Let technology serve all mankind
Shenzhen Translation Bureau is its compilation team, focusing on the fields of technology, business, workplace, life, etc., focusing on introducing new technologies, new ideas, and new trends abroad.
Editor ’s note: Recently, TSMC, the most advanced foundry in the chip industry, has made a major decision to set up a 5nm fab in the United States. How should this matter be interpreted? If we look at the opinions of people in the US industry, it may provide some useful reference. Ben Thompson, a well-known science and technology blogger, summed up three major lessons from the perspective of the United States. Please review them critically. The original title is: Chips and Geopolitics
The debate about who should engrave the Mount Rushmore will continue for a long time. But what is certain is that Zhang Zhongmou should be on the list. He is undoubtedly the leader in terms of visibility-related influence.
integration and modularization
Clayton Christensen’s “Innovator’s Answer” in 2003 explained how the industry’s natural development process has evolved from an interdependent architecture to a modular architecture :
Customer buying behavior depends on whether your product can help them solve major problems. However, the composition of a “solution” is divided into two different scenarios, as shown in Figure 5-1: The current products on the market are “not good enough” or “too good”. We found that for products that are “not good enough” on the market, internal development is more advantageous, and outsourcing (or professional, independent production methods) is applicable to those products that are “too good”.
Figure 5-1
On the left side of Figure 5-1, we can see that when a product has a performance gap (that is, when the product’s function and reliability cannot meet a certain market-level user demand), the enterprise The best products must be invested to compete. In this process, companies with patented and symbiotic product architectures can often gain a competitive advantage, because the default standards for modular product architectures greatly deprive engineers of design freedom and reduce product performance optimizability.
This has an intuitive meaning: optimizing everything brings better performance, but at the expense of long-term reliability and flexibility. Of course, long-term reliability and flexibility are not bad, but the priority is lower. However, once that highest priority is satisfied, these secondary priorities become the most important issues.
“Excessive service” does not mean that customers are unwilling to spend money on product improvement, but that customers are willing to spend money to buy products that are no longer improved on the original type. Once the product meets their requirements for certain functions and reliability, they will begin to redefine which aspects are “not perfect”. The new “not perfect” is the characteristics that can not immediately meet the requirements of customers, so customers are willing to pay extra fees for the speed, convenience and personalization of products on the new innovation trajectory. When this happens, we can say that the basis of market competition at this level has changed.
This is a big problem for companies that are in a dominant market in this transition period. After all, companies dominate because they perform best, that is to say, they are highly integrated. For business models and deeper cultural reasons, it is often difficult to break this integration. And this transition opens the door for new entrants:
Competitive pressure on the new improved trajectory has gradually evolved the product architecture. As shown in Figure 5-1, the product is gradually moving away from the symbiotic patented architecture With competitive advantage), began to enter the era of modular design, ushering in a competitive market with excellent performance. Modular architecture can help companies win low-end markets in disruptive competition. They can independently upgrade a certain subsystem without the need for a complete redesign, thus accelerating the speed of new products to market. The cost of using standardized component interfaces is at the expense of system performance. Nevertheless, the product can be easily sold to customers because the product’s functionality is now “too good”.
Modular studentsProduction has had a profound impact on the industry structure. It has spawned a series of independent, decentralized organizational departments to complete the tasks of sales, procurement, and assembly of subsystems and components. In the world of symbiotic products, enterprises must personally produce all important components, while in the world of modular products, it is only necessary to outsource or contract the production tasks of a certain component to invigorate an enterprise. Ultimately, the specifications of the modular interface will eventually form an industry standard. When an industry reaches this step, then companies in the industry can mix and match the components produced by the best suppliers to easily meet individual customer needs. .
Chang Zhang Zhongmou ’s TSMC, established in 1987, is arguably the best example of the process described by Christensen.
Intel and TSMC
The microprocessor was invented by Intel in 1971, and it was not good enough for the following decades. The 4-bit Intel 4004 is followed by the 8-bit Intel 8008, then the Intel 8080. In 1978, the Intel 8086 appeared, which is a 16-bit processor that is backward compatible with programs written in 8080 and 8008. Then came Intel 80286, and by 1985, the 32-bit Intel 80386 appeared. 80386 defined the benchmark x86 instruction set, which laid the foundation for modern processors used in most laptops, desktops, and servers. But the root of x86 is still 8008. By integrating the design, manufacturing, and software of the 1970s, Intel defined and controlled the processor market, which has been for decades.
