+44 (0)24 7671 8970
More publications     •     Advertise with us     •     Contact us
 
Technical Insight

Taiwain's foundry race will have few winners (Cover Story - GaAs Manufacturing)

Taiwan's GaAs foundry companies have ambitious plans to capture large slices of the device wafer market, but most of these will not come to fruition, writes Tim Whitaker.
Taiwan s foundry race will have few winners The last two years have seen the emergence of a number of companies intent on establishing GaAs foundries in Taiwan. Buoyed by the strong growth in demand for GaAs ICs, these companies sought to emulate the success of the giant Taiwanese silicon foundries, notably TSMC and UMC. There has been impressive progress; Taiwan can already boast the world s first "pure play" 6 inch GaAs foundry company, WIN Semiconductors (see ). However, prospects are much less rosy than they were as recently as six months ago. The precise level of demand within the GaAs market is unclear at present, but it is apparent that the GaAs industry as a whole has plenty of capacity to sustain itself for at least one or two years. Coupled with the inherent problems in establishing a foundry company in the GaAs industry, no-one expects that all the GaAs foundry projects in Taiwan will survive. Several plans have already gone up in smoke. Last summer, a company named G-Advanced Semiconductor held a groundbreaking ceremony for a 6 inch GaAs fab, then promptly vanished without trace. More recently, Arima, a major OEM supplier of notebook PCs and wireless handsets, reportedly shelved its plans for Arima Broadband, a subsidiary company which was to have built a GaAs fab and offered foundry services. Even so, there are already four GaAs fabs in Taiwan (Hexawave, Transcom, Advanced Wireless Semiconductor Company (AWSC), and WIN Semiconductors), with three more under construction (Compound Semiconductor Manufacturing Corporation (CSMC), Global Communication Technology (GCT), and Suntek Compound Semiconductor Company). In addition, most of these companies already have plans to expand or to build second fabs. These projects will now be described in more detail, based on the information supplied by the companies themselves. AWSC Dating back to April 1998, AWSC is the oldest of Taiwan s GaAs foundry companies. The company operates a 4 inch HBT fab in Tainan, with a current capacity of 600 wafers per week. The facility is being expanded to double its capacity for 4 inch wafers. When Compound Semiconductor spoke to AWSC one year ago, the company was about to start the installation of a 6 inch manufacturing line, but these plans were subsequently put on hold. "Originally, we planned to do 6 inch for phase 2 of our FAB 1," says AWSC s Charles Wang. "However, it now seems that this is not a good time to jump into 6 inch GaAs. Therefore, phase 2 of FAB 1 will be 4 inch." AWSC now offers two InGaP/GaAs HBT processes, one suited for wireless power amplifiers and the other targeted towards ICs for high data-rate OC-192 fiber-optic networks. AWSC has a 6 inch fab (FAB 2) on its technology roadmap (see ), along with InP HBTs. According to Wang, the 6 inch fab is likely to be built in 2003, and will be expanded in two phases, each providing a capacity of 1500 wafers per week. AWSC has a foundry agreement in place to supply HBT wafers to Conexant, but reports suggest that the AWSC process has still to be qualified. WIN Semiconductors WIN (an abbreviation of Wireless, Information and Networking) was founded in October 1999, and completed construction of its fab, including a 20 000 sq. ft cleanroom, in July 2000. The company rightly claims to be the world s first "pure play" 6 inch GaAs foundry. Following equipment installation in August, the company demonstrated 6 inch, 2 m HBT wafers in November of last year, and 6 inch PHEMT wafers in December. In March, WIN demonstrated another milestone the first 6 inch MMIC wafer fabricated using a 0.15 m gate length PHEMT process (see ), which has fT and fmax values of 85 and 220 GHz, respectively. As shown in the company s technology roadmap (), WIN plans to enter full production in April 2001, initially offering 2.0 m HBTs for L-band PCS/ cellular phones, WLAN and two-way pagers. PHEMT processes of 0.15 m and 0.35 m, together with a 1 m HBT process for high data-rate fiber-optic applications, will follow towards the end of this year. WIN aims to process 10 000 6-inch wafers this year, rising to 30 000 wafers in 2002, and will reach its full annual capacity of 100 000 wafers in 2005. Annual revenues are expected to reach $0.5 billion. WIN believes the current fiber-optic and broadband wireless markets require around 250 000 6-inch equivalent GaAs wafers every year, and that this figure will reach 1 million wafers by 2005. If these figures are correct, WIN s capacity will supply 10% of the world market in 2005. CSMC Founded in July 2000, CSMC plans to start mass