01 Dec 2014 What is the best way to measure coating thickness?

Our coating thickness measurement needs are very simple. We want a technique that is cheap, instant, safe, accurate, works on everything from 50nm to 1mm, can work in spot mode or scan mode and, perhaps, makes the coffee too. In reality we have to make do with compromises. But what are the facts behind the compromises? An excellent talk at last year’s AIMCAL Fall conference inspired me to create a set of apps at Thickness that show, as best I can, all the issues. The apps so far cover: Interference; Optical Density; Near-IR; Beta; Gamma. Also included are Cross-web issues and Defect detection. As with all my apps, if I’m missing something or have something wrong I’m always pleased to fix issues or add extra functionality.

For this blog I’ll focus on the Near-IR app.

The assumption is that your coating absorbs at some wavelength (λcoat) and sufficiently strongly that if you had a 50μm coating it would have a %Transmission defined by the %TCoat50 input. The lower that %T the stronger the absorption. Why 50μm? For app purposes I had to choose something simple. Unfortunately the substrate also absorbs IR with its λsub and %TSub50. If these are nicely separated then there’s no problem. But as the app shows, you very often have overlap of the peaks. Now you must select a probe wavelength λProbe at which to measure absorption. In real machines you tend to have a selection of probe wavelengths via a filter wheel. In the app you choose the one most relevant. In the example, the probe is somewhere between the two peaks – a deliberate choice on my part to show what problems arise from a bad choice or bad luck.

What we want to know is how accurate the coating thickness measurement will be. Let’s assume that the %T measurement has an error of 1%. The app shows that in this case the 1% error in %T leads to a 6.1% error in the measured thickness of the 10μm coating. Why? That’s best explained by playing with the app. If the substrate peak is at 2.15μm (giving more overlap) then the error is 10%, if it is beyond 2.28µm then the substrate has no effect and the error is 5.6%. If the probe is now moved to the centre of the coating peak the error reduces to 3.6%. If the coating is 30µm instead of 10µm, the error reduces to 2.8%. In other words, it all depends.

Clearly the ideal solution is to have a detector with very low errors in %T, tuneable to exactly your peak wavelength which has very strong absorption, with no interference from the substrate. That’s not going to happen. There are always cost/performance trade-offs. The point of the app is to allow you to decide if the available trade-offs give you the cost, convenience and performance you need. If not, then you need to see if one of the other techniques has a better set of trade-offs.

Use these apps for your own internal discussions and in meetings with potential vendors. The good vendors will have more sophisticated models to discuss with you. The bad vendors … well, you can work that out for yourself.

This blog first appeared in the Converting Quarterly Magazine from AIMCAL and I am grateful to the Editor, Mark Spaulding, for permission to re-use the text.