Fig. 1. Schematic diagram of silicon wafer sample consisting of air layer， film layer， and air layer

Fig. 2. Flowchart for self-attention neural network processing spectral data

Fig. 3. Configuration of spectral interferometry film thickness measurement system

Fig. 4. Reflection spectrum of wafer and its Discrete Fourier Transform（DFT）

Fig. 5. Thickness distribution and RMSE curve of wafer sample

Fig. 6. Measurement fitting results and deviations of samples using SANN and DFT methods

Fig. 7. Thickness measurement results of wafer with surface defects

Fig. 8. Analysis of normal sample by Fourier transform and self-attention neural network

Fig. 9. Reflectance spectra of outlier samples

Fig. 10. DFT curves of outlier samples

Fig. 11. Visualization of weight matrix calculated by multi-head self-attention mechanism on input spectrum in self-attention neural network

Fig. 12. Weight analysis of outlier samples by self-attention neural network

Fig. 13. Average attention weight curves of self-attention neural network for samples