production in June of this year, at a level of 2000 4-inch wafers per month. Products will include HBTs and PHEMTs for both wireless and fiber-optic applications. CSMC s current 800 m2 clean room space will be expanded later this year to 2500 m2, and the company expects to have as many as 10 fabrication lines in place by 2003. Capacity is forecast to reach 16 000 wafers (4 and 6 inch) per month. GCT GCT is setting up a 6 inch GaAs fab in Hsinchu that will have a capacity of 5000 wafers per month. Mass production is expected to commence by the end of 2001. GCT also plans to construct a second, larger fab in 2002 (see Compound Semiconductor February 2001, p57). Suntek Suntek Compound Semiconductor Company was founded in August 2000 as a joint venture between Procomp Informatics, Taiwan s leading supplier of microwave epiwafers, and Acer Communication and Multimedia. Suntek is working closely with at least two established IDMs (integrated device manufacturers). According to Suntek s James Lin, the foundry has a technology transfer agreement with Mitsubishi, under which the Japanese company will transfer its 4 inch HBT processes to Suntek. "Mitsubishi also has a foundry service agreement to buy back HBT-based ICs from Suntek," says Lin. Furthermore, the US GaAs manufacturer Celeritek has invested $2.4 million in Suntek, and also has a foundry service agreement in place. Suntek s plan to implement its first fab, which will process 4 inch wafers, calls for construction to be completed by the end of April 2001. Following equipment installation and test runs, the production ramp-up phase is scheduled for the final quarter of 2001. Mass production is expected to commence by the end of this year. Suntek s initial products will be AlGaAs and InGaP HBTs. In 2002, the company expects to offer InP HBTs and 0.5 mGaAs PHEMTs, as well as optical devices such as photodiodes, lasers and VCSELs. Suntek also has plans to build a 6 inch line for the production of GaAs HBTs and PHEMTs. As shown in , 6 inch production is expected to commence in the final quarter of 2002. According to its current plans, Suntek plans to ramp its capacity to a level of 21100 6-inch equivalent wafers by the end of 2004. Analysis Even without the downturn being experienced by the GaAs industry at the moment, it will be impossible for all these companies to be successful to the full extent of their original plans. As shown in , Taiwan s GaAs production would exceed 1 million 6 inch wafers per year by 2005 if all the projects went forward, and this would be almost entirely composed of foundry capacity. The likely scenario is that most, if not all, of the "fab 2" projects will be put on hold indefinitely, or simply cancelled. Many observers have predicted that only one or perhaps two of the Taiwanese foundries will be successful. Of the larger projects, WIN has achieved a "first-mover" advantage, GCT has strong but complicated links with an established US GaAs foundry, GCS, and Suntek already has two foundry partners, Mitsubishi and Celeritek. Picking a winner will certainly not be easy. The challenges of founding a foundry Taking a high-performance GaAs process from an R&D environment to a high-volume manufacturing facility offering commercial foundry services is fraught with problems. One person with first hand knowledge of this area is Ding Day, president of Network Device Incorporated (NDI), established in February 1997 as the GaAs industry s first "pure play" foundry. In February 2000, NDI was acquired by Alpha Industries. Attracting investment Taiwan s GaAs boom is based on the success of the foundry model in the silicon industry and the stellar performance of companies such as TSMC and UMC. By drawing parallels between TSMC and their own business plans, the technical teams behind Taiwan s GaAs foundry projects were able to attract plenty of investment. "The technical teams told their investors that the market would be big; that they had the best technology and would capture the whole of the GaAs foundry market; and that they would become the TSMC of the GaAs industry," says Day. "Many fab projects were funded on this basis alone." From his experiences with NDI, Day understands that new foundry companies face several major challenges. The first is that the technical teams involved in setting up foundry companies tend to lack operational knowledge. "In most foundry companies, the technical teams have a background in R&D or in small-scale manufacturing," says Day. "They lack real exposure to high-volume manufacturing." He contends that wafer fab operation skills are equally as important as the basic technology. "Foundry is basically a service industry. You need to have the right infrastructure in place for high-volume production, such as establishing foundry manuals, setting up quality control and assurance, getting ISO qualification and installing manufacturing execution software for