Over the past decade the biggest investor in chip manufacturing has been Samsung Electronics. With a burgeoning flash memory market driven by solid-state drives and tablets, coupled with the push below 10nm for advanced microprocessors, the Korean manufacturer has ploughed money into high-end fab capacity so it can keep up. And, over the past couple of years, the semiconductor business has been on a tear.
With such high-volume markets to serve, you might expect chipmakers like Samsung to be lobbying for a shift to larger wafers. In the wake of the financial crisis, the company was, along with Intel, keen to convince its equipment suppliers to support an increase in wafer sizes from 300mm to 450mm so that it could cut productions. Although making equipment to support wafers as big as a family-sized pizza would be complex and expensive, the doubling in chips per wafer promised increased gross margins for manufacturers serving high-volume markets.
However, Samsung’s enthusiasm waned as it became clear that wafer size is not the path to lower costs that it used to be. Steps such as etching and vapour-phase deposition reap the benefits of larger wafers because they act on the entire surface at once. But lithography has become a much bigger source of cost and that remains fixed to the reticle size – something that has not increased in terms of area in decades. As a result, bigger wafers take longer to process. Consequently, chipmakers seem to be locked into 300mm for some time to come.
Not a wasted investment
“The [R&D] investment in 450mm hasn’t been wasted,” says Malcolm Penn, president of market-analyst firm Future Horizons. “Spin-offs have gone into 300mm and made those 300mm factories much more efficient than they would have been under the natural roadmap.”
But the dominance of 300mm has come under unexpected pressure from its immediate predecessor, the 200mm wafer. First put into production 20 years ago, lines designed for 300mm wafers accounted for more than half of total installed fab capacity by the start of this decade.
As chipmakers shuttered their old 150mm and 200mm plants, partly driven by the long-term trend of transferring production to foundries, that share rose. It ticked up to just over two-thirds by the end of 2017. Although market analyst firm IC Insights expects continued capacity additions by the largest suppliers will push the share to more than 70% by the end of 2019, the old 200mm lines are fighting back in unexpected ways.
Since the early 2000s processes have been optimised for 300mm lines. As a result, the smallest geometry that was developed with 200mm wafers in mind was 180nm. But at the start of this decade, suppliers such as NXP and TSMC started to look at ways they could extend the life of their existing fabs and take the 200mm lines at SSMC in Singapore down to 110nm. Quietly, even Samsung has done the same, with its Line 6 fab in Giheung, Korea starting production of 65nm-geometry chips earlier this year.
Some of the resurgence in 200mm is based on the reuse of equipment from the twenty-odd fabs that have shut since 2009.
According to Penn, supplies are soon snapped up. “You can buy secondhand pieces of kit for 200mm, that’s workable. But those are now rarer than hen’s teeth,” he says. At the same time, the equipment suppliers are seeing their sales of 200mm machinery tick up. Samsung has, for example, used extensive automation technology, which was part of the 300mm revolution, to cut costs at Line 6.
During the company’s second-quarter conference call held in mid-May, Applied Materials CFO, said: “Our 200mm systems revenue should be up over 20% this year. In fact, we are now building brand-new systems to keep with the demand for the billions of sensors and other low-cost devices needed in the Internet of Things, advanced automotive applications, and Industry 4.0.”
Applied Materials president and CEO Gary Dickerson said capacity expansion at ‘trailing edge’ nodes, which are now classed as geometries of 28nm and coarser, has been something the company has been seeing for about a decade. A decade ago, 10% of the company’s business was serving those nodes. The vast majority went into the leading-edge processes. “Over time, that’s evolved to 80/20, 60/40 and this year we see it evolving to 50/50.”
| “The R&D investment in 450nm hasn’t been wasted. Spin-offs have gone into 300nm and made 300nm factories more efficient.” - Malcom Penn |
A lot of that supply is for 28nm on 300mm which has steadily increased and is demonstrating the interest among customers for SoCs that integrate sensors, memory and other circuitry that is expensive to provide on the finer-geometry finFET-based technologies. “Initially, we called peak capacity at 28nm [to be] 330,000 wafer starts a month. Then it was 400,000, then it was 450,000, and now it’s 500,000 wafer starts a month.”
Cost-volume trade off
One of the reasons why manufacturers have opted for 200mm production rather than simply expand 300mm lines is one of cost-volume trade off. For devices that have small die sizes and do not ship in the millions, it is difficult to manage the smaller batches of wafers that would be needed for 300mm production.
In a number of cases, the push for 200mm is coming from novel technologies such as gallium nitride on silicon. Suppliers such as EpiGaN and Kyma Technologies have claimed success in developing equipment and techniques to deposit the III-V semiconductor – useful for both power transistors and LEDs – reliably onto 200mm wafers.
Where chipmakers see high-volume markets, some are keen to push the cost advantage of bigger wafers even for speciality process technologies, such as those needed for mixed-signal and power devices.
Texas Instruments (TI) was among the first companies to manufacture mixed-chip chips on 300mm wafers with the conversion of its DMOS6 fab in its home state. The company has argued that manufacturing some of its devices on 300mm wafers gives it a 40% cost advantage before the cost of packaging is taken into account. According to IC Insights, in 2017 about half of TI’s mixed-signal revenue came from chip built on 300mm wafers.
Last month, Infineon Technologies said it would pump €1.6bn into a 300mm fab in Villach, Austria, that would focus on power devices.
CEO Reinhard Ploss said the expected demand for power semiconductors “is underpinned by global megatrends such as climate change, demographic change and increasing digitisation”. He expects the devices to go into electric vehicles, renewable-power generators and data centres.
The widespread use of sensors based on microelectromechanical systems (MEMS), which are often made on wafers measuring just 150mm or smaller, has similarly convinced Bosch that it should move production to 300mm.
Penn says the margin improvements Bosch should see when its 300mm Dresden fab moves into production in the early 2020s will hurt competitors such as STMicroelectronics, which still uses 150mm wafers for some high-volume sensors.
But in an environment where factors other than size often contribute heavily to margins, the shift to 300mm is not necessarily the right answer.
