Parameter Detection For Surface Acoustic Wave Filter Based On Image Processing

With the rapid development of the fifth-generation network communication technology,signal base stations of 5G cell phone are laid on a large scale,global smartphone sales have surged to a record 250 million,surface acoustic wave(SAW)filter,as the core device of the smartphone RF front-end,is responsible for receiving the base station signal and transmitting port frequency.In the frequency range up to 1GHZ,the SAW filter occupies a monopoly position due to its stable performance and low price,and SAW filter is competing for share with the bulk acoustic wave filter in the higher frequency bandwidth,the parameters of the SAW filter components will significantly affect its performance.Currently,the main method for microscopic dimension measurement of semiconductors is to use various high magnification microscopes with scale to take microstructure images,manually mark certain feature positions,and compare the selected feature positions with the scale to finally obtain the desired actual dimensions.Usually,because of the low signal-to-noise ratio in the region of interest,the boundary position is not marked and acquired accurately enough,which leads to large errors.With the development of computer arithmetic and image processing technology,it is an inevitable trend that the traditional SAW filter parameter detection method is replaced by the intelligent detection method.Based on this,this paper proposes an image processing-based SAW filter parameter detection technique and applies it to the production line yield inspection,including the accurate measurement of SAW filter number,width,and period.Digital image processing technology and deep learning technology are used to process SAW filter images to achieve parameter measurement.The main work of this paper is as follows:(1)In this paper,we construct an image dataset of SAW filter.First,we cut SAW filter chip substrate to separate the epoxy resin from the circuit structure to obtain the exposed piezoelectric layer;then,we used a high magnification laser microscope to photograph the internal structure of the SAW filter and labeled the filter edges with Labelme software.The structure of the SAW filter is simple and the sample similarity of the dataset is large,so we construct the dataset by acquiring filter images with different magnifications and filter images with residual epoxy noise,and we perform data enhancement and optimal amplification to further strengthen the sample diversity.(2)Based on a large amount of data collected,this paper proposes a method based on image processing to detect SAW filter parameters.We use the morphological operator to recover an image and identify the number of interdigital transducer finger bars under noise conditions.The feature points in the color image are selected to extract the scale.The Canny edge detection is enhanced through using an adaptive filter,increasing the gradient operator,bilinear interpolation.We use a minimum shape to enclose the edges to determine the interdigital transducer width and period and apply Fourier transform to verify the detection results.The experimental results show that the number of interdigital transducer finger bars is accurately identified.The width and period are the same as the real values at the 10 th percentile,and the detection requirements are met.(3)Based on the previous method,a channel attention-based HED algorithm is proposed for SAW filter parameter detection.We add SE module to the HED model to strengthen the connection between network channels and improve the attention of the region of interest.The method has good identification of the number for both normal and noisy images,and there is no need to deal with noise images,the edge detector is evaluated and the results are compared with other advanced computer vision architectures;the deep learning-based method and the image-processing-based method are validated against each other,and both methods have good accuracy,but the deep learning-based method is more stable in the noise attack.Through experimental analysis: the detection algorithm using deep learning has better robustness while accomplishing the task of detecting the parameters of SAW filter.