Technical Insight
The pure-play GaAs foundry business: is it a viable model?
With a glut of GaAs capacity still in place, how can pure-play GaAs foundries survive and be profitable? Not easily, in the view of some senior industry figures. Michael Hatcher catches up with WIN Semiconductor and reports on what some see as a flawed business model.
The GaAs foundry model is, like its silicon equivalent, driven by volume. The key difference is that, while companies such as TSMC and UMC can exploit the huge volumes of the silicon industry, the volumes in GaAs are nothing like as significant.
Rob Christ is the marketing director at TriQuint, the US-based manufacturer of both RF and optoelectronic devices that also has a significant GaAs foundry business. His view is that low volumes make the pure-play GaAs foundry option a flawed one. "The [silicon] model does not apply to compound semiconductors," Christ said.
Christ estimates that last year, TriQuint had nearly $60 million in revenue from its foundry business (with over 60% of the GaAs foundry market, TriQuint offers 18 different GaAs foundry production processes), equivalent to nearly 20% of the company s total revenue in 2003. He says that this proportion has stayed consistent for the past few years, with the size of the foundry business following the boom and bust of the overall industry, albeit with substantial changes in the product mix. At the moment, the share of military-related foundry business at TriQuint is roughly 20% of the total, estimates Christ.
While military applications provide a reliable revenue base - what Christ calls "ballast" - the proportion of foundry business from the commercial sector at TriQuint is now growing, thanks to the success of cell phones, wireless local-area network (WLAN) protocols and the emerging "WiMax" broadband version of this technology.
Despite the fact that TriQuint s foundry revenue is growing (in spite of competition from the Taiwanese foundries), and that the company recently extended its foundry offering with a next-generation GaAs HBT process, Christ says that the business would not survive as a stand-alone entity, and he is puzzled at the Taiwanese pure-play model.
The reason is volume: Christ estimates that the current total market for 6 inch equivalent GaAs wafers is 130,000-150,000 wafers per year. According to Christ s figures, 80% of this volume is controlled by what he calls the "big five" GaAs manufacturing companies: RF Micro Devices, Fujitsu-Sumitomo, NEC, Skyworks and TriQuint. Only the remaining 20%, equivalent to about 25,000 6 inch wafers, is available to other GaAs manufacturers.
"The TSMC business model works in the silicon business because there is enough required capacity for one or two big players to operate as pure-play foundries," said Christ. "The same cannot be said of GaAs. There simply isn t the volume." Christ reckons that the three biggest Taiwanese foundries (GCSC, WIN Semiconductor and Suntek) have to sell wafers for about $2000-$2500. This means that the total addressable market of, at most, around 25,000 6 inch wafer equivalents translates to a total possible revenue of $50-60 million per year for the Taiwanese foundries to share with the other smaller players in GaAs manufacturing.
Christ s belief is that this figure is substantial enough for one big player to be profitable, but no more. Therefore, he is expecting further consolidation between the Taiwanese foundry companies.
Christ s estimate is that there were just under 50 GaAs fabs in the world at about this time last year, equivalent to a total capacity of 700,000-800,000 6 inch wafers. Current production levels are still well below the total capacity that is available. Inevitably there will have been further reduction in capacity over the past year, but Christ says there is still a need for further shut-downs.
While relatively successful companies like RFMD may be operating at about 70% capacity, many others are operating at more like 20%. Christ himself admits that some of TriQuint s fab capacity is currently mothballed. "The excess capacity might be used up to some degree by WLAN and WiMax applications," speculated Christ.
Although TriQuint s foundry business is not viable on its own, it is useful both as an extra revenue stream and strategically, as it allows the company to nurture close relationships with its customers. Christ s belief is that the only way for GaAs foundries to survive in the long term - apart from consolidation - is for them to add services or product lines. There are some reports of this already happening, with Suntek believed to be relying on LED chip manufacture for half of its business. WIN Semiconductor has muscular financial backing from US-based Fairchild Semiconductor, which bought an equity stake in the Taiwanese operation when it acquired the commercial unit of Raytheon s RF activity.
