TSMC Stuck to Plain Contract Manufacturing — What Role Does It Play in the Chip Industry, and Why Was Morris Chang's Retirement a Big Deal? | QDaily Business History

TSMC changed chip manufacturing, and with it the course of the personal computer and the mobile phone

"No one in this field is going to write a Principia (Newton's three-volume Principia) and have it last 200 years. The world of computers has no paintings that people gaze at for centuries, no churches people revere for hundreds of years. No — in this field a person does his work, and ten years later it is out of date, and in another ten or twenty years you can't even use it."

"The computer field is a bit like sedimentary rock. You contribute one thin layer inside a mountain, and the mountain gets a little taller. But in the end people only stand on the summit; you'd need an X-ray to see what is inside."

That is how Steve Jobs described his own work in 1994, when he was struggling to keep his second company, NeXT, alive, in an interview with the Silicon Valley Historical Association.

Jobs himself is still remembered, because three years later he went back to Apple and launched the iPod, the iPhone and the iPad under the spotlight. But most of the people who built the computer industry have faded from memory, exactly as he said.

On June 5, after chairing TSMC's 2017 annual shareholders' meeting, 87-year-old Morris Chang formally retired from the company he founded. Next to the fanfare around Li Ka-shing's retirement, the news was quiet, and most coverage was over within two days. Taiwanese media mostly called him the "godfather of Taiwan's semiconductor industry," wrote about the end of the industry's "era of strongmen," and, of course, reached again for the worn-out phrase "the pride of Taiwan."

What TSMC gave the world is far larger than a high-tech industry for Taiwan.

TSMC is worth more than $200 billion today, one of the 30 largest listed companies in the world. But in the 31 years since it was founded it has done one thing only: build chips that other companies designed. It was the first company anywhere to specialize in that business, and it started the division of labor in chipmaking — some companies design, others manufacture.

Because companies like TSMC took on the ever more complicated work of manufacturing, design specialists — Nvidia, ATI, Qualcomm, Broadcom, even Apple — could put all their effort into chip design. That division of labor brought the 3D graphics revolution to the PC era, and it was one of the direct causes of the smartphone era. Now that the chip industry talks about autonomous driving, TSMC is still the pillar holding it up.

Like Robert Noyce, Bill Gates and Steve Jobs, the company Morris Chang built changed the course of technology over the past several decades, laying an indispensable layer inside that mountain.

The Integrated Circuit: the Bedrock of Today's Computers and Internet, and the Start of TSMC

TSMC stands for Taiwan Semiconductor Manufacturing Company. As the name says, it manufactures integrated circuits — chips — in Taiwan.

The integrated circuit is where the modern computer industry begins: it packs more complicated circuitry into less space. Before it was invented in 1958, a single computer assembled out of transistors could fill most of a room.

But the integrated circuit had nothing to do with TSMC or Morris Chang. That year the 27-year-old Chang had just joined the old semiconductor firm Texas Instruments. In the same year Jack Kilby, an engineer who had joined not long before him, built the first integrated circuit.

"While I was working day and night at Texas Instruments, something world-shaking was quietly happening right in front of me. But I didn't understand then what the integrated circuit was actually for — it always felt far away." That is what Morris Chang wrote decades later in his autobiography. He was not wrong at the time: Kilby's first design used germanium, which could not be mass-produced.

Fairchild Semiconductor solved that. In the 1950s eight young PhDs quit Shockley Semiconductor Laboratory and founded the company with help from the banker Arthur Rock. Among them were Gordon Moore, who would later state the law that carries his name, and Robert Noyce, the future founder of Intel. Half a year after Kilby's invention, Noyce built an integrated circuit on a silicon wafer, which made mass production possible.

Digital chips today are countless times more complicated, but the underlying principle is still Noyce's silicon integrated circuit.

Two huge government programs set off the boom. Purchasing for the Apollo moon-landing program and the Minuteman missile program brought growth to the whole semiconductor industry. Over the next half century integrated circuits filled the electronics of daily life, from satellites to microwave ovens, from rockets to phones. They also kept shrinking, until they became the microprocessors — the chips — inside those devices.

Over the following 20 years America's most important investors and scientists began to see that the semiconductor industry behind the chip was a stage with enormous opportunity. Money, technology and young people poured in.

Through all of it Morris Chang stayed at Texas Instruments, rising to vice president, running an integrated-circuit manufacturing business with 30,000 employees.

