Why legacy semiconductors are time-consuming

Why legacy semiconductors are time-consuming

1.
From the second half of the year, I've heard a lot that it's time for legacy semiconductors. Samsung Electronics and Hynix produce about 1.4 million DRAMs, but since the number of DRAMs used for HBM exceeds 300,000 based on wafers, shortage will naturally occur. That's the logic. No matter how hard it is to sell smartphones due to the bad economy in China, the news that Chinese smartphone makers cannot make smartphones next year is coming out no matter how bad the Foxconn conference said last week.

However, it is not the first time that analysts have made such a call. Back in 2018, they also called it "Shortage," referring to demand for DRAM from the cloud. And it was wrong then. The reason why it was wrong at that time was because it overlooked the speed of miniaturization of DRAM. As the DRAM came in at 20~30nm and 10nm, the number of DRAMs per wafer (net die) increased by more than 50%, which is why supply increased faster than demand.

Currently, even if 1a process is introduced in DRAM, net die increases by 15%, which is a definite analysis considering the impact. After all, the HBM market is becoming the size of DRAM market in the 10s, so why consider other things.

2.
Moore's Law was arguably the most significant social phenomenon over the past 30 years. Companies like Microsoft and Apple have changed the world in both the Internet and mobile devices, but the reason why all this is possible is that semiconductor performance has doubled exponentially per year.

In other words, because mobile phones are about millions of times faster than 20 years ago, and both semiconductor performance and total cost have been improved exponentially, it is possible for us to deliver on mobile phones.

On the other hand, among all goods on the planet, household PCs are the ones with the least inflationary impact. All goods have been affected by inflation, but humanity has overcome inflation, at least in computing power, through cost-cutting and performance growth that exceeds that.

The simplest explanation for the change in the hegemony of tech companies over the past decade can be explained by the slowness of Moore's Law. There is a saying that the reason Intel fell is because he is a CFO-turned-CEO, but the most important thing is that Moore's Law became physically impossible, and the performance improvement strategy of multi-core did not work, and the quality advantage maintained by Moore's Law was lost.

So, on the other side, the unfamiliar BM of foundry emerged as the answer. In the process of using Moore's Law, which is becoming slow, and the increasingly complex EUV, the value chain differentiated, and the unique value chain of fabless and tsmc is accepted as the answer.

In conclusion, exponentially improved performance and total cost (Moore's law) are the most important in understanding semiconductor companies, and the ability to lead the industry through them is the second most important. And in general, it led the industry by maintaining a 'super gap' with preemptive CAPEX and R&D.

Additionally, from the perspective of semiconductor producers, no matter how much money they make, this 'super gap' would have been considered a hell of a BM for CFOs as additional CAPEX input is required. On the other hand, from the perspective of consumers on the other side, it was not only an incentive to change cell phones every two years, but also because of such an exponential increase in performance, the aspect of life changed completely.

While performance increased exponentially and S/W companies or fabless companies rose in the middle, H/W companies did not grow that much. In the value chain from H/W to consumers, it is a representative indicator of who did the value capture.

Incidentally, in fact, the best winner is the consumer. Consumers don't pay as much as they feel convenient. They pay only for the monopoly power of service providers, and the rest remains their utility. It seems that there is also a talk about improving GDP based on value or utility in this dimension. If statistics are reorganized around utility, the quality of life improvement exponentially exceeds GDP. I think the intention to pay for a 1 million won mobile phone and a 100,000 won monthly communication fee is over 10 million won. In other words, we have gained considerable utility compared to the money we pay, and the size is increasing.

After all, capitalism seems to have a large captured value, which is the consumer's share. And I think this social system is more of a blessing. Thank God.

3.
Anyway, when I come back, legacy time is uncomfortable. The core competency of semiconductor manufacturers, including memory, was the direction of R&D that followed Moore's Law and widened the gap. However, people talking about legacy now say that profits for Samsung Electronics and Hynix will skyrocket because Moore's Law has ended (because net die growth cannot keep up with demand growth).

The fact that Moore's Law ended rather than simply a 10-fold increase in Korea-U.S. semiconductors means that the direction of existing social development, including semiconductors, has completely changed. Therefore, it feels a little uncomfortable for those who shout memory buy while raising their EPS forecast according to the existing formula.

In fact, I think the industry is being reorganized in a way that loses its long-term differentiation point. If running speed used to be their advantage, now they argue that they will make a lot of money because they have been running far. I don't think these advantages are sustainable. In the end, going far is bound to win, and slowing down is not advantageous in the long run.

It feels like their strengths are shifting from technological harmonic to the N100 trillion CAPEX that has been in place. So, I think it will be a natural threat if we take a policy drive and invest that much. China and the United States are the two countries that have capital and incentives to drive. And in that way, China distorted and ruined the display, chemical, and secondary battery markets.

A super-difference, whether cost or performance, must be maintained in order for semiconductors to become hegemonic, but the period of physical maintenance seems to have passed. And I believe that these characteristics will bring long-term opportunities to China, the United States, or Vietnam.

HBM is