In its early days, ASML was convinced that optical litho would only last for another decade. It seriously considered investing in next-generation lithography and even sold direct-write systems.
This week at the online SPIE Advanced Lithography symposium, ASML’s Jos Benschop made it clear that optics will continue to advance chip manufacturing at least until 2030. His developers are now working on high-NA EUV and he even talked about hyper-NA. Nine more years of cost reduction in IC production, that will add up to a breathtaking half-century of exponential growth – in hardware, not in software.
It’s also a fifty-year adventure in optics. What’s currently known as EUV was called soft X-ray in the mid-1990s, but the semicon industry labeled it “extreme ultraviolet” because that sounded more acceptable. Still, you could manipulate it with optics, although the multilayer mirrors for masks and the light path made it extremely complicated.
Half a century of optics – who would have thought? Certainly not the people who worked on ASML’s business plan in the spring of 1984. They were seriously considering a path to phase out diffractive optics – EUV wasn’t on their roadmap yet. It seemed clear that litho with lenses and light would reach its peak around 1990. All in full agreement with the assumptions of the time. Everyone seemed to concur that the wavelengths of the available light sources were the limiting factor in chip imaging.
One of ASML’s top-three objectives in 1984 was to mature into a recognized supplier of “other lithographic equipment” for volume production by the end of the eighties. The statement clearly shows how uncertain the industry still was at the time regarding optical methods for manufacturing chips. That’s why the company’s ambitions for developing electron-beam optics and possibly X-ray lithography were its number-two priority – below the desired top-three market position, but above turning a profit and paying off all debt.
At that time, most industry bets for next-gen litho were on e-beam. In 1984, ASML even sold direct e-beam writers. It didn’t own these Electron Beam Pattern Generators. The technology was being developed at Philips’ research lab in Redhill and manufactured at Philips S&I in Eindhoven. A few units were already being sold by Philips and ASML could make some money by selling the instruments as well. The expectation was that e-beam lithography would occupy a central role in chip fabrication and might be a future necessity for the young litho company.
“There are signs of increasing interest in e-beam technologies for wafer processing equipment, which makes them crucial for continuity in the late eighties,” ASML’s general manager Gjalt Smit wrote in his business plan. Demonstrating that it was completely unclear in 1984 that optical technology would have such staying power. The plan predicted that imaging was nearing the end of the line.
In an interview I had in 2012 with former Philips Research manager Marino Carasso, he told me about the struggles. In the eighties, he had hefty discussions with the developers of electron beam lithography in Redhill. “At that moment, I thought that printing details on chips under 200 nanometers would become extremely problematic. At least, I didn’t dare to guarantee that to the chip production guys,” said Carasso. “We did see that you could write small things very nicely with an electron beam like that, but it took forever to produce an image. The electron beam developers couldn’t explain to us how to speed up the process. To match the speed of a stepper, they had to feed their machine with a lot of information. Invariably, my argument was: you guys can invent the world, but you can never transmit information at terabytes per second.”
The economics of optics
Philips Research at the time was heavily involved in developing lithography for future chips. It supported ASML but developed its own steppers in parallel. These machines, called Silicon Repeaters, were intended for the Mega project, the catch-up race Philips and Siemens had started to get on par with the Japanese in memory chips. The research steppers were a safety belt. In case ASML and other commercial lithography providers couldn’t deliver, Philips planned to manufacture its own litho machines.
Carasso discussed the challenges with top-level executives, including Cees Krijgsman, director of Philips Elcoma and head of the Mega project. During one of the research meetings with Elcoma, Krijgsman said to Carasso in an informal moment: “Marino, be frank, how far does your Silicon Repeater go and when should I switch to another technology? Half a micron?” Carasso replied, hesitatingly: “Four-tenths of a micron, three-tenths? Very possibly two-tenths, but below that, I’m starting to have doubts.”
However, around the time ASML was founded, the first cracks appeared in the e-beam promises. In early 1984, General Signal decided to halt its development, just after Varian had announced that it was unable to deliver the first shipment of its direct-write equipment. But the latter company was still in an upbeat mood. “We continue to feel that for short-run VLSI and gate arrays, e-beam will be viable,” Bruce Doyle, general manager of Varian’s Lithography Products division, said publicly at the beginning of 1984. However, he didn’t want to write off optics: “There’s plenty of room for multiple lithography processes. And steppers, X-ray and e-beam will wind up complementing, not competing with each other.”
But Charles Biechler, deputy general manager of Perkin-Elmer’s Electron-Beam Technology division kept boasting: “Given a 1-micron minimum line width, and a not overly complex chip … we can offer a throughput of 30 4-inch wafers an hour.” With some exaggeration, he added that such a rate was comparable to what wafer steppers could do (in fact, ASML’s PAS 2000 did 60 wafers per hour at that time). “And when you talk about future 0.5-micron chips,” he maintained, “there’s nothing that can handle them like a direct-write system.”
Officials at both Perkin-Elmer and Varian attributed the delay in shipments primarily to the basic problem of designing machines for rugged production environments. “The challenge is in automating, as well as increasing throughputs,” noted a Varian spokesman in 1984. “An R&D direct-write machine from Cambridge Instruments or Philips might have 103 knobs for engineers to play with. Eliminating the knobs is a tremendously difficult job.”
But eventually, like Carasso correctly predicted, the inherently slow direct-write electron beams couldn’t match the economics of optics. When massive compute power became available, Mapper courageously attempted to provide semicon manufacturers with an alternative to EUV, but it sadly stumbled, after which most of its engineers joined ASML to help the giant with its e-beam metrology systems.
In the 1980s, many engineers considered the Japanese company JEOL, which also produced electron microscopes, as the best provider of direct-write equipment. JEOL is still active in e-beam for semicon, but direct write is still a niche market.