skip to main content

The Last Word

David Lammers


The Advanced Semiconductor Manufacturing Conference (ASMC) celebrated its 25th anniversary in Saratoga Springs, New York, in late May, with a record 308 participants enjoying the spectacular spring weather while reflecting on the considerable challenges facing the semiconductor industry.

Mukesh Khare, director of research at IBM’s semiconductor division, kicked off ASMC with a keynote that provided plenty of optimism on the technology front. After silicon FinFETs come silicon germanium FinFETs, nanowires in a gate-all-around architecture, and then carbon nanotubes (CNTs) with their spectacular mobilities. IBM has long been bullish on CNTs. (Gartner analyst Dean Freeman is more skeptical, saying “using CNTs is like trying to line up worms on a wafer.”)

“The technology roadmap is very rich, and can continue for a good period of time,” Khare said. IBM believes its 10nm technology will involve fifth generation ArF lithography with sidewall image transfer technology. “It took the industry 15 years to develop high-k,” he added. “EUV lithography is another technology that will take a lot of resources, but it is coming together.”

Charlie Pappis, vice-president and general manager of Applied Materials’ Global Services business, said while traditional patterning shrinks have become more difficult, “the industry has changed the innovation equation to new materials and the third dimension [vertical, 3D structures] out of necessity, as we get to sub-ten nanometers.”

Dean Freeman,
Industry Analyst at Gartner

Mukesh Khare,
director of research, IBM semiconductor division

Robert Maire,
president of Semiconductor
Advisors, LLC.

Economic Challenges Loom

Until the 28nm node, the chip industry has consistently been able to “double the number of transistors for the same number of dollars,” said Robert Maire, president of Semiconductor Advisors, LLC. Maire summed up the core cost problem nicely, noting that as double patterning became necessary at the 28nm node, “the industry fell off the [cost-reduction] wagon. We need to fix or replace EUV. Wouldn’t it be nice to have direct-write lithography, so we wouldn’t need a $5 million mask set?”

Since 1965, Moore’s Law has been translated into economic terms, as companies were able to reduce the cost-per-function by about 25–30% per year. Getting back on that cost-reduction curve is the key problem as scaling continues to move ahead. To be sure, we can improve the efficiency of design by using leading-edge transistors more creatively. Older fabs can also be upgraded and improved, partly by networking tools and analyzing the big data streaming from 200mm and 300mm production tools.

It used to be harder for semiconductor-related companies to raise money due to the cyclicality of the industry. Pappis said that DRAM cycles often exacerbated the industry’s ups and downs, noting that even the traditional seasonality swings have modulated. Instead of manufacturing volumes rising solely to meet consumer demand during the Christmas season, the Chinese New Year is exerting its own pull. “Our focus on Western ways is going away, as demand rises in other parts of the world, especially in Asia,” Pappis said.

The attention paid to consumer trends carried over into a closing presentation by Gartner’s Dean Freeman. “The consumer is king, so if they get scared things could change. But we see a pretty nice year shaping up,” he said. Capital expenditures should increase by 5.6% on average in 2013–2018, with an increasing share going to packaging.

And as low-cost devices aimed at Internet of Things (IoT) applications begin to reach higher volumes, it will put new life into 200mm fabs. By 2020, about seventeen 200mm fabs, each capable of 50,000 wafer starts per month, will be required to produce the devices needed for the estimated 10 billion “things” going into IoT applications. “The IoT should help the underutilized fabs around the world. Right now, we count 20 shuttered 200mm fabs, and some of them could be put back into service if our scenario of a reasonable—but not spectacular—growth rate for the IoT occurs,” Freeman said.

I could barely count all the new fab lines being built now by TSMC. In his ASMC keynote, John Lin, general manager of operations at the G450C center in Albany and a general manager of operations at TSMC, described a flurry of investments. These fabs will be “nearly people-less,” Lin said, with 80–90% of the operators working in glassed-in control rooms, monitoring the production lines. A TSMC fab can have about 1,000 different products underway at any one time. The company builds about 4,000 new masks a month. To keep improving, the foundry is building what Lin called a “big data neural system.”

“We want to feed [data] forward and feed backward in an integrated manufacturing system, including design and packaging,” he said. GLOBALFOUNDRIES has similar goals. Dave Gross, in charge of automation technology, said networks between the fabs and suppliers are expanding, even as security concerns continue. “It used to be that not one bit of data could go outside the fab. Not one bit. Now, we stream data constantly.”

David Lammers
is an Austin-based technology journalist.