Not surprisingly, Peter Cheang, WIN Semiconductor s marketing director, doesn t agree with Christ s prognosis. "WIN firmly believes that the pure-play foundry model is viable," Cheang told Compound Semiconductor. "WIN has witnessed more and more fabless design houses moving from the engineering stage to the production stage."
Cheang also hinted that more business would be coming WIN s way in the near future: "[Our] success with the Raytheon RF component group (now Fairchild) was the first sign that existing major GaAs manufacturers would release production needs to a foundry house. More arrangements like this will be disclosed in the near future."
WIN s current set-up is designed to serve the market s needs for high-volume, high-quality 6 inch GaAs wafer manufacture, says Cheang. The main processes that WIN is involved in are HBT manufacture for power amplifiers and HEMTs for switching and high-frequency applications. Technologies available at the Taiwan foundry include 1, 2 and 3 µm HBTs, and 0.15 µm PHEMTs and MHEMTs, among others.
So does WIN plan to diversify its product offering, as Christ believes is necessary? Cheang does not rule this out: "WIN definitely believes that it can survive and become profitable by staying as a pure-play foundry supplier," he said. "We will consider offering other services that align well with our business model and market needs."
Cheang says that WIN s foundry is currently operating at about 60% of its manufacturing capacity. Emerging applications in the WLAN market have helped to drive revenue upwards recently: "The surge in the WLAN market has enabled independent fabless design houses to gain market acceptance easier and faster than the handset market. Consequently, WIN has seen a [more] substantial growth in WLAN orders compared with any other applications in the past year."
This translated to a large increase in yearly revenue at WIN from 2002 to 2003. Cheang said that the 2002 figure of NT$251 million ($7.6 million) jumped to NT$500 million last year. "WIN has seen revenue growth more than double each year since we started to offer foundry services," he noted. While Cheang won t reveal the company s break-even revenue, he said that it expects to reach the monthly cash flow break-even point this year.
Christ s view is that the fabless design houses will increasingly tend to merge with companies such as WIN and TriQuint that have available manufacturing capacity. Whether further consolidation is required to see foundries like WIN move into profitability remains to be seen.
Rob Christ is the marketing director at TriQuint, the US-based manufacturer of both RF and optoelectronic devices that also has a significant GaAs foundry business. His view is that low volumes make the pure-play GaAs foundry option a flawed one. "The [silicon] model does not apply to compound semiconductors," Christ said.
Christ estimates that last year, TriQuint had nearly $60 million in revenue from its foundry business (with over 60% of the GaAs foundry market, TriQuint offers 18 different GaAs foundry production processes), equivalent to nearly 20% of the company s total revenue in 2003. He says that this proportion has stayed consistent for the past few years, with the size of the foundry business following the boom and bust of the overall industry, albeit with substantial changes in the product mix. At the moment, the share of military-related foundry business at TriQuint is roughly 20% of the total, estimates Christ.
While military applications provide a reliable revenue base - what Christ calls "ballast" - the proportion of foundry business from the commercial sector at TriQuint is now growing, thanks to the success of cell phones, wireless local-area network (WLAN) protocols and the emerging "WiMax" broadband version of this technology.
Despite the fact that TriQuint s foundry revenue is growing (in spite of competition from the Taiwanese foundries), and that the company recently extended its foundry offering with a next-generation GaAs HBT process, Christ says that the business would not survive as a stand-alone entity, and he is puzzled at the Taiwanese pure-play model.
The reason is volume: Christ estimates that the current total market for 6 inch equivalent GaAs wafers is 130,000-150,000 wafers per year. According to Christ s figures, 80% of this volume is controlled by what he calls the "big five" GaAs manufacturing companies: RF Micro Devices, Fujitsu-Sumitomo, NEC, Skyworks and TriQuint. Only the remaining 20%, equivalent to about 25,000 6 inch wafers, is available to other GaAs manufacturers.
