TSMC is too scary. 5nm trial production is completed and will enter mass production

While we are still struggling to decide whether the newly purchased mobile phone or computer is equipped with the latest 7nm or 10nm chip, TSMC announced this month that it has completed the 5nm architecture design and has entered the trial production stage. The semiconductor process has achieved such a rapid iteration in such a short period of time, which is beyond the expectations of many people.

Especially for those DIY players who have been waiting for AMD's 7nm and Intel 10nm new products, this news is like a bolt from the blue. Does it mean that 7nm and 10nm, which many people have not used yet, are about to become outdated again? This not only makes people sigh at the rapid development of technology, but also makes people feel a little unhappy...

TSMC's 5nm is ready to go

In fact, from the user's perspective, we should support the accelerated iteration and upgrade of products, because this allows us to get products with better performance for the same money. However, if the frequency of product updates is too fast, it seems to have the opposite effect of "too much is as bad as too little". Because no one wants the latest product they just bought today to become "outdated" within a few days of use!

TSMC has benefited greatly from being the first to seize the 7nm process, and has received large orders from international giants such as Apple, Qualcomm, AMD, and NVIDIA. Even AMD, which has been suppressed by Intel, has also received more attention and expectations from players thanks to TSMC's 7nm.

This also prompted TSMC to make a concerted effort and decide to conquer the 5nm high ground in 2020!

According to official data, compared with 7nm (first generation DUV), the new 5nm chip based on the Cortex A72 core can provide 1.8 times the logic density, 15% performance improvement, and 15% chip area reduction. According to TSMC's development plan, 3nm and 2nm are already on the list in the future...

Is it possible that the nm (nanometer) unit that we have used for many years is about to become obsolete?

In fact, it is not absolute. According to TSMC's official disclosure: the layout of the 5nm process technology market is to assist customers in realizing the 5nm system-on-chip design that supports the next generation of high-performance computing application products, targeting the 5G and artificial intelligence (AI) markets with broad development prospects.

This reminds us of some things that happened last year regarding the first launch of 7nm chips. Who would have thought that the first company in the world to mass-produce 7nm chips was actually a mining machine company from China, Canaan Creative. The company's technical level and shipment volume are naturally not comparable to Apple and Qualcomm, but it was able to launch the first chip before them.

The reason is that the structure of the mining chip is extremely simple, so it is very easy to adapt to the latest production process from the perspective of design and production. For general computing processors such as computers and mobile phones, because the structure is much more complex and the transistor density is extremely high, there are more stringent requirements for design, production, and the impact on yield.

If a mining chip is like a basketball court, then a computer processor is like building an aircraft carrier...

TSMC's use of 5nm process technology as a layout for the future 5G and AI markets is similar in nature.

For example, the most important baseband chip in the 5G field, the 5nm baseband with smaller size, lower heat generation and stronger performance will undoubtedly bring a direct improvement to the experience of mobile terminals. In the field of artificial intelligence, the design of edge computing chips is somewhat similar to that of mining machine chips, with a simple structure and single function, and it is easier to be deployed in more smart terminals.

Therefore, judging from the current trend, if TSMC's 5nm is mass-produced in 2020, it will not have much impact on the traditional PC market and the mobile phone processor market in the short term. The 7nm and 10nm processes that are currently in the ascendant will continue to be used for many years. At least the R&D costs will limit these processor manufacturers from updating to the new process faster.

This seems to indicate that the traditional chip field and the emerging 5G+AI field will form a clear watershed in the 7nm era. Previously, the global semiconductor process has always been constrained by the process step, but in the future, dedicated chips with relatively simple structure, smaller size and more advanced process will open up a new path.

Therefore, empiricism that simply distinguishes whether something is more advanced based on the size of the numbers will no longer be applicable.

Even now, it is not advisable to simply conclude that "the smaller the number, the more advanced it is". According to foreign media's evaluation of Intel's first 10nm processor Core i3-8121U, which has achieved small-scale shipments, although "10" seems to be a little behind "7", in fact, Intel's 10nm technology is more advanced.

To put it simply, a processor is about cramming more transistors into a limited chip area and ensuring that these components can operate efficiently and stably. Intel's 10nm process uses the third-generation FinFET stereo transistor technology, with a transistor density of 100.8 million per square millimeter (in line with official claims), which is 2.7 times that of the current 14nm!

In comparison, Samsung's 10nm process has a transistor density of only 55.1 million per square millimeter, which is only a little more than half of Intel's. The 7nm process has a transistor density of 101.23 million per square millimeter, which is barely higher than Intel's 10nm. As for TSMC, its transistor density is even lower than Samsung's. This is also the fundamental reason why many industry insiders complain that "TSMC's 7nm process has shrunk."

Therefore, if we only look at the key data of transistor density, although Intel's 10nm technology seems to be "backward", it is actually not backward compared to other 7nm technologies, and is even ahead.

Looking back at this issue from the perspective of our users, do we really need to blindly pursue novelty and excellence?

The answer must be no. We must clearly understand that technology is developing endlessly, but our needs have a threshold, so how should we understand this sentence?

To give a simple example, when the processor was still in the era of relatively backward technology, the power consumption and heat generation were so huge that it directly gave rise to the popularity of professional radiators, such as the well-known Kyushu Fengshen and Overclocking Three, and high-end ones such as Noctua, etc. Because without these huge radiators, it really couldn't be suppressed!

At that time, every improvement in the manufacturing process would actually bring a "solid" improvement, because the basic power consumption of the processor at that time was really too high. With the continuous iteration of the manufacturing process, the base of the improvement became smaller. Constrained by physical limits, the percentage of improvement was also shrinking.

As a result, people have gradually become less sensitive to improvements in manufacturing technology.

If you still don't understand, let me give you a simpler example. A few years ago, Android phones were hot to the touch and the battery drained quickly, but now even a thousand-yuan phone won't give you such a miserable experience. So before, you were eager to change your phone, but now it's basically enough and there's nothing special about it, so your need to change your phone is not so urgent.

Don’t worry too much about 5nm, 7nm, or 10nm.

If you pursue the latest, you will never be satisfied. Only choose the suitable ones, and you will be at peace.