In a 160-minute interview with CommonWealth Magazine, Taiwan Semiconductor Manufacturing Co. (TSMC) Chairman Morris Chang may have spoken slowly but directly answered every question with clear logic. He said Moore’s Law, a main tenet of the semiconductor industry which projects computing power to double every two years, may come to an end in 2025, and also analyzed the limitations faced by TSMC in investing in new businesses and what the future may hold in this area.

He also touched on Taiwan’s recent run of labor-management disputes and had some interesting views on labor unions. The following are translated excerpts of the interview:    

CommonWealth Magazine: The high-tech sector is now busy looking for the next big opportunity after smartphones. Aside from the Internet of Things, which you’ve talked about, other possibilities that have been suggested are augmented reality (AR), virtual reality (VR) and artificial intelligence (AI). How do you see these new technologies?

Morris Chang: We recently launched four new technology platforms – mobile, high performance computing (HPC), automotive and the Internet of Things (IoT).

HPC covers cloud computing and server-related products, and the AR, VR and AI you just mentioned are mainly part of the HPC category, but they are also seen in the automotive and IoT domains.

CW: Which one do you think will contribute the most to TSMC’s growth in the future?

Chang: In the next five years, I think it will still be mobile and high performance computing. But if you extend the time frame to 10 years, then the last two will pick up. In the second five years, IoT and automotive electronics may surpass mobile devices.

As for high performance computing, it will be a big part of both the first five years and second five years.

CW: The ITRS 2015 (International Technology Roadmap for Semiconductors) said Moore’s Law will come to an end by 2021, and companies will no longer have the economic incentive to continue shrinking the transistor. Some companies are predicting that semiconductors will gradually become a traditional manufacturing sector that will be dominated by price competition. Do you agree with this outlook? 

Chang: Saying it will “come to an end” seems overly dramatic. I sense that (the end of Moore’s Law) will not be 2021, but I do believe it will happen sometime between 2020 and 2029, maybe around 2025.

I agree that the semiconductor sector is gradually moving toward traditional manufacturing, but that doesn’t necessarily mean it will be dominated by price competition.

Just look at the 28 nanometer process [a less advanced semiconductor process]. We have many different types of 28 nanometer processes, and there is also this 3D IC. These are all new technologies unrelated to Moore’s Law.

CW: Soon after Rich Templeton took over as CEO of Texas Instruments in 2004, the company pulled out of the smartphone chip market and moved into the analog IC field. He has now led Texas Instruments to its highest market capitalization in more than 10 years, returning it to glory. How has it been able to repeatedly reengineer itself in the face of pursuers? What part of that experience can Taiwan’s high-tech sector learn most from?

Chang: It took 20 years for Texas Instruments to revive itself. The chief engineer was TI’s former chairman Tom Engibous [now one of TSMC’s independent directors]. What did he do? Outside of semiconductors, he sold all of the company’s businesses, including the defense and consumer electronics businesses, to concentrate on semiconductors.

Later, Templeton did in fact pull out of smartphone chips, but the logic behind that strategy was to focus on what TI was good at. It was the same strategic logic that Engibous had.

What can Taiwanese companies learn from that? Play to your strengths is maybe one lesson. Of course your strengths cannot be in a market that is shrinking. If the market is not shrinking, then you can learn from this experience, but if the market is shrinking, then you have to try something different.

CW: TSMC has previously invested in solar power and LEDs. Are there any plans in the future to develop new businesses outside of your core IC foundry business? Would you consider doing like Google and Apple and make acquisitions to get into new technologies?

Chang: We currently have no plans for new businesses outside of contracting. Our focus for the past 30 years has been on the IC foundry business. This “focus” has led to our success, but it has also limited us.

Our people are very specialized in this business, but that means that if they want to enter other businesses, they would have to learn them from scratch.

In 2009, I felt that solar power and LEDs were relatively similar to our IC foundry business, so I decided to invest in them. But the results were not very satisfactory, so we pulled out.

New People Needed to Succeed in New Businesses

You mention new businesses in the future. I think you have to hire a group of new people. With new people, new businesses are more likely to succeed.

But speaking of acquisitions, I won’t talk about Google or Apple. I’ll use another company as an example – Cisco. There was a period of time when it aggressively acquired new businesses, but I later asked Cisco CEO John Chambers how much of Cisco’s overall revenues those new businesses accounted for, and he told me it was negligible.

Intel bought a number of small companies to get into the mobile device market, but it also wasn’t successful.

Some acquisitions, however, are done to find new people, and I can’t rule out that possibility.

Overall then, we don’t have any acquisition plans at present. If something does happen, it would be to find new people.