This integrated approach takes a long time to “over-serve” the market. Intel’s 80386 was first replaced by 80486 and the latter was replaced by Pentium. Each version further accelerated the speed of the computer, so that use cases that could not be imagined 1 or 2 years ago suddenly seemed within reach. It is commendable that under the pressure of AMD’s 64-bit version and backward compatibility with 80386, Intel still did this.
Despite this, Intel also manufactures general purpose processors. At least in theory, processors made for specific tasks are much faster, but the start of manufacturing special processors is difficult: Zhang Zhongmou, a senior executive at Texas Instruments, was established in the 1980s The cost of a new chip company needs between 50 million and 100 million US dollars, mainly because of manufacturing costs. Yes, you can sign a production contract with Intel, Texas Instruments or Motorola, but this may not be reliable-because they are also your competitors!
A few years later, in 1987, Zhang Zhongmou was invited to return to Taiwan and asked him to help develop a business to establish a semiconductor industryplan. Zhang Zhongmou explained in an interview with the Computer History Museum that he has no foundation at hand:
Before formulating, I want to see how many families we have. My conclusion is that there are very few things. We do not have R & D capabilities, and even if we do, they are very small. We do not have the advantages of circuit design and IC product design. We have little advantage in sales and marketing, and almost no advantage in intellectual property. The only possible advantage of Taiwan is even a potential advantage, not an obvious one. It is semiconductor manufacturing and wafer manufacturing. So, what kind of company will you build to adapt to this advantage and avoid all the other disadvantages? The answer is pure wafer foundry …
When choosing a pure OEM model, I managed to take advantage of Taiwan ’s only advantages and managed to avoid many other disadvantages. However, there is a problem with the pure wafer foundry model, which may be a fatal problem, that is, “Where is the market?”
What happened later is what Christensen will describe a few years later: TSMC created the market by “promoting independent, non-integrated organizations to sell, purchase, and assemble components and subsystems.” To be more specific, Zhang Zhongmou asked chip designers to start their own companies:
When I was working for TI and General Instruments, I saw that many IC design engineers wanted to leave to start their own businesses, but the only biggest obstacle to leaving these companies was that they could n’t raise as many Money to form their own company. Because at the time, everyone thought that every company needed to manufacture, all need wafer manufacturing, these are the largest part of the capital investment of semiconductor companies and IC companies. I have witnessed all those who had the idea of leaving but ended up halfway because they lacked the ability to raise a lot of money to build a fab. So, I think maybe TSMC (as a single-fab wafer foundry) can correct this. Because we can correct this, these designers can successfully build their own companies and then become our customers, which will build a stable and growing market for us.
This approach worked. The graphics processor is an early example: Nvidia was founded in 1993 when it only cost $ 20 million. The company never owned its own fab. Qualcomm lost millions of dollars in the production of its own earliest designs. In 2001, it stripped off the chip manufacturing department to focus on design. Ten years later, Apple began to manufacture its own chips without a fab. . Today, tens of thousands of chip designers are in various fields, from electrical equipment to jet fighters. They make special chips for various niche products, and they do not have their own fabs.
Also issued along the wayOne thing happened: As I detailed in “Intel and the Danger of Integration” in 2018, TSMC eventually surpassed Intel not only in terms of flexibility but also in terms of pure performance:
However, with the passage of time, TSMC is getting better and better, in large part because it has no choice: soon, its manufacturing capacity is only one step away from industry standards, and in It caught up within ten years (although Intel has always been ahead of everyone). At the same time, the existence of TSMC has created conditions for the explosive growth of design-free “fabless” chip companies … The growth of the business has made TSMC invest more in its own manufacturing capabilities.
In short, no matter what level of performance you care about, TSMC is the best chip manufacturer in the world. What followed was a completely new problem, not only for TSMC, but also for Taiwan.
International Relations Issues
Sino-US relations are complicated.
You will notice that although most of the foundry Taiwan is for self-use, the home country of Samsung, which produces the highest-end chips, is located at the end of the Pacific Ocean. At the same time, the United States is on the other side of the Pacific Ocean. Oregon, New Mexico, and Arizona have advanced foundries, but they are all operated by Intel, and Intel only produces chips for its own integrated use cases.
This is important because the chip is also very important in many use cases other than PCs and servers (which is the focus of Intel), that is, TSMC Very important. Nowadays, whether it is military or other uses, almost every device is equipped with a processor. Some of them do not require particularly high performance and can be manufactured by fabs built in the United States and around the world a few years ago. However, others require the most advanced technology, which means that they must be manufactured by TSMC in Taiwan.