Twenty Years of Moore's Law Later, the Personal Computer Took Shape — and Taiwan Began Planning TSMC

Gordon Moore, who had studied chips for years, stated Moore's Law in 1965: roughly every 18 months the number of transistors on a chip would double, the price would halve, and performance would double. More important than the law was that Intel delivered on it. Intel's 4004, launched in 1971, held 2,250 transistors; the 386, launched in 1985, held 275,000. That was the performance the personal-computer revolution was built on.

As chips matured, the personal computer took shape. The first IBM PC appeared in 1981, and IBM then opened the standard to third-party makers. As those makers fought each other, the cost problem was solved.

In 1984 the Apple Macintosh, developed by a team Steve Jobs led, was released. It cost too much, but it settled the direction of the personal computer: a graphical interface and a mouse. Computers became easy to use.

And then there was Microsoft, which through the late 1980s steadily improved Windows and the Office software. The personal computer stopped being a geek's garage toy; companies started buying them in bulk to get more work done.

Another clear trend of the 1980s was that manufacturing began moving from the United States and Western Europe to Asia. Back in the 1960s the world's garment-making center had already shifted for the second time, from the United States and Japan to newer regions such as South Korea and Taiwan.

In those places, Taiwan included, wages were half of Japan's or less, and labor was plentiful. So the big firms in electronics, electrical machinery and synthetic fibers — industries that needed large numbers of workers — began expanding into Taiwan. After 20 years of doing that simple processing work, Taiwan decided to move up from basic garment assembly. Sun Yun-suan, then premier of Taiwan, set the goal of building a semiconductor industry in the 1970s.

Sun's idea was to turn Taiwan into a semiconductor business center by building one company that could do everything — architecture design, production, packaging and testing. That was exactly Intel's position at the time. To do it, in 1982 he approached Morris Chang, who ran a 30,000-person production organization and was the most experienced Asian executive in the industry.

Sun did not persuade him on the first try. After 25 years in semiconductors, Chang thought Taiwan's foundation in the field was close to zero and that it had no seasoned professionals. A company like Intel, designing and building its own chips, was bound to fail in an industry that heavy in assets and technology. Years later VIA Technologies did in fact fail at exactly that.

In 1985 Texas Instruments decided to move into consumer electronics along Apple's lines, which ran against everything Chang had planned. Meanwhile the government in Taiwan had been inviting him back for several years.

In the end Chang, after 25 years at Texas Instruments, resigned — and made up his mind not to build another Intel, but to do only the back half of the business: chip manufacturing. In his memoir Chang wrote that he had watched colleagues leave to start companies around chip-design ideas, and watched many of them fail because they could not raise enough money to build a fab and make the chips.

No one had taken this road before: he was carving a division of labor out of an industry that had none. At 54, Morris Chang started his first company.

TSMC Started with a Huge Pile of Capital, and Spent Without Limit on Manufacturing Technology

"Not many people liked the idea of a business that only fabricated chips. The government in Taiwan was probably my strongest backer then, but even they didn't really understand it — they were just willing to invest and let me try." Twenty-two years later, Chang, by then in his seventies, recalled the experience in a conversation with Nvidia CEO Jensen Huang at a Computer History Museum event.

Investors were not the only doubters. When Chang took the idea to Gordon Moore, Moore rejected it outright. Moore, and plenty of others then, believed that both manufacturing and packaging and testing demanded very high capital from a chip company — manufacturing above all, which needed enormous sums poured into research on new technology.

Talking with Jensen Huang, Chang recalled that "in TSMC's early days one investor who held 5% took me to dinner three times, and each time he had a long list of questions. But he knew nothing about semiconductors. The only thing he worried about was how TSMC was going to compete with a giant like Intel."

"TSMC isn't competing with them. This is a completely new model, and we may even work together," Chang answered, laughing. But there was no one who had already succeeded on this road to point to, and that frightened the investor more.

To raise more money Chang wrote to semiconductor companies including Sony and Mitsubishi, and every one of them turned him down. In the end it was the government in Taiwan that helped him from behind the scenes. Chang said three members of the fundraising team were government ministers: "As I remember it, there were five outside investors in the end. I talked two of them into it; the other three were told to invest by the government."

Taiwan had set up policy funds of NT$800 million in 1984 and NT$1.6 billion in 1986 to guide venture investment, paid for by the Executive Yuan and state-owned banks, with the aim of developing new technology.

TSMC was announced in 1986. The government in Taiwan helped bring in Royal Philips as a co-investor, and together they put in about $220 million, of which the government's share was 48%. Adjusted for more than 30 years of inflation, that is $3.6 billion in today's money. Unlike Apple, Microsoft and the other small companies that started in garages, TSMC was a large, heavily funded company from its first day.

Chang needed that much money because he knew chip manufacturing burns cash.

Chip production is extremely complicated, and Moore's Law keeps making it more so.

Put simply, silicon rods are first refined out of silicon dioxide (SiO2) ore — quartz sand, for instance — through a series of chemical and physical smelting steps, then sliced into round single-crystal discs. Those discs are wafers.

The wafer then goes through repeated etching, photomasking and other steps that lay tiny circuits across its surface. Only after strict testing, dicing, packaging and testing again does it become a finished, usable logic chip — perhaps a CPU, perhaps the baseband processor that handles mobile communication in a phone, perhaps the small chip that gets a computer onto Wi-Fi.

Building the fab does not end the spending. Every step up in wafer size and in process node takes another large helping of capital.

The "process node" you hear about with Intel, Apple or Qualcomm chips is really the spacing between the tiny circuits on the wafer — from 3 microns at the start down to 10 nanometers and 7 nanometers today. As Moore's Law predicted, the chip industry moves up a node roughly every year and a half. Wafers have grown with the technology too, from the original 2-inch, 4-inch, 6-inch and 8-inch discs to the 16-inch wafers of today.

A smaller node and a bigger wafer mean more circuits on one wafer, more chips cut from it at the end, and a lower cost per chip.

For a young technology company without much capital, absorbing the costs that each new generation brings — let alone building a fab to make its own chips — is close to impossible.

What TSMC does is take all that complicated production off their hands, so a new company can start designing chips from nothing.

With the money in hand, Chang put up TSMC's first fab in 1987. In its first year the plant could run only 3-micron and 2.5-micron processes, its annual capacity was under 7,000 six-inch wafers, and its yield was poor. It could not win orders from big companies, and the whole company ran at a loss.

But Chang never stopped upgrading the plant. He brought in Intel to certify it and raise its level, and 200 technical problems were fixed within a year. From then on, to keep TSMC ahead in chip technology, Chang insisted on heavy capital spending for as long as he ran the company.

That spending has never let up. Even in 2001, when the dot-com bubble burst and internet and computer companies were failing in batches, Chang still believed demand for chips would not go away, and he pushed the fabs' research spending to 80% of net profit that year — a year in which the bubble had already crushed TSMC's profits.

By 2015 Intel's capital spending had fallen to $7.3 billion, while TSMC's held at $8.1 billion.

On the figure that matters most, research spending, TSMC has kept it at 20% of net profit ever since 2000. As a share of revenue it has grown from 3% in the 1990s to 8% today.

While holding that level of investment, TSMC's revenue grew 44-fold between 1994 and 2015, an annual growth rate of 19.7%. By 2015 it was turning out 9 million wafers a year, the largest wafer foundry in the world, with half of all foundry output.

Looking back now, pouring money into chip manufacturing technology seems obvious and highly profitable; Samsung has made a fortune from its semiconductor business too.

But look back further and there were too many hot new technologies in the 1980s. Voice recognition, VR and artificial intelligence were all taken as symbols of the future at one point or another. Picking the right one was never that easy.

TSMC Made the Division of Labor in Chips Possible — the Foundation a New Generation of Chip Companies Grew On

As TSMC grew, chip foundry work became an industry of its own — not only Taiwan's UMC, which gets mentioned often, but later Samsung and even Intel went into the foundry business.

That left chip-design companies of every kind with nothing to worry about on the production side.

In the 1990s a wave of new semiconductor companies appeared in North America: Broadcom, Nvidia and Marvell were founded one after another. Each had its own specialty — networking chips, graphics processing, Wi-Fi chip design — but they had one thing in common. None of these startups owned a fab, and none had the money or the people to handle production themselves. They badly needed outside wafer lines.

TSMC became the choice for most of them. In 1998 Nvidia signed a strategic partnership with TSMC and handed over production of all its graphics accelerator cards. In the second year of that partnership Nvidia introduced the idea of the GPU, the graphics processing unit, which could take some of the computing load off the CPU, the central processor — dynamic graphics rendering, for instance. The whole semiconductor industry later adopted the idea.

The GPUs Nvidia went on to release were chased hard by the market, and TSMC, which built them, grew fast as well. Between 1987 and 1991 TSMC's compound sales growth reached 74%; over the years since, its average compound sales growth has been 19.7%. At the shareholders' meeting where he announced his retirement, Chang said outright: "Even I think TSMC's growth over its 31-year history has been something of a miracle."

Nvidia, for its part, benefited from TSMC's fast, efficient production, holding a compound growth rate of nearly 70% a year through those years. Jensen Huang, who is friendly with Chang in private, once half-joked: "If I'd waited to build my own fab to make GPU chips, I'd probably be sitting on a company worth a few tens of millions of dollars right now, an easygoing CEO."

