For Taiwan Semiconductor Manufacturing Corp. (TSMC), the earthquake was just one shakeup in an unusually volcanic year. Earlier in the year, faced with rising demand for its services, it laid plans to aggressively build capacity. In June 1999, it acquired a 30 percent interest in Acer Semiconductor Manufacturing Inc. (ASMI). In January 2000, it acquired the remaining 70 percent of ASMI and all of a new Taiwan competitor, Worldwide Semiconductor Manufacturing Corp. (WSMC).
Observers are now wondering whether TSMC has taken on too much capacity and too many market risks to straddle the fault lines of rapidly changing technologies in a global economy. Will profits and customers evaporate as demand slows? Will the competition catch TSMC with its margins down?
Value Chain Visions
TSMC has led a revolution in semiconductor manufacturing. Just thirty years ago, almost all integrated circuit (IC) or “chip” manufacturers were “integrated device manufacturers” – IDMs – that designed, fabricated and marketed their own products from start to finish. Gradually, however, more and more IDMs contracted out much of their chip and component manufacturing, especially for their low-profit products, rather than build new or expanded factories or “fabs.” A new fab can cost up to $3 billion and take two years to bring on line.
Morris Chang, who founded TSMC in 1985 and continues as its CEO and chairman, was one of the first to see the potential of specialization in the semiconductor value chain, specifically the potential of the “pure foundry,” a contract manufacturer of IC wafers serving countless independent customers but not manufacturing its own products. Today, TSMC is the world’s largest pure semiconductor foundry.
TSMC’s early customers, who included Texas Instruments and Intel, were gained largely from contacts Chang made during his earlier 25-year career with TI where he ran TI’s worldwide semiconductor business. According to Chang, however, these early customers tended to use TSMC as extra capacity to manufacture “the kind of stuff they didn’t want to do themselves.” “They used us as a buffer,” he says.
The Fabulous Fabless
Chang and his team steadily built a growing customer base, but the biggest boost came in the late 1980s when entrepreneurs, with great ideas for chip designs but limited or non-existent resources to fabricate them, began outsourcing the manufacture of their products. Such firms became known as “fabless” companies, a significant new segment of the semiconductor value chain, and today they constitute about two-thirds of TSMC’s business. Examples of fabless companies are Broadcom Corp. and PMC-Sierra, Inc., both producers of networking semiconductors; Transmeta Corp., a producer of low-power microprocessors for mobile Internet computers; Silicon Laboratories Inc., a producer of communications semiconductors; and Synaptics, Inc., a supplier of human interface solutions for laptops and other mobile consumer appliances.
The fabless percentage of the world IC market continues to grow, from 3.9 percent in 1994 to 7.6 percent in 1999, and it is projected by Dataquest to be 11.8 percent in 2004. Some fabless companies use IDMs to manufacture their products – Transmeta uses IBM – but more and more of them are using foundries like TSMC. Fabless companies account for about two-thirds of TSMC’s sales.
“The fabless model permits companies like ours to focus on what we are especially good at doing – designing and developing great chips,” says Greg Aasen, chief operating officer of PMC-Sierra, which has its networking chips manufactured by TSMC and Chartered Semiconductor Manufacturing Ltd. of Singapore. “We focus on our part of the supply chain – on intellectual property development and on our customers and what they need. Our foundries focus on their part – turning out flawless chips at low cost. The entire chain is a series of high value-added links.”
IDMs, wary of the cost of building their own new fabs in times of rapidly changing technologies, are also adding to TSMC’s customer base. Motorola, for example, entered into a large production agreement with TSMC in 1999 as part of a plan to outsource 50 percent of its chip manufacturing. According to the Gartner Dataquest, IDM demand for foundry production is expected to triple in the years 1998 to 2003, from 27 percent to 33 percent of the foundry market. For TSMC, its IDM business rose from 25 percent of sales in the first quarter of 1999 to 32 percent in the fourth quarter. Its fabless business represents the remainder.
TSMC’s financial results for the first half of 2000 reflected the robust growth of the foundry business. Its first half sales revenues more than doubled, to $2.1 billion from $959 million, and its earnings more than doubled as well, to $760.2 million from $326.1 million.
