Refine
Year of publication
- 2019 (2)
Document Type
- Journal article (2)
Language
- English (2)
Has full text
- yes (2) (remove)
Is part of the Bibliography
- yes (2)
Institute
- Life Sciences (2)
Publisher
An ultraviolet visible (UV–Vis) spectroscopy method was developed that can quantitatively characterize a technical copper surface to determine oxide layers and organic impurities. The oxide layers were produced by a heating step at 175 ℃ for four different times (range = 1–10 min). Partial least squares (PLS) regression was used to establish a relation between the UV–Vis spectra and film thickness measurements using Auger electron spectroscopy depth profiles. The validation accuracy of the regression is in the range of approximately 2.3 nm. The prediction model allowed obtaining an estimation of the oxide layer thickness with an absolute error of 2.9 nm. Alternatively, already known methods cannot be used because of the high roughness of the technical copper surfaces. An integrating sphere is used to measure the diffuse reflectance of these surfaces, providing an average over all angles of illumination and observation.
The detection and characterisation of oxide layers on metallic copper samples plays an important role for power electronic modules in the automotive industry. However, since precise identification of oxide layers by visual inspection is difficult and time consuming due to inhomogeneous colour distribution, a reliable and efficient method for estimating their thickness is needed. In this study, hyperspectral imaging in the visible wavelength range (425–725 nm) is proposed as an in-line inspection method for analysing oxide layers in real-time during processing of copper components such as printed circuit boards in the automotive industry. For implementation in the production line a partial least square regression (PLSR) model was developed with a calibration set of n = 12 with about 13,000 spectra per sample to determine the oxide layer thickness on top of the technical copper surfaces. The model shows a good prediction performance in the range of 0–30 nm compared to Auger electron spectroscopy depth profiles as a reference method. The root mean square error (RMSE) is 1.75 nm for calibration and 2.70 nm for full cross validation. Applied to an external dataset of four new samples with about 13,000 spectra per sample the model provides an RMSE of 1.84 nm for prediction and demonstrates the robustness of the model during real-time processing. The results of this study prove the ability and usefulness of the proposed method to estimate the thickness of oxide layers on technical copper. Hence, the application of hyperspectral imaging for the industrial process control of electronic devices is very promising.