SANTA CLARA, Calif. -- At 70, David K. Lam is as excited as ever about the future of the semiconductor industry and what he says is his billion-dollar opportunity to nudge it forward for the second time.
Thirty years ago, Lam helped establish the plasma etcher that became a workhorse tool for cutting fine lines in chips. By fielding a reliable, automated version of the tool, the company that bore his name became the first founded by an Asian American to go public on NASDAQ.
Today he has a new venture he thinks could give the industry a lift at a time when it's getting hard to keep making chips faster, smaller, and cheaper. This time Lam is taking what many have seen as a rival technology -- direct electron-beam lithography -- and bringing it into the conventional process flow.
"In the old days most new semiconductor technologies were disruptive now they have to be complementary -- you have to find ways to use what you have," said Lam.
Chip giants Intel and Taiwan Semiconductor Manufacturing Co. each pitched in a cool billion dollars last year to help finish development of extreme ultraviolet lithography. EUV is widely seen as the successor to today's immersion lithography used to print ever finer lines on chips.
EUV has been delayed many times. It involves a huge set of engineering efforts, perhaps rivaled only by putting a man on the moon. The delays forced engineers time and again to find ways to extend the current immersion lithography to keep alive Moore's Law of producing a generation of smaller, faster cheaper chips every two years or so.
But now immersion is showing signs it is running out of gas and EUV is still not ready. This year for the first time fabs are running wafers twice through the immersion machines to print lines for 20nm chips, adding time and cost to an already complex and expensive process.
If EUV doesn't come in time for the next generation at 14 nm they may have to use four passes, a.k.a. quad patterning, for some critical layers. Intel has said it sees a path to using quad patterning cost effectively even down to 10nm, but others shudder at the thought.
"When you go to quad or more patterns, the cost are astronomical," said Lam. "It's not just the mask costs but cycle times, time-to-market and the complexity of the process that really effects yield -- we lump it all these together in total costs."
Lam shows estimates (above) of lithography rising from 25 percent of the costs of making a chip in 1984 to 70 percent today. "By 2018 if we don't have EUV and have to go to quad or octal patterning, it will drive costs through the roof -- it's unsustainable," Lam said.
Multibeam's e-beam system could eliminate many of the extra steps, Lam claims. The exact number of reduced steps varies based on how a fab uses the technique.
Multibeam got its start as one of a handful of efforts to find an alternative to EUV, using electron beams to write directly on wafers without using masks. But recently, Lam made a course change at Multibeam to parallel rather than collide with the lumbering industry effort.