsung announced in mid-2019 that it will launch its "wrap-around-gate (GAA)" technology in 2021 to replace FinFET transistor technology, FinFET can still be calm; to this day, Intel has stated that its 5nm process will abandon FinFET and switch to GAA , There are already signs of turning the age. The three major foundry giants have already chosen GAA. Although TSMC ’s circuit line as the leader of the foundry is “not moving”, there seems to be no suspense. Is FinFET really at the end of history?
FinFET's glory
After all, when FinFET debuted as a "savior", it carried the important "mission" of Moore's Law to continue to advance.
With the upgrading of process technology, the manufacture of transistors becomes more difficult. The first integrated circuit flip-flop in 1958 was built with only two transistors, and today the chip already contains more than 1 billion transistors. This motive force comes from the continuous advancement of the flat silicon manufacturing process under the command of Moore's Law.
When the gate length approaches the 20nm mark, the ability to control the current drops sharply, and the leakage rate increases accordingly. The traditional planar MOSFET structure appears to be at the "end". Prof. Zhengming Hu from the industry has proposed two solutions: one is FinFET transistor with three-dimensional structure, and the other is FD-SOI transistor technology based on SOI ultra-thin silicon-on-insulator technology.
FinFET and FD-SOI allowed Moore's Law to continue the legend, but the two have taken different paths afterwards. The FinFET process tops the list first. Intel first introduced the commercial FinFET process technology in 2011, which significantly improved performance and reduced power consumption. TSMC also achieved great success with FinFET technology. Subsequently, FinFET has become a global mainstream. The "Fuji" choice of Yuanchang.
In contrast, the FD-SOI process seems to have been living in the shadow of FinFETs. Although its process leakage rate is low and its power consumption has advantages, the manufactured chips have applications in the Internet of Things, automotive, network infrastructure, consumer and other fields, plus the power of giants such as Samsung, GF, IBM, ST, etc. Pushing has opened up a world in the market. However, industry veterans pointed out that due to its high substrate cost, it is difficult to make the size smaller as it moves upward, and the highest level is up to 12nm, which is difficult to continue in the future.
Although FinFET has taken the lead in the "two-choice one" competition, with the application of the Internet of Things, artificial intelligence, and intelligent driving, it has brought new challenges to ICs, especially the manufacturing and R & D costs of FinFETs are getting higher and higher. 5nm can still make great progress, but the flow of process history seems destined to "turn" again.
Why GAA?
With Samsung taking the lead, and following up with Intel, GAA has suddenly become the upstart to take over FinFET.
The difference from FinFET is that there are gates around the four sides of the GAA design channel, which reduces the leakage voltage and improves the control of the channel. This is a basic step when reducing the process nodes. By using more efficient transistor designs, coupled with smaller nodes, better energy consumption can be achieved.
Seniors also mentioned that the kinetic energy of process nodes is to improve performance and reduce power consumption. When the process node is advanced to 3nm, the FinFET economy is no longer feasible and will turn to GAA.
Samsung is optimistic that GAA technology can improve performance by 35%, reduce power consumption by 50%, and chip area by 45% compared to the 7nm process. It is reported that the first batch of 3nm Samsung smartphone chips equipped with this technology will begin mass production in 2021, and more demanding chips such as graphics processors and data center AI chips will be mass-produced in 2022.
It is worth noting that GAA technology also has several different routes, and future details need to be further verified. Moreover, the shift to GAA undoubtedly involves a change in architecture. Industry insiders point out that this puts forward different requirements for equipment. It is reported that some equipment manufacturers are already developing special etching and thin-film equipment.
Xinhua Mountain on the sword?
In the FinFET market, TSMC stands out, and Samsung and Intel are struggling to catch up. Now it seems that the GAA is already on the string. The question is, what will happen to the stalemate of the "three kingdoms"?
From Samsung's context, Samsung believes that the GAA technology bets are one or two years ahead of its rivals, and it will lay down and maintain its first-mover advantage in this field.
But Intel is also ambitious, aiming to regain leadership in GAA. Intel announced that it will launch 7nm process technology in 2021 and will develop 5nm based on the 7nm process. It is estimated that the industry will see its 5nm process "true capacity" as soon as 2023.
Although Samsung is the leader in GAA technology, considering Intel ’s strength in process technology, its GAA process performance has improved or become more obvious, and Intel has to introspect itself and no longer follow the "Long March" road of 10nm process.
In the past, TSMC was extremely low-key and cautious. Although TSMC announced that the 5nm process for mass production in 2020 still uses the FinFET process, it is expected that its 3nm process will be advanced to mass production in 2023 or 2022. Process. According to TSMC officials, details of its 3nm will be announced at the North American Technology Forum on April 29. By then, what kind of tricks will TSMC offer?
The battle of the GAA has already started.