Every flat optic carries a number that quietly decides its price: the flatness callout. λ/4, λ/10, λ/20 — fractions of a wavelength of light, measured across the clear aperture. Get the callout right and you pay exactly for the performance your system needs. Get it wrong in either direction and you are buying scrap or burning budget. Here is how an inspector reads those numbers — and the one trap that catches even experienced buyers.
The reference-wavelength trap
Flatness is specified in fractions of a reference wavelength, and the industry convention is the HeNe laser line: 632.8 nm. So λ/10 means a peak-to-valley figure error of 63.3 nm — about one seven-hundredth the thickness of a human hair. The trap: a drawing that says "λ/20 @ 10.6 µm" sounds tighter than "λ/20 @ 633 nm," but it is actually 16.75× looser, because the wavelength it references is 16.75× longer. Two vendors quoting "λ/20" can be quoting surfaces that differ by more than an order of magnitude. Always state the reference wavelength on the drawing — and demand it on the cert.
| Flatness Decoded (reference: 632.8 nm HeNe) | |
|---|---|
| λ/4 PV | 158.2 nm — general-purpose precision |
| λ/10 PV | 63.3 nm — precision laser & multi-element systems |
| λ/20 PV | 31.6 nm — reference-grade |
| "λ/20 @ 10.6 µm" | 530 nm — 16.75× looser than λ/20 @ 633 nm. Read the fine print. |
What λ/4 actually buys — and when λ/10 earns its price
The classical Rayleigh/Maréchal criterion links surface figure to image quality: a transmitted wavefront error of λ/4 peak-to-valley (≈ λ/14 RMS) corresponds to a Strehl ratio of 0.8 — the accepted floor for "diffraction-limited" performance. One λ/4 surface, alone, keeps you at that floor. But wavefront errors add up across a train: put three or four "just barely λ/4" surfaces in series and the system budget is gone. That is why λ/10 exists — not as a vanity spec, but as margin. Each λ/10 element consumes only a sliver of the system wavefront budget, leaving room for the elements you cannot control.
Super-polishing: below one angstrom
Figure (flatness) says how flat the surface is; roughness says how smooth it is at microscopic scale — and they are polished, measured, and priced separately. Conventional polishing leaves single-digit-angstrom roughness; super-polishing drives it below 1 Å RMS — a published Edmund Optics example takes a surface from 7.42 Å to 0.45 Å RMS. Why bother? Scatter follows TIS ≈ (4πσcosθ/λ)²: it falls with 1/λ², so a rough surface scatters far less at 10.6 µm than in the visible — but absorption and laser-induced damage threshold (LIDT) still reward the better surface, because contamination and subsurface damage anchor themselves in the texture that polishing leaves behind.
How we hold it: the metrology chain
A spec is only as real as the instrument that checks it. Figure is verified by interferometry against a reference flat; mid-spatial-frequency texture by white-light interferometry; sub-angstrom roughness by atomic force microscopy. On the fabrication side, American Photonics runs P.R. Hoffman double-side lapping and polishing machines — the Servo RS 5400, RS 6600, and RS 7600 class, the largest in the line (P.R. Hoffman Machine Products, Carlisle, Pennsylvania — building precision lapping machines since 1938) — processing both faces of a substrate simultaneously against matched plates — the geometry that makes true parallelism and repeatable flatness a machine property instead of an operator's good day.