Photolithography Oxide Masking 1957 Align Photomask Wafer
# Align the Photomask to the Wafer: The Step Where Geometry Becomes a Registration Problem
## 1. Why Alignment Is Easy Once and Hard Every Time After
This step brings a photomask — a glass plate carrying an opaque pattern — into near-contact over the baked resist and positions it so the pattern lands exactly where the process intends, and for this first masking level on a bare oxidized wafer the task is almost trivial, because there is nothing yet on the wafer to align *to*. That ease will not last; from the second masking level onward, every new pattern must register against features already diffused into the silicon, and the margin for error shrinks to whatever the earlier pattern's own tolerance allows. Jay Lathrop and James Nall at the Diamond Ordnance Fuze Laboratory worked this problem directly, viewing the mask and wafer together through a microscope to position one against the other by eye — and the total positioning error such a system accumulates is best understood as several independent sources added the way independent errors always add:
where each $\sigma$ is the standard deviation contributed by the mechanical stage, the mask's own pattern placement, the viewing optics, and the human operator reading them. Combining in quadrature rather than by simple addition means the largest single contributor dominates the total almost regardless of the others — so alignment effort is best spent entirely on whichever term is currently worst, a lesson this project will not need to relearn but will need to keep applying.
## 2. Real Diagram: What a Few Micrometers of Offset Does to a Junction
Correct alignment places the window entirely within the region the process intended it to open. Misalignment shifts that window — and because the shift is set once, by the mechanical act of positioning the mask, it is identical across every die on the wafer. A single alignment error is therefore not a scattered yield loss; it is a systematic one, striking every device the same way at once.
## 3. Why 1956 Never Needed to Ask Where Anything Was
The 1956 drift-transistor process also used a mask to define a pattern — the mesa boundary at Step 8 of that series — but that mask only ever needed to land *somewhere reasonable* on the wafer, because each mesa was self-contained and nothing in the design required it to line up against a feature that already existed in the silicon. This step, and every masking level from here forward, is different in kind: the pattern must register against structures diffused earlier, and the tolerance for that registration is set by the earlier structure's own dimensions, not by anything optical. 1957 does not simply add photoresist to this project's toolbox; it introduces an entire new class of error — registration error — that no process this project has documented before could even experience, because none of them had two separate patterned operations that needed to agree with each other.
Step 5 does not yet place a single atom of dopant; it decides whether every atom placed in the steps that follow will land where the design, and not merely the mask, intended.