Nano, when a wavelength is huge

You are trying to print a stripe so thin that ordinary light is clumsy around it, and the stripe may be meant to carry light itself. The picture at the top is that regime under an electron microscope: lines packed on a surface, some alone, some in a crowd. On a screen each of them was a confident rectangle. In the lab, the wavelength you care about may be larger than the stripe, and neighbors donate energy while the pattern is written.

The drawing is smaller than the wave

Visible green light is about 0.55 µm. A line 0.05 µm wide is a tenth of that wavelength. Telecom light at 1550 nm — 1.55 µm — is about thirty times a 50 nm gap. The wave does not “see” your rectangle the way a voltmeter sees a copper trace. An isolated line can vanish. A gap can close. A neighbor can print wider because extra dose leaked over. The failure is a waveguide that does not guide, a slot that is fused, or a resonator at the wrong color. The file can still be clean.

The second picture is the comparison the first one implies up close: an isolated line and a dense bundle that were drawn the same width and did not print the same width. That is proximity effect whether the writer was light through a mask or a scanning electron beam. Dense regions bloom. Isolated lines starve. Correction is a bias or a dose map, agreed with the person who owns the process, not a thicker stroke in the editor.

What you still owe the file

You still need a clean integer grid and no slivers, because the writer will try to honor vertices. Circles become many-sided polygons. Gaps close if the bias is larger than the gap you drew. A waveguide here is often a strip whose width sets the mode. A few tens of nanometers of width error shifts the color the device likes.

Do not treat these numbers like board traces. A 50 nm gap is not “0.050” in a unit you did not name. Say nanometers or micrometers in the same sentence. And do not believe the microscope photograph is the file. It is the file after wavelength, scatter, and process bias took a cut. Talk to the process owner about how much to grow or shrink the drawing so the print matches the intent. Nano layout is still layout. It is layout with a clumsy printer.

Keep the vertices honest before that measurement. A curve faceted every 5 nm is a different write from a curve faceted every 50 nm, and both can look smooth when you are zoomed out. The microscope will show the facets if they are large beside the wavelength, and the beam writer will charge you for them if they are many. Choose the facet with the person who owns the tool, then freeze it.

On the process owner’s micrograph, the isolated line beside the bundle in the second picture, measure the printed width of that lonely line and the printed width of one line in the middle of the bundle. The drawing gave them the same width. Write both printed numbers beside that drawn width. Measure each line in three places along its length and average them, because the edge in the micrograph is rough and a single cursor click will flatter you. If the bundle line is wider, neighbors donated dose, and one bias for the whole chip will not repair both.

Ask, before you believe the averages, what bias the tool owner expects. If that bias is 10 nm and the smallest gap in the file is 20 nm, the gap may close. Draw the gap wider, or accept a different process, and only then freeze the two printed columns. A file that has the drawn column alone is not yet a device. The isolated line and the bundled line are the two rows that tell you whether the correction was real. Keep the three raw widths, not only the average. One narrow spot on the isolated line can open a device even when the average looks fine. Write the expected bias in that same note, next to the smallest gap in the file, and read the pair before you freeze the width. A bias that is half the gap is a closed gap waiting on the wafer, and you can see it in the note without another pass over the micrograph. Circle the narrower printed width, usually the isolated line, and use that number when you decide whether the gap survives. A bundle line that printed wide will flatter you if it is the only width you keep.

1550 nm, and a gap of 20 nm

The wave you are drawing for is 1550 nm, which is 1.55 µm. A feature of 0.05 µm is a small fraction of that wave, so the drawing is not a picture of the light. On the process owner’s micrograph, the isolated line and one line in the middle of the bundle were given the same drawn width. Write both printed widths beside it. If the tool owner’s bias is 10 nm and the smallest gap in the file is 20 nm, that bias is half the gap. The gap can close on the wafer while the file still shows daylight.

Facet the curve at 5 nm or at 50 nm only after the person who owns the tool picks one, and freeze that choice in the note next to the bias and the 20 nm gap. A file that keeps the drawn width and drops the two printed widths is not yet a device. Circle the narrower printed width, usually the isolated line, and decide from that number whether the gap survives.