Jon Y challenges the semiconductor industry's most expensive assumption: that Extreme Ultraviolet (EUV) lithography is the inevitable final frontier of chip manufacturing. While the wider media accepts the current trajectory, Y argues that the physics of the current light source—tin droplets hit by lasers—is hitting a hard wall of randomness and power limits that could stall progress at the 3-nanometer node. This is not just a technical deep dive; it is a warning that the $150 million machines dominating the industry may soon be obsolete, replaced by a radically different, massive-scale solution.
The Stochastic Wall
Y begins by dismantling the confidence surrounding TSMC's latest "3-nanometer" process. He notes a peculiar regression: despite having the most EUV machines on Earth, TSMC is actually using fewer EUV exposures for its newest "real" N3 process compared to previous generations. "N3E apparently scaled that back, going from 25 EUV exposures to 19," Y writes, highlighting a counterintuitive move that suggests the technology is struggling rather than thriving.
The core of Y's argument rests on the concept of "stochastic" failures. In the current method, laser-produced plasma (LPP) fires tin droplets to create light, but the resulting photons behave like shotgun pellets—random and unpredictable. "Stochastic print failures are far smaller, completely random and thus non-repeating, and come via a law of nature," Y explains. This randomness creates broken lines and microbridges on the wafer that cannot be fixed by simply cleaning the machine; they are fundamental to the physics of the current approach.
To combat this, the industry has tried to blast the problem with more power. Y points out the inefficiency of this strategy: "It takes 14 times more energy to output an EUV photon than a 193-nanometer photon." While ASML is pushing for 800-watt light sources, Y argues this is insufficient. He cites a 2019 estimate from KIOXIA suggesting that future nodes require 1.5 to 2.8 kilowatts—a gap that current technology cannot bridge without sacrificing production speed. "The fabs won't pay $150 million for a slow machine," Y notes, underscoring the economic impossibility of simply increasing exposure times to fix the physics.
Beyond Tin and Lasers
If the current path is a dead end, Y proposes a radical alternative: the Free Electron Laser (FEL). Unlike traditional lasers that rely on bound electrons within a gas or crystal, an FEL uses a beam of electrons traveling at near light speed, "wiggled" by magnetic fields to generate light. This eliminates the tin contamination that plagues current mirrors and offers a tunable wavelength.
Y describes the mechanics with clarity, contrasting the chaotic nature of current plasma with the controlled environment of an FEL. "If the electrons are similar enough to each other, and fired in the right way ... then something remarkable happens. The electrons will interact with each other's own emitted radiation," he writes. This self-amplification creates a coherent, high-power beam without the debris issues of tin droplets.
The evidence for this shift is already emerging. Y highlights a proof-of-concept from Japan's KEK (High Energy Accelerator Research Organization), which successfully generated EUV light using a linear accelerator. The efficiency gains are staggering. "The KEK machine uses about 7 megawatts of electricity to generate 10 kilowatts of EUV power," Y notes, compared to the current method's massive energy drain. This efficiency could be a game-changer for fabs, where electricity is a primary variable cost.
Dumping the tin laser approach might cause some disappointment to tech enthusiasts. But the thought of TSMC putting a $400 million, 200-meter long linear accelerator underneath their next fab is kind of awesome too.
Critics might note that the sheer scale of an FEL—requiring a 200-meter linear accelerator and a $400 million price tag—makes it a difficult pill to swallow for existing infrastructure. Integrating such a massive machine into a cleanroom environment presents logistical hurdles that Y acknowledges but treats as surmountable engineering challenges rather than dealbreakers.
The Economics of Disruption
Y frames the economic argument around the concept of shared infrastructure. A single FEL could theoretically power multiple lithography machines, distributing its high fixed costs across the fab. "A laser can generate 10 kiloWatts of EUV power for one machine, or 1 kilowatts for ten machines," Y writes, suggesting a shift from buying individual expensive tools to subscribing to a centralized light source.
This reframing challenges the entire business model of equipment suppliers like ASML, whose current strategy relies on selling more units of increasingly powerful LPP systems. Y suggests that the industry is approaching a tipping point where the "Promised Land" of EUV has been delayed by the limitations of the tin-laser method. "EUV was supposed to take us to the Promised Land. It hasn't yet because the amazing, double-tin-shot-with-a-laser EUV light source everyone loves to talk about is not powerful enough," he asserts.
The article concludes by noting that research into FELs is active in Europe and the United States, not just Japan. Y speculates that ASML has likely evaluated this path but remains silent. "Someone should get the semiconductor Avengers together to talk about it," he urges, calling for a collaborative effort to solve the power and stochastic crisis before it stalls the entire industry.
Bottom Line
Jon Y's strongest move is connecting the abstract physics of stochastic failures to the concrete business reality of TSMC's production scaling issues, proving that the current EUV roadmap is hitting a wall. The argument's biggest vulnerability lies in the massive capital expenditure required to retrofit fabs with linear accelerators, a hurdle that may delay adoption despite the technical superiority. Readers should watch for any shift in ASML's R&D spending or new partnerships with accelerator labs, as these would signal the industry's acceptance that the tin-laser era is ending.