wafer tracking, data collecting and engineering, production control and customer interfaces. You also need a very strong management team that can deal with issues like quality, delivery and customer support. It s a huge task to try to emulate TSMC in the GaAs world." In fact, TSMC started life as a spin-off from Philips, so a lot of the manufacturing infrastructure was already in place. The other problem, says Day, is that the financial backers for the Taiwanese GaAs foundries probably made their decisions based on insufficient information or overly optimistic data. "Investors have been told that the overall market demand for GaAs wafers and the foundry market are several times higher than they actually are." According to reliable estimates, made prior to the current slowdown in demand, the entire worldwide market for wireless and broadband GaAs ICs and devices in 2002 could be supplied by the equivalent of a large silicon fab producing 30 000 to 40 000 6 inch wafers per month. "Consider the number of IDMs around the world" says Day. "Many of these have recently expanded to 6 inch operation. There s a huge amount of GaAs capacity available, and a good portion of it is standing idle right now." Over-estimated markets Unrealistic numbers are also used for the foundry market. "The new companies tend to assume they will capture a huge share of the foundry market," continues Day. "Also, they over-estimate the ratio of business that will go to the IDMs and to the foundries." In the silicon industry, foundries have somewhere around 20% of the total wafer fabrication market. Reaching this level has taken well over 10 years, as well as a huge investment in capital and talented personnel. Design houses When dealing with IDMs, the foundries face the lengthy challenge of getting their process qualified to customers various design specifications. There is also a problem with fabless design houses, which are very significant customers for silicon foundries. "Most of the fabless companies in the GaAs industry are very small, and there are not many of them," says Day. "It can be very difficult for small companies to gain a foothold with big OEMs such as handset manufacturers, which tend to form close working relationships or supply chains with their established IDM suppliers." Furthermore, like the wafer foundries, the design houses need to establish the necessary infrastructure to ship products at high volume. "They have to demonstrate their capabilities before the handset manufacturers will give them a chance," Day adds. One area that might prove successful is the second- and third-tier suppliers in the handset industry, the so-called "handset foundries" such as Arima and Acer that manufacture low-end phones initially. "These companies would need RF expertise from the design houses and could prove to be ideal customers for fabless design houses," says Day. Limited success Perhaps the best hope for the Taiwanese foundries comes from Japan. The major Japanese manufacturers have been slower to expand than their US counterparts; several, notably Fujitsu, have been assessing foundry partners (see Compound Semiconductor August 2000, p31). Also, as described above, Mitsubishi has established a foundry relationship with a Taiwanese company. "Taiwanese foundries will have a better chance of success if Japanese IDMs transfer a large portion of their manufacturing to Taiwan," says Day. Even then, profit margins are likely to be very tight, since the IDMs understand all the cost structures involved. The foundries may not be able to demand a premium for highly advanced processes, since their technologies will probably trail behind those available internally to the IDMs. In contrast, TSMC or UMC are able to offer leading-edge processes that are not available elsewhere. So what does the future hold? "In my opinion, probably only one or two GaAs foundry companies in Taiwan can prosper by working according to the pure foundry business model," says Day. "The successful companies will be the ones with strong management focused on establishing efficient, service-oriented manufacturing infrastructures to realize economies of scale." However, because the private sources of capital behind these companies will be reluctant to abandon their investment, they will almost certainly try new directions and move into other markets, such as optoelectronics, if their GaAs foundry business does not pan out.
×
Search the news archive

To close this popup you can press escape or click the close icon.
×
Logo
×
Register - Step 1

You may choose to subscribe to the Compound Semiconductor Magazine, the Compound Semiconductor Newsletter, or both. You may also request additional information if required, before submitting your application.


Please subscribe me to:

 

You chose the industry type of "Other"

Please enter the industry that you work in:
Please enter the industry that you work in: