The hard part of a deposition tool is never the first good wafer. It is the ten-thousandth identical one. When we moved the plasma-photon MOCVD process from 200 mm to 300 mm wafers, the physics did not change — but the engineering problem inverted. A 300 mm wafer has 2.25× the surface of a 200 mm wafer, and across every square millimeter the plasma density, the UV dose, the precursor flow, and the temperature have to stay flat. At the same time the tolerance for variation shrinks, because a production tool is judged not by its best die but by its worst.
The fight happens at the edge
The center of a wafer is easy; the edge is where uniformity goes to die. Backside contact varies, the plasma sheath bends, heat escapes faster. We rebuilt the wafer handling around a rotary, edge-clamp susceptor to take backside-contact variation out of the equation, split the heater into six independently controlled SCR zones so the rim could be driven as hard as it needed, and re-profiled the ICP coil until the plasma the edge sees is indistinguishable from the plasma the center sees. Each was a separate multi-month program. None of them produced a single new capability — they produced sameness, which is the actual product.
Repeatability you can schedule
Throughput only counts if it is interchangeable. Four chambers run in parallel to reach 60 wafers per hour, and that number is meaningless unless a recipe qualified in chamber one runs identically in chambers two, three, and four. Chamber matching to within metrology noise became its own engineering discipline, backed by in-situ monitoring tight enough to certify every 60-second growth window rather than spot-checking one wafer in twenty.
Two years of the 8-inch demonstrator taught us the recipe. The 12-inch program, which we unveiled at IEDM 2025, was about something less glamorous and far harder: teaching that recipe to behave identically everywhere, every hour, in four places at once. That is the whole difference between a result and a tool.