What is critical dimension (CD), and where is a chip's line width decided?

What does CD mean in semiconductors?

Line width is how wide a line on a chip is. The industry often calls it the CD, short for critical dimension. More broadly, the gap between two lines or the diameter of a small hole can also be a CD: any size in a layer that matters most and has to be held tightly. It is measured after litho prints the pattern and again after etch cuts it, and it is one of the numbers a fab looks at most.

How metrology differs from inspection is covered in "How does a chip fab know each step went right?". This article is only about line width itself.

What happens if the line width is off by a few nanometers?

The switches in a chip are transistors: current flows from one end to the other, and a gate in the middle turns it on or off (see "What is a transistor?"). The gate is a thin line laid across the path, and how wide that line is printed decides how long the path for the current is.

• A narrower line: The path gets shorter, so the switch turns on more easily and runs faster; but when it is off, it also leaks more easily.
• A wider line: The path gets longer, the current gets smaller, and the switch is slower.

So it isn't enough for the average to be right; every line has to be close to it. The center and the edge of a wafer, the first and last wafers of a lot, this lot and the next all have to match. That is called uniformity, and without it some chips come out fast and some slow. Uniformity requirements are tight, and are often set as a statistical range.

Where is line width decided? Step by step

Step 1: It starts on the photomask

• What: Lines on the photomask are usually drawn four times larger and shrunk when exposed; whether other lines sit nearby is also worked in, so some are drawn deliberately wider or narrower (see "What is OPC?").
• Why: The lines are finer than the wavelength of the light, so a dense line and an isolated line drawn at the same width print at different widths.
• If not: When lines are finer than the wavelength, a small error on the photomask doesn't always shrink in proportion on the wafer; it can even be amplified.

Step 2: Exposure: dose and focus

• What: The scanner exposes with a set amount of light (the dose), with the image in focus (see "What is photolithography?").
• Why: A little more or less light changes the line width; out of focus, the image blurs, the line width changes and the sides tilt. So for a new layer, one wafer is exposed with the dose and focus stepped field by field, the line width is measured in each, and the range where both can be a little off while the line width stays in spec is found. It is called the process window.
• If not: Tools drift a little every day, and a wafer is not perfectly flat everywhere; with a narrow window, the line width keeps going out.

Step 3: Bake and develop

• What: After exposure the wafer is baked on a hot plate (the post-exposure bake), then developed, leaving lines of photoresist.
• Why: How far the reaction in the photoresist goes depends on temperature and time (see "What is photoresist?").
• If not: If a hot plate is cold on one side, the whole wafer comes out wide on one side and narrow on the other.

Step 4: Measure once after develop

• What: A few wafers are sampled and the width of the resist lines is measured, a check often called after-develop inspection (ADI).
• Why: The pattern is still only in the photoresist; if it's wrong, the resist can be stripped and the step redone (rework).
• If not: Once etch starts, the wrong width is cut into the material, and there is no going back.

Step 5: Etch, which may not match the resist

• What: The layer below is etched away following the resist lines (see "What is etching?"). If the sides get eaten in a little, or the resist edges get etched away a little, the line gets narrower; if too much protective film builds up on the sidewalls, the line comes out narrow at the top and wide at the bottom. That difference is called etch bias.
• Why plan for it: The difference can be set in advance and even put to use: the final lines on a chip are often finer than what litho can print, done by etching or by trimming the lines further.
• If not: When the etch conditions drift, the resist measures fine but the etched line comes out off.

Step 6: Measure again after etch

• What: The etched lines are measured again.
• Why: These are the real lines on the chip. If the after-develop number is fine and the after-etch number is off, suspect the etch step first; if both are off, look at litho first.
• If not: You can't tell whether the problem is in litho or etch, or which tool to adjust.

How is line width measured? From above, or worked out from light

An electron microscope looking straight down

Most often it is measured with an electron microscope built for line width (CD-SEM). When electrons hit the wafer, they knock other electrons out of the surface, and many more come out at the edges of a line, so in the image both sides of every line are bright. Software measures the distance between the two bright edges, and that is the line width. Three things to watch:

• Edges aren't straight: At nanometer scale an edge wanders like a ragged hem, so what is measured is the average over a stretch of line.
• The sides are hard to see: The sides aren't perfectly vertical and the bright edge has some width, so where the "edge" is depends on the method; numbers compare well only when measured the same way. To know how much the sides slope, the image is compared with computer-simulated images, or the wafer is cut open to look at the cross-section, and a wafer that has been cut can't be used again.
• Measuring changes it too: Some photoresists shrink a little when hit by electrons, most of all on the first measurement, so how many electrons to use is chosen with care.

Scatterometry: worked out from light

The other way takes no picture. A small patch of densely packed, repeating lines is printed in the scribe line (the narrow strip between chips that is cut away at the end), light is shone on it, and how much of each color comes back is measured. How wide, how tall and how sloped the lines are all change the reflection; a computer has worked out ahead of time what many shapes would reflect and picks the closest match. It is fast and can also tell how sloped the sides are, but it needs the right model and the optical properties of the materials, and it only measures that small dedicated pattern.

What happens after it is measured?

1. Onto a control chart: Each lot measured adds one point to the control chart, to see whether it is slowly heading one way (see "What is an SPC control chart?").
2. Out of spec after develop, redo it: The lot goes on hold first, then the resist is stripped and printed again (see "What is rework?").
3. Slow drift, adjust the next lot: The measured line width goes back to the scanner, and the next lot's dose is adjusted a little; the after-develop line width can also be passed ahead to etch, and the etch time adjusted a little (see "What is APC?").
4. A sudden jump, find the cause: A hot plate, a new bottle of photoresist, an etch tool; trace back which one changed.

Is the "nanometer" in a process name the line width?

No. Long ago, a process generation's name roughly matched the finest line that generation could make; later the name came to mean only "which generation" (see The news says "2 nm". What exactly is 2 nm? and The news says "angstrom era"). Line width is a different thing: every layer has its own target, and what is measured is how wide that layer really came out.

In fab24hr

Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab", swipe to the process engineer and tap "Litho" under the name: you're the litho process engineer. The four shifts follow the time you open the site; open it between noon and early evening and you get the afternoon shift, "Rework", whose first part is called "Line width": the coater/developer TRK-01 came back from PM this morning, and in the two lots out of it this afternoon, the line width is wide on one side and narrow on the other. The equipment engineer says: "If the post-exposure bake is a little off, the line width moves with it." Then come 12 line-width maps: one square per measurement site, blue where the line width drifted, and on some wafers all the blue sits on one side. Once you've marked the wafers to rework, the equipment engineer tells you: the four hot plates take turns, and wafers 3, 7 and 11 landed on HP-3, the one that runs cold on one side.

Tap "Etch" under the name instead and you're the etch process engineer. Open it in the morning for the morning shift: on metrology tool MET-01, the post-etch line width has 8 points in a row below the center line, each lower than the last. At 10:40 Ya-Ting calls: one lot's line width is a little above the control limit, still inside spec. Let it go on, or keep it back?

You can watch the electron microscope that measures line width at work in "Chip fab machines, taken apart". The company, people, lot numbers, tools and figures in the game are fictional; what line width is and how it is measured follow public, general sources. For other jargon, see Chip fab jargon in plain English.