Announcement from TSMC
This is the background that TSMC announced last week to establish a fab in the United States. From The Wall Street Journal:
TSMC, the world ’s largest contract manufacturer of silicon chips, said on Friday that it will spend $ 12 billion to build a chip factory in Arizona because of increased U.S. concerns about key technology dependence on Asia. TSMC said the project disclosed by the Wall Street Journal earlier on Thursday was supported by the federal government and Arizona. This is just when everyone is more and more worried that the United States will rely heavily on Taiwan, China and South Korea to produce microelectronic products and other customs.Key technology, when the Trump administration tried to start building a new chip factory in the United States.
Informed sources said that TSMC decided to embark on the project at a board meeting held on Tuesday, adding that both Arizona and the US Department of Commerce participated in the plan. The company said in a statement that the plant will begin production next year, with a target production time of 2024. TSMC’s new plant will produce chips with 5-nanometer transistors, which are the smallest, fastest, and most energy-efficient chips that can be manufactured today. TSMC has also only launched 5nm chips in a factory in Taiwan in recent months. TSMC said that the integrated factory will produce 20,000 wafers per month. For a company that produced more than 12 million wafers last year alone, the scale of this factory is relatively small. TSMC’s Fab 18 plant, which is currently producing its 5-nanometer chips in Taiwan, has a monthly target production capacity of 100,000 wafers when it breaks ground in 2018.
First of all, although this announcement is similar to Foxconn ’s unlucky factory in Wisconsin, this project was suspected of political performance from the beginning, and more importantly, for anyone That said, it doesn’t make much sense. Frankly speaking, the current situation-a vacant innovation center and a factory whose intention was changed before completion-is the default result.
There are several reasons why TSMC ’s project is different. First of all, the construction of the fab will not be abandoned halfway; TSMC will either invest billions of dollars or not at all. Second, the US federal government obviously bears a large portion of this cost. Third, this is exactly what the US federal government should do, because its impact on national security is real.
This does raise a question, that is, how much TSMC has invested in this project. As the “Wall Street Journal” puts it, relatively speaking, Arizona’s fabs are very small. Although 5 nanometer chips are today’s top products, they will not appear when the fab opens in 2024. In addition, it is worth noting that TSMC set up a fab in Washington State in 1998. The plant is still operating, but so far, TSMC has not made any other investments in the United States.
However, I think, at least from the American perspective, this is an overly pessimistic interpretation of this news. First of all, TSMC certainly has to start on a small scale, and it will use the construction technology it has already come up with to build factories. It is one thing to build a “gigafab” (large-scale fab) next to a factory that has been built in Taiwan, even if the best employees who have pushed TSMC to the top in the past three decades are still contemplating the next processing node But making similar attempts across the ocean is another matter.
However, what is more significant is that in terms of processor technology, the Taiwan region in 2020 is not the last but the vanguard. Authorities with significant stakes have been working to keep the best technology locally. But the move to set up factories in the United States has already taken place.
Lessons and lessons to be learned from the technical circle
In this news, the technology circle and the United States in general refer to three lessons.
First of all, although we learned that technology is inseparable from domestic politics in 2016, the lesson to us in 2020 is that technology and geopolitics should be inseparable. It is silicon that has made Silicon Valley famous, and everything related to chip decisions is related to geopolitics, not economic factors.
Second, at some point, every technology company may have to choose a side station. Maybe in the end everyone will be in trouble like TSMC, and Switzerland will not be an option.
Third, much like Compaq, Intel is also a fable that the United States seems to have lost. Out of the endless pursuit of efficiency and shareholder value, the United States gave up flexibility and flexibility and turned to the top performance. Intel is one of the most advanced chip manufacturers in the world, but it turns out that its functions are too constrained by its own needs to be universally applied. To make matters worse, to the extent that Intel is willing to become a contract manufacturer, it hopes that the federal government will pay the bill so that it can better meet the requirements of shareholders. In my opinion, the US government has correctly selected an operator that is actually used to operating around the world, rather than an operator who was used to doing so.
At the same time, TSMC ’s reasonable prudence in setting up a fab in the United States actually provided Intel with an opportunity. As early as 2013, in the first article of Stratechery, I urged the company to embrace manufacturing and give up its integration. If Intel took action at the time, Intel and the entire United States would be in a much better position. But as the saying goes, the second best opportunity is now-this not only applies to Intel, Intel should spend the money to develop contract manufacturing by itself, but also applies to the United States. The world has changed, and it is time to act accordingly.
Translator: boxi.