Reasons why Japanese companies are still in time to enter the foundry industry

　In Japan, which is originally skilled in manufacturing technology, there is still no foundry (a service provider that undertakes semiconductor manufacturing), and 30 years have passed without a growth engine. Intel has decided to acquire Israeli foundry Tower Semiconductor for $5.4 billion.

　Tower's sales in 2021 are only $1.5 billion. Nevertheless, Intel buys it for 3.5 times that amount. Why doesn't Japan start a foundry business?

More than 10 years ago, I wrote in October 2010 that we should start a foundry business in Japan as soon as possible (reference material 1). On the other hand, there was also a voice saying "Too late" (too late). But what about the reality?

50% of sales at TSMC are 16nm and above

　In the foundry business, Taiwan's TSMC is far ahead of the curve, and no one has moved to establish a foundry in Japan until now. TSMC is certainly the leader in 5nm and 7nm, but it does not only produce products for such cutting-edge process nodes. As seen in Figure 1, 50% of product sales are on technologies looser than the 16nm process node.

　Also, the foundry UMC, which was overwhelmingly farther away from TSMC in Taiwan, has only the technology to make it with the process node of 22 nm or more, which is the finest rule. Still, sales in 2021 are up 20.5% year-on-year to $7.698 billion (approximately ¥885.3 billion). In other words, it's a lucrative business even if it's not as advanced a process node as TSMC. By the way, UMC's operating profit is 1.868 billion dollars (about 214.8 billion yen) with a profit margin of 24.3%.

Intel buys Tower for automotive applications

　The biggest reason Intel decided to buy the Tower is because the foundries are profitable. Moreover, he himself declared that he would focus on the foundry with IDM2.0. Intel once started a foundry business. In 2012-13, FPGA maker Altera was using Intel as a foundry. But since acquiring Altera, Intel has shown little interest in foundries. Since Pat Gelsinger took over as CEO in 2021, Intel has once again committed to its foundry business.

However, Intel does not fully understand the foundry business. Tower has been a veteran of the foundry business for more than 10 years, even though sales have not grown for the past year or two. We are well versed in the design technology that is indispensable in the foundry business, and have provided PDK (Process Development Kit) to support our customers. If Intel learns the foundry business with Tower as a teacher, it has a good chance of succeeding.

There are production lines called foundries

　Unfortunately, the foundry business is not understood in Japan. There is a section that simply thinks that if there is a production line, customers will come. Unfortunately, a business that only waits is not viable as a business. The more actively you try to get customers, the more you need knowledge of semiconductor design.

　A typical customer can't put their image into the design data even if they want a chip like this. In semiconductor design, logic description data called RTL is necessary to convert it into design data, so it is necessary to program in LSI design languages ​​such as Verilog and HDL. After verifying that there are no bugs or mistakes in the RTL, it is converted to a schematic using logic synthesis. The circuit diagram is converted into the layout arrangement of small-scale circuits such as IP blocks, wiring layers between IP cores, etc., and finally the mask format GDS-II is output. A photomask is created based on this output data.

　At any design stage, the foundry must be able to respond to customer requirements. For this reason, the presence of a design house that carries out design work is essential. TSMC had a division that acted as a design house. This division spun off to become Global UniChip, but TSMC is still the largest shareholder with a 35% stake (reference material 2).

Get the Tower to get into the car market

Another reason for Intel's acquisition of Tower is a full-scale entry into the automotive industry. Intel has acquired Mobileye, which handles image data from cameras used for autonomous driving, but this alone is lonely. Therefore, Intel Investor Day 2022 held on February 17 introduced three technologies to strengthen the automotive market (reference material 3).

The first is the domain computer architecture called the Open Central Compute Architecture. Since the number of ECUs (Electronic Control Units) in automobiles has increased too much, a domain computer that integrates several ECUs into one is indispensable from now on. Intel is aiming for a domain computer that takes charge of the IT related part of automatic driving. Another is an automotive-grade foundry manufacturing network. Expect collaborations with Tower and Mobileye here. Finally, there is the packaging technology that enables the use of advanced technology. Implement by making full use of CPU and accelerator chiplets and IP cores.

As far as Intel's new entry is concerned, it cannot be said that it is too late to enter the foundry business. In the United States, the foundry Skywater Technology Inc., which has barely mastered the 90nm process, has made a bid for semiconductor manufacturing, and GlobalFoundries, which was driven out by the Trump administration, is making a comeback. Regardless of the process node, the base of semiconductors is expanding, and it is now possible to do business in various fields.

　Semiconductors have now become brains from circuits, and have basically become compute + alpha. Without semiconductors, DX (Digital Transformation) and Metaverse cannot be realized, and smart xx such as smart cities and smart factories cannot be realized. As you look at the progress of IT in the future, I would like you to think about how the system side can advance more favorably than other companies without semiconductors. Of course, software can be differentiated. However, software alone is slower. Semiconductors are essential if you want to dominate system performance.

　If system companies around the world want to differentiate themselves with semiconductors, but there are only a limited number of places that can manufacture them, like GAFAM, Japan should seize this business opportunity. I strongly hope that foundry companies will emerge as soon as possible.

References

1. Kenji Tsuda, "Japan should establish a foundry as soon as possible", Semicom Portal, (2010/10/29)

2.Overview on Global UniChip Corp.,

3. Intel Investor Meeting 2022