Research On Dead Pixel Detection And Inpainting Technology Of 3D Stacked Image Sensor

3D stacked CMOS image sensor is one of the mainstream directions of image sensor,which has the advantages of high-density interconnection,large fill factor,and independent optimization of different layers.A typical structure of 3D stacked image sensor includes a pixel array,an analog-to-digital converter(ADC)array,and an image signal processor(ISP),which are connected vertically by through silicon vias(TSVs)or micro-bumps(?bumps).In the vertical interconnect path,any device or TSV,?bump failure will result in the loss of information in the corresponding area of the image,forming a continuous dead pixel.In this paper,a fault-tolerant 3D stacked image sensor readout architecture is studied.The fault-tolerant pixel-interleaving scheme changes the location distribution of the latent defective pixels,which is convenient to subsequent processing.A detection method for locating dead pixels based on the information of the neighboring pixels and simple arithmetic operations is proposed.An inpainting method used for the damaged block of pixels is proposed,the image is progressively reduced through a process of the discrete wavelet transform(DWT)and down-sampling until the damaged block converts to one pixel only.The reduced image is inpainted and optimized,and then the original image size would be restored by up-sampling.Finally,the hardware implementation of inpainting method is performed.The experimental results show that the detection method of the dead pixels in the case of random dead pixels and 2×2 defective pixel blocks is above 99.7%.The inpainting method is implemented in the case of single or multiple TSV/?bump failures.Evaluations over a range of images demonstrate that the average peak-signal-to-noiseratio(PSNR)in the case of single TSV/?bump failure improves 2.35-3.26 dB compared with the exemplar-based method and the TV method and the average peak-signal-tonoise-ratio(PSNR)in the case of multiple TSV/?bump failures improves 1.93-2.39 dB compared with the exemplar-based method and the TV method.