"The TSMC business model works in the silicon business because there is enough required capacity for one or two big players to operate as pure-play foundries," said Christ. "The same cannot be said of GaAs. There simply isn t the volume." Christ reckons that the three biggest Taiwanese foundries (GCSC, WIN Semiconductor and Suntek) have to sell wafers for about $2000-$2500. This means that the total addressable market of, at most, around 25,000 6 inch wafer equivalents translates to a total possible revenue of $50-60 million per year for the Taiwanese foundries to share with the other smaller players in GaAs manufacturing.
Christ s belief is that this figure is substantial enough for one big player to be profitable, but no more. Therefore, he is expecting further consolidation between the Taiwanese foundry companies.
Christ s estimate is that there were just under 50 GaAs fabs in the world at about this time last year, equivalent to a total capacity of 700,000-800,000 6 inch wafers. Current production levels are still well below the total capacity that is available. Inevitably there will have been further reduction in capacity over the past year, but Christ says there is still a need for further shut-downs.
While relatively successful companies like RFMD may be operating at about 70% capacity, many others are operating at more like 20%. Christ himself admits that some of TriQuint s fab capacity is currently mothballed. "The excess capacity might be used up to some degree by WLAN and WiMax applications," speculated Christ.
Although TriQuint s foundry business is not viable on its own, it is useful both as an extra revenue stream and strategically, as it allows the company to nurture close relationships with its customers. Christ s belief is that the only way for GaAs foundries to survive in the long term - apart from consolidation - is for them to add services or product lines. There are some reports of this already happening, with Suntek believed to be relying on LED chip manufacture for half of its business. WIN Semiconductor has muscular financial backing from US-based Fairchild Semiconductor, which bought an equity stake in the Taiwanese operation when it acquired the commercial unit of Raytheon s RF activity.
Not surprisingly, Peter Cheang, WIN Semiconductor s marketing director, doesn t agree with Christ s prognosis. "WIN firmly believes that the pure-play foundry model is viable," Cheang told Compound Semiconductor. "WIN has witnessed more and more fabless design houses moving from the engineering stage to the production stage."
Cheang also hinted that more business would be coming WIN s way in the near future: "[Our] success with the Raytheon RF component group (now Fairchild) was the first sign that existing major GaAs manufacturers would release production needs to a foundry house. More arrangements like this will be disclosed in the near future."
WIN s current set-up is designed to serve the market s needs for high-volume, high-quality 6 inch GaAs wafer manufacture, says Cheang. The main processes that WIN is involved in are HBT manufacture for power amplifiers and HEMTs for switching and high-frequency applications. Technologies available at the Taiwan foundry include 1, 2 and 3 µm HBTs, and 0.15 µm PHEMTs and MHEMTs, among others.
So does WIN plan to diversify its product offering, as Christ believes is necessary? Cheang does not rule this out: "WIN definitely believes that it can survive and become profitable by staying as a pure-play foundry supplier," he said. "We will consider offering other services that align well with our business model and market needs."
Cheang says that WIN s foundry is currently operating at about 60% of its manufacturing capacity. Emerging applications in the WLAN market have helped to drive revenue upwards recently: "The surge in the WLAN market has enabled independent fabless design houses to gain market acceptance easier and faster than the handset market. Consequently, WIN has seen a [more] substantial growth in WLAN orders compared with any other applications in the past year."
This translated to a large increase in yearly revenue at WIN from 2002 to 2003. Cheang said that the 2002 figure of NT$251 million ($7.6 million) jumped to NT$500 million last year. "WIN has seen revenue growth more than double each year since we started to offer foundry services," he noted. While Cheang won t reveal the company s break-even revenue, he said that it expects to reach the monthly cash flow break-even point this year.
Christ s view is that the fabless design houses will increasingly tend to merge with companies such as WIN and TriQuint that have available manufacturing capacity. Whether further consolidation is required to see foundries like WIN move into profitability remains to be seen.