Bedrock Concerns
Are there potential fissures in the bedrock of TSMC’s solid-looking business? One concern is whether TSMC has added more capacity than it needs. By the end of 2000, with the addition of ASMI and WSMC, its capacity will expand to 3.4 million silicon chip wafers per year (8-inch wafer equivalents) from 2 million wafers in December 1999. [See Semiconductor Manufacturing Process sidebar below.] But despite this substantially larger capacity, industry analysts expect that demand for foundry services will exceed supply well into 2001. Moreover, adequate capacity is helpful in assuring customers that their chips will be produced in times of strong demand. TSMC’s new capacity, therefore, should prove to be a competitive advantage at least in the near term.
A second concern is whether TSMC will be able to maintain customers’ confidence and attract new customers. A key factor here is the quality of its work, one that has as much to do with the Taiwanese people as it does with TSMC itself.
When Chang was with TI, which maintained IDM semiconductor fabs in both Japan and United States, he observed that the Japanese fabs achieved consistently higher yields of usable chips per wafer than the American fabs. He concluded that this was due to the better technological training and lower turnover of the Japanese fabs’ workforce, and he bet that the Taiwanese, with their strong work ethic and manufacturing know-how, could do even better. While it is difficult to establish that Taiwan’s semiconductor workers are the best in the world, it is clear that they are very, very good. Taiwan’s workers in all of its industries are renowned as exceptionally hard-working and adept in their manufacturing skills. In the case of TSMC, they are also highly educated. Some 5,000 of TSMC’s 13,000 employees are engineers, 3,000 of them with advanced engineering degrees. The quality and output of TSMC’s workforce, especially in terms of time-to-market, has proved to be a key factor in retaining and attracting customers.
Keeping Confidences
Another factor in maintaining customers’ confidence is protecting their intellectual property (IP). TSMC often has the work of 30 or more different customers on its shop floors at any given time, and individual TSMC engineers routinely work with the designs of several competing customers, theoretically presenting risks to IP. But this does not appear to be a concern to customers. Ed Healy, vice president and general manager of the Wireless Division of Silicon Laboratories in Austin, Texas, for which TSMC is the sole fab, says that TSMC is very careful in protecting customers’ IP. “They have to be. Customer confidence is their bread and butter.” Aasen of PMC-Sierra, which uses TSMC and Chartered as its foundries, adds that misappropriation of chip designs is unlikely since the manufacturing process is so complicated. “The chances of successfully manufacturing a stolen design elsewhere are very low,” he says. And there are additional reasons why this is not a concern, according to Steven Chang, a former official of WSMC and now managing partner of China International Development Consulting, Inc., a Taiwan venture capital firm. (The three Changs mentioned in this article are not related.) “With product cycles so short today, a stolen design would be useless within a matter of months,” he says. “Also, because TSMC is a pure foundry, with no products of its own, there is no real motive for misappropriation.”
Notwithstanding this, TSMC has worked hard to instill high ethical values and performance standards throughout its organization. “This was a high priority effort in 1991 and 1992,” says Don Brooks, who was president of TSMC from 1991 to 1997 and is now chairman of KLM Capital, a venture capital firm in California, and also a director of TSMC’s principal competitor, United Microelectronics Corp. (UMC). “We made a major effort to translate the values, needs and expectations of our customers to the TSMC staff. This involved many operating meetings talking about what it takes to succeed in the foundry business, including the need to convince customers that their IP was safe with TSMC and that TSMC, as a pure foundry, would not go into the product business in competition with them.”
Competing Conditions
A third concern is TSMC’s competition. Strong competition in the long run can lead to the loss of customers, price erosion, and overcapacity in the industry. While TSMC is currently the world’s largest foundry, with about half of the industry’s total shipments, another Taiwan foundry, UMC, has a market share of about 30 percent and is substantial and growing [see UMC sidebar below]. And there are other competitors, including Chartered in Singapore, with a market share of about 12 percent, Anam Semiconductor in South Korea, with a market share of about five percent, and smaller players such as Newport Wafer in the United Kingdom and Tower in Israel. In addition, profitability in the industry is attracting new competitors, including two in Malaysia: First Silicon in Kuching and Silterra in Kulim.
Separately, there is competition from several dozen IDMs that provide foundry services with their idle capacity, including IBM, Seiko, Mitsubishi and Toshiba. However, these do not constitute strong competition for the foundries because many customers feel that their IP is not safe with an IDM, which might produce competing products, and that their orders could be put on a back burner in times of strong demand.
Harvey Chang, TSMC’s senior vice president and CFO, answered the competition issue in an interview with Upside. He said that TSMC has three important competitive strengths. First, its operations are based on a “cluster” concept. The majority of its “fabs” or factories are in close proximity (seven of its twelve fabs are in the Hsin-Chu Science Park, south of Taipei), enabling the sharing of facilities and personnel, the coordination of technical support, and a high utilization of equipment. Second, a foundry is a service-driven business, and TSMC’s customer service, which has “Jump Through Hoops” as its slogan, is second to none, he claimed. Third, it makes economic sense for TSMC’s customers to stick with one foundry. In a complex manufacturing process that requires careful integration of the customer’s design with the foundry’s systems and equipment, a working relationship develops between customer and foundry that itself has economic value. For customers, there is a cost to changing foundries.
Ron Norris, TSMC’s senior vice president of worldwide marketing and sales, added to the list of TSMC’s competitive advantages. “Increasingly, because of the growing complexity of semiconductors, customers must match their designs with a foundry’s technology, making it more difficult for them to change foundries,” he claimed. “Also, as to competition from IDMs, foundries have more flexibility in utilizing capacity and can therefore produce chips in a shorter time and in greater quantities.”
Trusting TSMC
How do customers feel about these competitive advantages, some of which appear to be anticompetitive tying arrangements, preventing them from changing to a less costly chip manufacturer? Healy of fabless Silicon Laboratories, for which TSMC is the sole foundry, has some concern that prices will rise and production will be rationed in times of strong demand. “But we have the highest regard for TSMC, and since we are a loyal customer we believe they will be fair to us in times of strong demand,” he says. “Because they are a large foundry, we believe that they will have capacity for us and that their size will help to moderate price swings.”
Norris acknowledges that good customer relations, which are a key ingredient of the foundry business, tend to curb price increases. “You can’t easily raise prices or cut production without damaging your relationship with a customer,” he says. “We have ongoing discussions with our customers about prices and capacity. We try to anticipate the future and manage their expectations.” He also acknowledges that good customers – those that place most or all of their foundry business with TSMC – receive more favorable treatment than other customers, in terms of their production needs, during periods of strong demand.
It is also true, however, that foundries give more favorable treatment to customers who have high growth and whose products are higher on the value chain, such as high-end logic semiconductors, for which premium prices can be charged. (There is about a 50 percent gross profit margin on high-end chips.) This means that customers with low-end products, such as memory chips, often end up at other foundries. Aasen of PMC-Sierra, which uses TSMC and Chartered in approximate equal amounts, says that TSMC is now producing the greater proportion of his company’s more advanced chips.
Many foundry customers hedge their bets by having chips produced at two or more foundries. As demand grows, this appears to be a trend. Andrew Hsu, technical marketing manager of Synaptics, reports that his company is considering using an additional foundry. Also, Transmeta, whose sole foundry is IBM, is rumored to be doing the same.
Success and Succession
A final concern, in the minds of some observers, is the business culture and management structure of TSMC, which is highly centralized and disciplined. While this style is appropriate for a company that manufactures complex products, it tends to drive out talent and suppress innovation. One consequence can be that strong leaders will not emerge from within the company’s ranks. Another consequence can be that the company’s manufacturing technologies will not keep up with its competition.
A related issue involves the succession of Morris Chang, the 69-year-old founder and CEO of TSMC. Although Chang has built a strong organization, he has not identified a successor nor has he loosened his tight grip on the reins. Chang – referred to by some observers as “Warlord Chang” – is clearly in charge. Other observers, however, believe that TSMC is such a strong organization that succession will not be a problem. And in the words of one of the TSMC executives interviewed for this article, “Any of the top management could step in.”
On September 20, 1999, the day before Taiwan’s big earthquake, TSMC’s fabs at Hsin-Chu Science Park were humming away with accustomed ease. The next day there was nothing but chaos. TSMC got through that day and, unlike many businesses in Taiwan, was back to full production within two weeks. But as competition grows and technology changes, do more earth-shaking events lie ahead for TSMC?
Sidebar – United Microelectronics Corp. (UMC)
Taiwan’s UMC is the second largest and the fastest growing dedicated semiconductor foundry in the world, its share of the world foundry business having grown from five percent to about 30 percent in the last five years. Like Taiwan Semiconductor Manufacturing Corp. (TSMC), its customers include both integrated device manufacturers (IDMs) and fabless companies, and its financial results are strong. For the first half of 2000, its net income was $627.5 million on sales revenues of more than $1.4 billion.
UMC started out as an IDM and began full production in 1982 as the first occupant of Hsin-Chu Science Park, south of Taipei. Its early production consisted of the simple chips used in calculators, watches, toys and other consumer products, and it remained a small player in the foundry business until 1995 when it formed joint ventures with a number of fabless companies, including Xilinx, S3 Inc. and SanDisk Corp., to build new “fabs” or factories. Restructuring and reorganization continued. In 1997, UMC spun off its own product lines in order to become a pure foundry and hired away two TSMC executives, its president, Don Brooks, and its CFO Gary Tseng. In 1999, UMC bought out its fabless joint venture partners, who continued as customers, and consolidated its operations in order to gain greater control over process technologies and reduce costs.
UMC now has nine fabs and more than 300 active customers, 50 percent of them in the United States and 20 percent in Europe. Its strategic focus is on developing more advanced production technologies, primarily on reaching smaller and smaller feature sizes – as small as 0.13 micron – and on perfecting copper interconnects. [See Semiconductor Manufacturing Process sidebar below.] Such advances are expected to permit smaller, faster and more complex chips to support new applications and higher prices.
Industry observers claim that UMC is not just a carbon copy of TSMC. Some believe that UMC is slightly ahead on process technologies, but Don Brooks, former president of TSMC and now a director of UMC, believes that the two companies are practically in a dead heat in all categories except for capacity, where TSMC now has a clear lead.
Is there a risk that these two companies, which hold a combined 80 percent market share of the world foundry business, would attempt to corner the market and raise prices? “No chance,” says Ron Norris, TSMC’s senior vice president of worldwide marketing and sales. “We’re fierce competitors!” Industry observers confirm this and point to the fact that each of the two companies is headed by a strong leader, Morris Chang of TSMC and Robert Tsao of UMC, whose management styles are sharply. Chang has built a company that is highly centralized and disciplined. Tsao has built one that is decentralized and flexible. Each wants to be better than the other. It is indeed difficult to imagine that the two would cooperate on anything.
– David James
Sidebar – Semiconductor Manufacturing Process
Integrated circuits (ICs), or semiconductor chips, come in many sizes, shapes and degrees of complexity. Among their many types and uses are microprocessors found in computers, digital signal processors (DSPs) found in communications devices, and application specific integrated circuits (ASICs) found in sound cards and graphics accelerators. What they seem to have in common these days is their decreasing size and increasing complexity.
There are five steps in the manufacture of a chip: specification, design, silicon fabrication, testing and assembly or packaging. The customer first sets the specifications for what it wants, its designer then lays out the logic design and makes certain that the chip would work “on paper,” and next the foundry manufactures large numbers of the chip on a single silicon wafer. Each chip is then tested in place, after which the wafer is broken up, and each chip is assembled/packaged and then tested again. After this, the chip is plugged into the customer’s product and then typically tested a third time.
The fabrication of the silicon wafer is the most difficult step and perhaps the most complex manufacturing process in the world. It consists of coating the wafer with insulating and photosensitive materials, projecting circuit images on the wafer surface in a lithography process (some circuit images being as small as 0.13 micron), etching the images into the silicon surface, and coating the layer with aluminum or copper that connects to circuits in an underlying layer. This process is repeated to form up to 30 layers of circuitry on the wafer. The entire task can involve as many as 600 different steps.
Foundry engineers work closely with designers during the fabrication process, and many adjustments are made during the process to achieve a working chip. Not only must there be adjustments to the manufacturing equipment, there must also be adjustments to the design to match it to the foundry’s process technologies. Once a chip’s manufacturing process is perfected, the “cycle time” for the manufacture of a wafer averages 1.6 days per layer. The more layers, the more time it takes to complete a wafer. But the smaller the chip or the larger the wafer, the more chips there can be on a single wafer, thereby reducing the cost of each chip. Whereas most foundries today produce wafers that are six inches in diameter, there are many now that produce eight-inch wafers and a few, the newest fabs, that produce 12-inch wafers.
– David James