Intel 18A returns to process leadership?TSMC: Impossible

Infineon / Mitsubishi / Fuji / Semikron / Eupec / IXYS

Intel 18A returns to process leadership?TSMC: Impossible

Posted Date: 2024-01-24

Electronic Enthusiast Network reports (Text/Zhou Kaiyang) If we want to say that the biggest competition in the semiconductor market in 2024 will definitely be the arms race in AI chips and computing power. But after experiencing a series of architectural innovations, if you want to further improve chip performance, you still have to go back to one of the key factors that determine AI chips, which is the semiconductor manufacturing process.

After Intel announced the launch of IDM 2.0, chip design manufacturers suddenly have more choices. Intel, Samsung and TSMC can all provide them with excellent process solutions. In particular, Intel has been desperately trying to catch up in recent years. It would rather spend a lot of money to build IFS, and even aims to be the first. But at the same time, TSMC has not slowed down. Nodes such as N3P and N3X are advancing steadily, and N2 is also approaching risk trial production. The competition between the two has also entered a fierce stage.

Intel 18A regains process leadership

Intel has been relatively conservative in the field of semiconductor processes in the past, especially during the period when TSMC and Samsung were trying to purchase EUV machines for use. Intel was still busy with external acquisitions to expand its business and stock buybacks, and these activities were of no use to manufacturing advanced semiconductor processes. There is no benefit, and it also causes its manufacturing process and equipment to lag behind its competitors significantly.

After the IDM 2.0 plan was announced, Intel changed its relatively conservative strategy in the past. It can be said that Intel is radical in advancing process nodes. The plan of five nodes in four years does not seem realistic. Even Intel CEO Pat Gelsinger himself emphasized Moore's Law cycle has slowed to three years. However, Intel's current actions are to catch up with its own progress. Under the madness of spending money to build factories and purchase EUV machines, Intel's goal seems to be completely feasible.

Intel CEO Pat Gelsinger said in an interview at the end of last year that Intel's 18A process is on par with TSMC's N2, and even slightly ahead of TSMC's N2 due to its excellent transistor design and power transmission characteristics. Not only that, Intel can provide lower packaging cost solutions, and Intel also plans to once again become the world's top semiconductor process manufacturer with the help of the 18A process.

You must know that Intel uses the most advanced GAA and backside power supply technology on its 20A and 18A process nodes. On the other hand, TSMC's N3 series nodes are just a normal technological evolution of TSMC's N5 process. You must know that today's advanced processes no longer blindly pursue transistor density. Reducing costs, improving yields and heat dissipation performance are also key factors in developing processes.

After advancing the trial production window by half a year, Intel's 18A will enter the trial production stage in the first half of this year. As for mass production, if it goes well, it may be realized by the end of the year, but judging from the process of general wafer fabs , it is expected that 18A may still be postponed to mass production in early 2025.

TSMC N3P conservatively tied with 18A, N2 is still a strong rival

Naturally, TSMC cannot give up its No. 1 spot so easily to Intel, which is both a customer and a competitor. Since TSMC's N3P process and Intel's 18A process are planned to be mass-produced at the same time, they are often compared. According to the transistor density and improvement rate given by Wiki Chip and Anandtech, the transistor density of N3P is approximately 224.2MTr/mm2, which is quite excellent.

At this year's TSMC fourth quarter financial report meeting, Wei Zhejia once again stated that they have further verified all specifications and transistor technology, and can confirm that Intel's 18A process is at the same level as TSMC's N3P process. Wei Zhejia did not clearly state what technical indicators they are at the same level, so we can only assume that the two are at the same level in terms of transistor density or PPA.

Wei Zhejia pointed out that although N3P and 18A will be launched in the second half of next year, TSMC is more advanced in terms of technology maturity. After all, N3P is just an improved version of the N3 process node, and TSMC has been polishing N3 for a long enough time. , and 18A and 20A are Intel's first attempts at GAA and PowervVia back power supply. As for TSMC's N2, he pointed out that even the version that does not use back-side power supply can completely surpass N3P and 18A in terms of PPA.

TSMC Chairman Liu Deyin added after Wei Zhejia’s speech that he believed Wei Zhejia’s description was still too modest, and once again confirmed that TSMC’s N3P is almost the same as Intel’s 18A. At the same time, he also encouraged everyone to look at the competitive relationship between the two from another perspective. Intel's 18A is most likely to be put into use in its own products first, and TSMC has been doing wafer foundry for so many years. Everyone knows that unless there is a shortage of production capacity, the latest process customers can use it, and it will also be based on the customer's needs. products to optimize their foundry processes.

This is precisely Intel's weakness. Their process catch-up and customer development are carried out almost simultaneously, but TSMC already has a leading reputation in the industry and has established alliances with a number of EDA/IP manufacturers. Its design and manufacturing ecosystem are currently Better than Intel.

At the same time, Intel's 18A process still uses 0.33NA EUV machines, and TSMC already has sufficient experience in using machines of this specification. TSMC is definitely more experienced than Intel in terms of experience in improving processes, yields, and energy efficiency. It is said that the yield rate of Sony Japan's CIS wafer fab has never been improved, so it found TSMC as a foreign aid to cooperate in building the factory. TSMC sent 20 engineers from the Kumamoto wafer fab to assist. In just one month, the yield rate increased from 8 The cost increased to about 90%.

Intel’s trump card may not be revealed yet

Although it seems that TSMC still has an absolute advantage, Intel may also reveal its follow-up trump card this year, which is the process roadmap after 18A. Intel has announced that it will hold an IFS Direct Connect event at the end of February, at which time it may disclose future advanced process planning and announce foundry customers for cooperation.

As for Intel's possible subsequent 16A and 14A processes, ASML's 0.55NA EUV machines will inevitably be used. You must know that this signal has not yet entered the large-scale mass production period, and Intel has also purchased most of the first batch of shipped machines. Therefore, after 18A, Intel is bound to use these 0.55NA machines, and the production scale may be higher than that of TSMC in the short term. From this point of view, Intel still has the opportunity to catch up.

However, TSMC and Intel are also always paying attention to the risks brought by the development of advanced processes. Both Intel and Samsung have experienced overturning in the past. And once it overturns, it may take another year to adjust and optimize the process and catch up again. Currently, Intel is still actively developing customers and obviously cannot afford such a failure.

#Intel #18A #returns #process #leadershipTSMC #Impossible