TSMC's enormous spending meant chip-design companies no longer had to fund production lines of their own. That lowered the barrier to entry in design and cut the risk of a product that failed, which in turn brought a crowd of small and mid-sized firms into the design side of the semiconductor business.

Intel still designs and builds its own chips, holding tight to the lucrative desktop CPU market. But bit by bit the CPU stopped being the only chip that mattered.

A Diverse Division of Labor in Chips Directly Set Off the Smartphone Boom

In the personal-computer era power was not much of a constraint, and one very fast CPU could handle everything. In a phone, specific jobs had to be pushed onto dedicated chips wherever possible.

By 2006, when Apple began the final sprint to build the first iPhone, the chip industry already had a rich supply of chips for almost any job — and most of the companies that designed them had no factories, leaving the manufacturing to TSMC or another foundry.

iFixit once took the first-generation iPhone apart. You can see the model numbers of every chip inside on its site. We counted up the function chips that were designed by chip companies and built by foundries.

• CPU: Samsung ARM-architecture processor ARM1176JZF (Samsung designed and built it)
• Memory: Samsung S3C6400 (Samsung designed and built it)
• Wireless flash memory stack chip: 1030W0YTQ2 (Intel designed and built it)
• GPU: PowerVR MBX Lite 3D GPU (design and manufacturing split — Intel did the manufacturing at first, TSMC after 2009)
• Audio: Wolfson audio WM8758BG (Wolfson only designed it; manufacturing went to TSMC)
• Battery management chip: Linear Technology 4066 (Linear only designed it; manufacturing outsourced)
• Wireless network chip: Marvell W8686B13 (Marvell only designed it; TSMC built it)
• Phone signal power amplifier: Skyworks GSM/Edge (Skyworks only designed it; manufacturing outsourced)
• Bluetooth chip: Skyworks CSR BlueCore4-ROM WLCSP (design and manufacturing split)
• Phone signal management chip: Infineon M1817A11 (Infineon only designed it; manufacturing went to TSMC in 2009)
• Multimedia chip: Infineon PMB8876 (Infineon only designed it; manufacturing went to TSMC in 2009)

Of the 11 chips found in that first-generation iPhone teardown, only 4 were designed and built in-house by companies like Samsung and Intel. All the rest came from the split between design and manufacturing that TSMC had worked out.

By focusing only on production and staying out of design, TSMC created the conditions for the many different function-specific chips a mobile phone needs. It also meant the chips inside a phone were no longer controlled by a handful of chip giants.

Fabless design companies such as Qualcomm and NXP then watched their revenue climb as the phone market exploded. Qualcomm, also a TSMC customer, could put more than 20% of its revenue into research every year — designing and selling the most advanced chips on one side, and building a large stock of core patents to license to downstream manufacturers on the other. That, too, came from not having to spend on chip manufacturing.

Morris Chang Has Retired, but TSMC's Business Is Not Finished

Chip manufacturing moved along with the rest of manufacturing, but because wafer processing asks so much technically, it never slid down-market the way garment work did.

TSMC is worth about $200 billion now, roughly the same as Intel, and it briefly passed Intel in 2017. Samsung Electronics, with the world's largest phone business on top of its chips, is worth $280 billion — only half again as much as TSMC.

At the June 5 shareholders' meeting where he announced his retirement, Chang said he expected TSMC's 2018 revenue to grow 10-15%, driven by high-performance computing, various Internet-of-Things uses, and cars, among other areas.

TSMC's newest 7nm process is expected to go into production this quarter and reach full capacity in the fourth, at which point it could supply around 10% of TSMC's revenue. Chang has said "TSMC already has 100% of the 7nm market," implying that Samsung, one of its biggest rivals, has not landed a single 7nm order.

Looking ahead, C. C. Wei, one of the two men taking over from Chang, said TSMC plans to spend $25 billion over the next three years developing 5nm manufacturing technology. For comparison, TSMC's total revenue in 2017 was $33 billion.

The sums going into chip manufacturing keep getting larger.

But what people remember is the brand printed on the back of the phone and the logos of internet companies on its screen. "Intel Inside" is a thing of the past; you will not see a chip's mark anywhere on a phone. Names like Snapdragon are still spoken only by the press and by enthusiasts. The three largest phone companies — Samsung, Apple and Huawei — all design their own chips now, and Apple is shifting more of its chip manufacturing away from Samsung, a competitor, to TSMC, which is not.

Chips keep getting more invisible, but they still drive everything. Inside that mountain of computing, the division of labor TSMC brought about in the 1980s is still one of its essential layers.

Cover image source: Dongfang IC