Two startups betting they can crack EUV cost
xLight wants to replace ASML's tin-droplet plasma with a free-electron laser and sell photons by the gallon. Substrate wants to skip EUV entirely and revive X-ray lithography. If either ships, who can build a fab changes overnight.
EUV cost per tool keeps rising and High-NA paid the half-field tax. The cost ceiling is the binding constraint on how many fabs can ever exist. Two startups are attacking it from first principles.
The cost problem from first principles
The conventional wisdom is that ASML is uncatchable. The tin-droplet light source took two decades to engineer; the Mo/Si mirrors come from one supplier (Carl Zeiss); the scanner has thousands of co-evolved components. A new competitor would have to invent all of it, and Zeiss is not going to sell its mirrors to a startup that competes with ASML.
Two startups are arguing that the right move is to stop competing on ASML's stack and rebuild the physics: change what generates the light, and let the rest of the stack follow.
xLight: free-electron laser, photons as a service
xLight (California, with Pat Gelsinger on the board) wants to replace the tin-droplet plasma with a free-electron laser (FEL): a particle accelerator that produces coherent EUV at much higher total power. One FEL could feed many ASML scanners (xLight keeps the scanner; it swaps only the light source), with the option to scale below 13.5 nm later.
The business model is the punchline: photons as a service. xLight builds the FEL on the fab's premises and bills per photon consumed, the way the fab already pays for gas, water, and power. The fab swaps capex for opex and gets higher throughput and better yield from the higher dose. xLight takes the engineering risk and the capital risk; the fab gets a utility bill.
Source: Chipstrat (chipstrat.com); xLight company materials
Substrate: revive X-ray, skip EUV entirely
Substrate (San Francisco) is throwing out the EUV playbook and reviving X-ray lithography. IBM and others built working X-ray prototypes in the 1980s and 90s; tabletop synchrotrons have existed for decades. The hard part is the optics: X-rays go through everything, so the trick is proximity printing, where the mask is held at native pitch right against the wafer with no reduction. Mask-making becomes harder, but the rest of the stack potentially becomes much cheaper.
If it works, an X-ray scanner could cost a fraction of an EUV one. That changes who can afford to build a fab.
Source: Chipstrat (chipstrat.com); Substrate company materials
The strategic read
If either bet ships, the geography of "who can build a fab" flips. Global Foundries, Texas Instruments, and even fabless designers could plausibly go vertical again on a fraction of the capex. Jerry Sanders said real men have fabs; for two decades only TSMC, Samsung, and Intel could afford the title. The next photon source could change who else gets to wear it.
Both bets are early. Neither has shipped a tool that prints a wafer at a leading node. But the question worth tracking is no longer 'when does ASML ship the next scanner'; it is 'does anyone make ASML's economics look obsolete'.