Anti-reflective microstructured silicon surface could greatly improve the light trapping performance for silicon-based energy harvesting devices, thereby increasing the optical absorption efficiency and reducing the surface reflection. In the manufacturing of anti-reflective silicon surfaces, the thermal effect produced by laser ablation may lead to the collapse of microstructures with high aspect ratios, thus resulting in the rapid increase of reflectivity by seriously affecting the light trapping effect on microstructured silicon surfaces. Therefore, the microstructure characterization on silicon surfaces is of great significance for product quality and loss assessment of parts. Currently, offline methods including spectrometer or scanning electron microscopy (SEM) are widely employed. It needs to transfer the samples for observation after the preparation, and the fabricating process could not be optimized in time referring to processing results. To improve the quality of laser processing on anti-reflective silicon surfaces and shorten the optimization periods, researchers have proposed corresponding solutions via real-time monitoring. Laser ablation is often accompanied by the generation of acoustic, optical, electrical and thermal signals, and can be monitored based on a variety of sensors in real-time. Compared with other signals, acoustic signals exhibit excellent spatial resolution. At present, real-time monitoring technology based on acoustic signals is mostly adopted to monitor the changes in laser parameters or processing quality, while few studies focus on the forming of surface microstructures. Therefore, a real-time acoustic signal monitoring and processing method is put forward based on time-frequency domain processing to analyze the forming process of surface microstructures.

In this paper, real-time monitoring of acoustic signals includes two steps of sampling and feature extraction. The acoustic signal is converted into electrical signal through a pre-polarized capacitive microphone with a frequency response range from 20 Hz to 31.5 kHz. Then, the electrical signal is collected by an oscilloscope at a sampling frequency of 240 kHz. The frequency composition of real-time signal and its intensity change with time could be obtained by MATLAB software for short-time Fourier transform and fast Fourier transform. The acoustic signal is normalized in the characteristic section, and the proportion of the amplitude in the frequency domain of the acoustic signal could be extracted as the characteristic parameter for analysis to improve the linearity between the characteristic parameters of the acoustic signal and the average depth of surface microstructures. Based on the sound source generation mechanism, the surface morphology of the silicon surface during laser ablation could be monitored by the time-frequency spectrum of the corresponding acoustic signal. An acoustic measurement method based on the fast Fourier transform is proposed to obtain the height and width of microstructures. The reflectivity of the sample is considered the evaluation standard of the sample processing quality, and the amplitude of each acoustic signal frequency is taken as the input. The artificial neural network is applied to predict the processing quality of the anti-reflective silicon surface.

Firstly, the correlation between the acoustic signals in 0-30 kHz and characteristic sizes of the surface microstructure fabricated by laser ablation could be analyzed by the time-frequency spectrum of the acoustic signal. The acoustic signal generated during laser processing can reflect whether the ablated silicon surface forms periodic microstructure (Fig. 3), and the frequency composition of the acoustic signal could be tuned by microstructured width (Fig. 4). Then the effects of laser power and processing times on the microstructure depth are analyzed in detail, corresponding to the amplitude changes of each acoustic signal frequency. In the case of a fixed microstructure width, the normalized acoustic signal characteristic parameters change linearly with the microstructured depth, and the influence of laser power changes could be ignored (Fig. 10). Additionally, an artificial neural network is applied to forecast the processing quality by utilizing the acoustic signal as the input. The reflectivity of 5% on the silicon surface is defined as the processing quality boundary, and the actual measurement accuracy of processing quality prediction could exceed 90% via the artificial neural network. This indicates that the acoustic online monitoring can be employed to evaluate the surface processing quality of anti-reflective silicon surface in real-time (Fig. 13).

The real-time acoustic signal analysis in this paper can effectively monitor the surface microstructure morphology and predict the machining quality. Results show that the acoustic signal with the same frequency could be always captured during laser processing, regardless of whether the microstructured surface exists or not. The acoustic signal frequency related to the microstructure could be influenced by the microstructure depth and laser scanning speed. The ratio of the frequency component intensity to the total signal intensity is linearly correlated to the average structural depth on microstructured silicon surfaces. The proposed method can effectively work even after partial structured collapse. The artificial neural network is adopted to verify the correlation between the acoustic signal and the machining quality. The accuracy of the classification prediction is higher than 90%. This study provides a theoretical basis for the real-time quality monitoring of anti-reflective silicon surfaces fabricated by laser ablation. In the future, more attention could be paid to the underlying mechanism and monitoring method of acoustic signal generation at the moment of microstructured collapse, which could be combined with other existing signal processing methods to yield better processing quality.