Measurements on Copper Substrates

Film thickness measurements on Cu materials has proven to be very difficult, though possible. The hardware requires coherent optical interference; surface roughness seen on copper makes the reflectivity data appear “noisy,” destroying some of the fringe information. Also, the native oxide and Cu reflectivity seem to vary considerably from process to process.

Enhancements have been made to the software to help with some of these issues. We recommend using the automatic baseline correction setting in the measurement program. This will help adjust for the varying reflectivity.

More importantly, the new material file, “Cu_underresist.m1,” more accurately represents the Cu index data we have seen. We removed resist from a Cu-coated Si wafer, then made spectroscopic ellipsometry measurements of the optical constants (index). Using the “freqency fit” type in the program settings may improve the ability to use fit to separate correct measurements from noise-dominated measurements.

Because of copper’s sloping baseline reflectivity and the “noise” mentioned previously, our software algorithms have difficulty making initial thickness guesses on copper. You will probably need to experiment with the search offset values to find the best results, shifting the initial guess positive or negative. If the model (red curve) is higher freqency (more compressed) than the measured data (black curve), reduce the offset to push the guess to a thinner value.

Thicknesses of Cu greater than about 200 nm appear infinitely thick. Thus ignore layers under the Cu, setting the Cu as the substrate. Some cases have seen that a CuO layer of a few nm (fixed or variable) may achieve better results; we recommend you experiment with this.