Foundry up, poor, poor and rich

Foundry up, poor, poor and rich

There was a time when semiconductor development cost was 50 million won. Maybe 350 nanometers 0.6-0.5 um (micrometer) Ni 600-500 nanometers. (1 um = 1,000 nanos)

With the development of process technology, it came down to 500, 350, 220, 180, 130, 90, 65, 45, 38, 32, 28, 22, 16, 14, 12, 8, 7, 6, 5, 3, and 2 nanometers.

If the interval between the transistors in the circuit is reduced from 600 nanometers to 2 nanometers, will it be 300 times more integrated? No. If a square measuring 600 mm by 60 cm in the 2D coordinate system is reduced to 2 mm, it is not reduced to 300 times.

600x600/(2x2)
= 360,000/4
= 90,000 times more aggregated.

That is, since the size of the wiring and the transistor (and-not-nand-xor gate component) is reduced to two dimensions, it is reduced to 1/90,000. If the gate size is from 600 nanometers to 5,000 nanometers, it becomes 90,000 x 5,000 gates when configuring 2 nanometers.

In other words, it will be 450,000,000 and 450 million gates in the same area. As the chip size has recently increased, 1 trillion transistors will be integrated in the future.

The problem is that as the chip size increases, the speed becomes faster in proportion to the gate size, so the required IP (intellectual property) and layout costs increase together. Nvidia needed $10 billion and 14 trillion won for development, which could have been designed for 50 million won in the past. IPs with unfamiliar names such as ARM CPU, DDR, HBM, PCIe, USB, SerDes, and HDMI have to be loaded at a huge cost.

Therefore, the cost of developing a 5-nanometer SoC has reached at least 50 billion won. The recent TSMC 3 nanometers will probably cost at least 200 billion won. At least several hundred billion won is required not only for chip development, but also for software, applications, device, driver, communication, LINUX, etc. and verification costs.

Hundreds of billions of dollars is not a small amount of money even for large companies, so once development is done wrong, the company will collapse. Exynos heat problems cause the mobile phone business to collapse, battery explosions, and such things happen frequently. If development needs to be done at once, but the development fails, the relevant executives and team leaders must take responsibility. Hundreds of billions of dollars have been lost in set-ups, shouldn't someone be responsible?

That's why no one wants to be innovative in semiconductor development. Should we put in this function and make it well? If you bring an idea at the agency level... What if executives and team leaders don't do that? Who will take responsibility if there's a problem in that area? Is there a record that was mass-produced by other companies? Is it verified? Does it work well and sell well? I'm asking this. No, bring me the idea of making something that doesn't exist in the world. How can there be such a record?

For example, put USB 4.0 in this chip and speed it up to 40Gbps. As soon as you plug it in, the data is delivered in 0.1 seconds, so consumers are satisfied. Even if you make such a proposal, who mass-produced it in the lab? Have you tried it? Has it been verified? If it doesn't work, you'll take responsibility.

Who can take responsibility for that and be confident? I'm nervous about making ideas. That's why innovation disappears. At times like this, I'll take responsibility if the chairman or the CEO is not available. Let's do our best. There are risks, but we can't do it. We have to try. We have to say this. If a president points to the window of a building and asks, "Are you going to jump out of this business?" You are not qualified as a president.

Why is the executive responsible for that? Basically, it's not what the chairman of a tech company should say, but such a mind itself is not qualified. Engineers can only take the risk to the extent that they do their best and do not repeat the same mistakes. Let's do the rest for the boss. You can do it. Don't take responsibility... You have to say this to make things work. The reason why Oman Electronics can't do it is because it doesn't try, so there is no successful log. In this case, you pass on the successful log to the customer.

We haven't tried this before, but we can't take responsibility. Waverlist is that if the development fails, the customer is responsible. It's something we haven't done at our foundry, but if you want to try it, you have to pay an additional fee because it's your first time trying it. This is the mindset.

What this means is there are two hospitals that are as big as Goman. I need to have heart surgery, but one hospital is that surgery gum. There was one yesterday, and there are hundreds of patients in our hospital who underwent that surgery. If you wake up with your eyes closed for a while, you will be hospitalized for a few days and discharged. The cost of the surgery is one million won. Let's have surgery. There is a doctor who says this

In another hospital, we haven't performed the surgery on the same patient, but a paper says it works. But the surgery costs 100 million won because we need to upgrade our internship equipment. The patient pays for it, and we're going to perform the surgery, but I don't know the outcome. Let's open our chest for heart surgery. Let's say there's a doctor who does it.

Which hospital would you go to? That's why Oman Electronics is losing ground to TSMC. There is a huge difference between 10% and 60% of the market share. Semiconductor development costs are hundreds of billions of dollars, so would you like to go into Oman Electronics and become a Mormot? Would you decide that once you make a mistake, hundreds of billions of dollars will fly away and the company will close?

TSMC said, "We've used that IP dozens of times, successful customers ordered it hundreds of millions of times, and we're currently testing an upgraded version. You don't have to worry. Oman Electronics has not tried it and failed, and the IP price is more expensive because it's a small purchase, and labor costs are added to it, so pay extra for the test. Which foundry would you use if you said this?

In this situation, since it is a domestic fab, will they ask us to use it from a long-term perspective with a patriotic heart? Isn't it a wonder that hundreds of billions of dollars are not small money, and that fabless companies that have no choice but to go abroad survive in Korea in such an environment? The selection of a fabless company's foundry is the same as the risk of entrusting heart surgery. Which doctor would you like to leave your heart to?

Those who have enough will take it, but those who do not have enough will take it