ASML immersion lithography is winning the 45nm node, and it is winning by doing something that sounds slightly unhinged: flooding the sliver of space between the final lens and the silicon wafer with ultra-pure water. I have watched this company for years, mostly as the quiet Dutch name stamped on equipment nobody outside a fab ever lays eyes on, and in 2006 it has pulled off something I did not see coming. It passed Nikon. A firm from Veldhoven, a town most people could not place on a map, now sets the pace for the entire chip industry.
Money bled out of ASML for most of its first decade. Philips spun it up in 1984 with a tools outfit called ASM International, ran it early on from a leaky prefab next to a company parking lot, and watched it flirt with collapse more than once before the PAS 5500 saved it. That stepper shipped in 1991, first unit out the door to IBM in Fishkill, and gave the Japanese leaders the first real reason to take ASML seriously. The 1995 listing brought in cash, none of which made it a leader. Through the late 1990s the names that mattered in lithography were Nikon and Canon, and ASML was the European challenger that finished third more often than not.
Two wafer stages instead of one. That is the whole TwinScan trick, almost embarrassingly simple once you see it. Most scanners handle a single wafer at a time, align it, expose it, swap it out, start over, with the brutally expensive optics sitting idle through every setup. TwinScan measures and aligns the next wafer on a second stage while the first is still under the lens, so the optics never wait. Nikon could not match that throughput without tearing its machines down to the frame and starting again. ASML brought it to market in 2001, and that is the moment it stopped imitating and started leading, even if the market took a few more years to say so out loud. Wafers per hour is the number a fab lives or dies by, and nothing else on the market moved it like this.
157nm light was supposed to be the next step down from 193nm, and it collapsed into a swamp. Shorter wavelength, fluorine lasers, exotic calcium-fluoride optics, and the whole approach fell apart in the lab: the lens materials cracked, the pellicles meant to shield the mask degraded under the laser, the costs ran past anything a fab would pay. Resolution scales with wavelength over the numerical aperture of the lens, so with the wavelength path dead the only lever left was the aperture. ASML’s answer was water. A film of it in the gap the beam crosses bends 193nm light at a refractive index near 1.44, lifting the effective numerical aperture and buying resolution without touching the laser at all. Burn Lin at TSMC pushed the idea hard when plenty of serious people called it a parlor trick, and ASML committed while its rivals hedged. The first immersion TwinScan machines are shipping now, and they are what drag 45nm into reach. Nikon backed immersion too, only later and slower, and in this business late is just a slower way of being wrong.
The 45nm orders are going to Veldhoven now, not Tokyo, and that sentence would have sounded absurd five years ago. Nikon and Canon owned the leading edge through the 1980s and 1990s without much argument, and they did not get worse; ASML simply got better faster, on two fronts at once, TwinScan throughput and an earlier, harder bet on immersion. The pull underneath that order book is relentless. The Cell processor Sony and IBM built for the PlayStation 3 keeps chewing through the most advanced nodes a fab can offer, and Apple’s freshly announced move onto Intel processors only lengthens the queue. The volume nobody is pricing in yet is mobile, where Qualcomm’s quiet push into application processors is the kind of design that, multiplied across hundreds of millions of handsets, decides which fabs stay full. Canon has already backed out of the bleeding edge to fight elsewhere. Nikon is still in it, but for the first time in a generation it is fighting from behind, and the speed of the reversal has clearly rattled it.
EUV is what comes after immersion, assuming anyone can make it work, and right now nobody can. Light at 13.5nm, more than ten times shorter than today’s 193nm, fixes resolution for years on paper. In practice it is savage to engineer. That light gets swallowed by almost everything, air and ordinary glass included, so the whole machine has to run in a vacuum and bounce off mirrors instead of passing through lenses. The source is the worst of it. Generating enough 13.5nm photons to expose wafers at production speed is something nobody has solved, and the source-power numbers I have seen sit nowhere near what a real fab needs. ASML is shipping its first Alpha Demo Tools to research institutes right now just to prove the concept can work at all. The industry’s optimistic line is that EUV reaches production near the end of the decade, and I would not put my own money on that date. The physics works. Turning it into a machine a fab can run around the clock is the part no one has cracked, and my hunch is that immersion stretched with double-patterning tricks carries the load a good deal longer than the roadmaps want to admit.
ASML in 2006 sits about where Nikon sat twenty years ago, out front and setting the pace, everything riding on a technology that may not arrive on schedule. Immersion already paid for itself. EUV is the open wound, a hole the company keeps pouring money into with no guarantee anything climbs back out, and that will not change for years. I have stopped calling any lead in this business permanent. Whoever owns lithography owns it because they got the last transition right, and the next one is just a story everyone repeats until the machines actually ship. My bet is ASML makes it. Just not on anything close to the timeline Veldhoven is selling today.