CW: The new labor system requiring workers to have at least one day off every seven days has again made “working hours” a central point of contention between labor and management. TSMC promoted a policy a few years ago of “50 hours of work per week.” Even though the amount of overtime employees could work was limited, TSMC accelerated the pace at which it introduced new processes and increased capacity. Why was TSMC able to shorten working hours yet become more competitive?   

Chang: That’s completely founded on my own experience. I started as a base-level engineer and have been going for 61 years. I rarely ever worked more than 50 hours a week.

There were a few periods when I did (work more than 50 hours a week), the longest of them about three to four months when I first joined Texas Instruments. At that time, my colleagues called me a workaholic. I was putting in about 70 to 80 hours a week. I was 27 then.

But in the 61 years I’ve been working, I’d say I worked more than 50 hours a week only 1/20th or 1/30th of that time, and most of the time it was under 50 hours.

At a TSMC sports meet on Oct.22, TSMC Chairman Morris Chang circled the track as he typically does to boost worker morale. He then announced to big cheers that he was issuing NT$400 million in "red envelopes" to workers.

As for why TSMC could get better results after reduced its working hours, effectiveness is very important, a lot more important than how many hours you work. There are two similar words – efficiency and effectiveness. Of the two, effectiveness is particularly important.

From a management perspective, first, I advised my colleagues and particularly top executives that we held too many meetings. As to how to hold meetings, I’ve talked about that many times so I won’t talk about it today. Middle and top managers have spent less time in meetings and there were fewer of them, but this is just one factor.

Second, you know there is a so-called knowledge pyramid, with “data” at the bottom of the pyramid, “information” above that and “knowledge” at the very top. We wanted to increase the proportion of “knowledge,” which brings about a higher degree of effectiveness.

Engineers spend a lot of time collecting data, and we wanted to reduce the amount of time spent in this area. Engineers need to spend more of their time on high value-added work like doing more analysis and finding ways to improve processes. If you knew what kind of data you wanted to collect, the time spent collecting data could be cut down. That’s knowledge.

The third point is that in our company, cycle time is very important. For every one of our new processes, we have to test over and over again, kind of like learning through trial and error. That’s a process that can be shortened. Reducing the number of times you go through trial and error requires a little “foresight.” You don’t want to test everything. Instead, you want to first test things that have a better chance of success.

The production cycle time of every test needs to be shortened and a little faster. It takes us many months to produce a wafer (using 16 nanometer processes and below) because of the complicated production process. If the cycle time can be shortened, you can really save a lot of time.

We have initiated a “Night Hawk Plan” [TSMC’s program to accelerate development of its 10nm and 7nm processes] in which R&D engineers work around the clock in three shifts. Very few people used to be willing to work the graveyard shift, so what we did is to have people in the same division rotate shifts, taking turns working night shifts. We have about 2,000 engineers working on the 7-nanometer process, so we can run multiple shifts.

The fourth point is to effectively use big data analysis technology.

CW: Labor-management relations have faced rarely seen tensions in Taiwan recently. You have experience managing companies in both the United States and Taiwan. Could you give us your views on how you see this issue?

Chang: There is nothing I want to see less than labor-management disputes. American high-tech companies do not have unions. Google, Amazon, Facebook, Microsoft – not one of them has a union, nor does Intel or Texas Instruments. 

I’m going to say something that may surprise you: When you look at why those companies are successful, I think them not having unions is a big part of it.

Labor Disputes Ruined U.S. Car Industry

When the high-tech sector first sprouted up in the United States, I was “present at the creation” to borrow the words of former American Secretary of State Dean Acheson [the title of Acheson’s 1969 book on the Cold War]. At that time, we had already clearly seen that labor-management disputes had brought down the American car industry.

So the American high-tech sector decided then that it would not allow unions.

In the 1970s, I was head of global semiconductors for Texas Instruments. At the time, a national union wanted to go to our Houston factory to organize a union there. So I went to the plant and talked to all of the 1,000 to 2,000 employees. Under American labor laws, you need the approval of at least one half of the employees to form a union. The votes were cast after one or two months, and very few of our employees endorsed the formation of a union, falling far below the 50 percent threshold. Other high-tech companies had similar experiences but the result was the same – no high-tech company had a union and it remains that way today.   

The chairmen and CEOs of these high-tech companies would absolutely agree with me: that one of the keys to their success is that they don’t have unions because for a company to succeed, everybody has to work together. When you have labor-management adversity, it’s really bad. Maybe it brings short-term benefits to workers, such as slightly higher wages and slightly lower working hours, but in the long run, it’s bad for the workers and bad for society as a whole.

But I also have to say Taiwan’s situation is a little different, so I want to modify what I just said. What I want to say is that good companies should be able to request that workers not form unions.

Translated from the Chinese by Luke Sabatier

---

Chinese Version: Google、臉書為何成功 張忠謀：沒有工會