We introduce a 1.58μm back illuminated CMOS image-sensor (BI CIS) which contains 4 Mega 2×1 shared on-chip micro-lenses (OCL). With this architecture, we realize a full phase-detection auto focus (PDAF) CIS, where all photo detectors (PDs) are L/R paired with a common OCL, the same color on-chip-color-filter (OCCF) and without metal grid between the L/R pixel boundary. The device has a good separation ratio of 2.59 and is more than twice sensitive under normal incident 550nm plane wave compared to current state-of-theart PDAF pixels with half metal-shielded aperture. Because the device has no obstacle structure on its optical path, it hardly suffers from reflection and diffraction of the incident light. This feature is adequate to realize PDAF function especially for the CISs with smaller pixels. In addition, the OCCF configuration of the device is a 45o rotated to Bayer CF pattern, and hence, every 2 lines contain only Green pixels. This CF configuration makes it suitable to detect the phase difference between L and R sub-images. Furthermore, this architecture is in particular suitable for HDR imaging by varying the exposure times for each L and R sub-images because it is possible to capture two set of sub-images with almost identical optical characteristics to the on-focus object. In this paper, we report on the device performance and propose versatile applications including high-sensitive AF and HDR for the full-PDAF CIS. In the past decade, the rapid progress of CMOS image sensor technology has drastically improved their image quality along with the trend of shrinking pixel size, increasing pixel number and smart signal processing. Recently, Back illuminated CMOS image sensor with a stacked logic structure were reported, which realize various on-chip functions including PDAF and HDR functions[1].---We have developed a novel prototype device with 1.58μm◇, 2×1 shared OCL technology, so-called micro-split lens (μSL) pixel, as shown in the pixel schematic of Figure 1. Each pixel is equipped with L and R PDs. There is no metal ridge at the L/R pixel boundary which minimizes the reflection and diffraction caused by such obstacle structures. Therefore, the device achieves a two times higher sensitivity than conventional architectures with half-shielded PDAF pixels and provides a better L/R discrimination capability[2].Figure 2(a) shows a block diagram of the chip. A chip is equipped with a 4 Mega 2×1 shared OCL pixel array, diagonally 4.60mm (Type 1/4) in size, load transistors, H/V scanners and column parallel ADCs in the peripheral. The inset (b) is a micrograph of the chip including a close-up view of the OCL of the pixel.------------------------------------0 0 0 0 0 0 0 0 0 0 0 0 We designed the pixel structure to achieve good L/R discrimination capability by using FDTD simulation. Figure 3(a), (b) show schematics of the pixel and (c), (d) are the snapshots of magnetic field |H| distributions cut along the sensor plane obtained with the simulations. The simulations were performed with the incident plane wave at λ = 550nm with the incident angles of θ = +15o (a), (c) and θ = 0 o (b), (d), respectively. Here the angle is defined to be θ = 0 o for the normal incident light and is changing along with the plane including the horizontal axis of the pixels. As shown in (c), (d), the incident light is mainly guided into the left pixel at the incident angle of θ = +15o and is equally distributed to both the L and R pixels at θ =0o.--------------Figure 4 shows the incident angle dependency of a green pixel sensitivity of the device with the simulation (c) and measurements (d).The data are taken along the horizontal plane denoted as the magenta lines on (a),(b). The incident light is set to λ = 550nm both in the simulation and measurement and the angle sweep ranges from θ = +30o to θ = -30o in steps of 2o. Solid lines represent the data of this μSL pixel (1.58μm◇), and the dotted lines are for those of the half shielded PDAF pixel (1.12μm□), showing good consistency between the simulation and the measurements. The μSL pixel at θ = 0o achieves a 2.27 times higher sensitivity compared to the PDAF pixel and almost equal sensitivity to the normal green pixel, because the area of each L/R pixel of the 1.58μm◇ PDs is same as that of 1.12μm□ pixel. The μSL pixel has a separation ratio as high as 2.59 whereas that of the PDAF is about 1.65. The definition of separation ratio is:----------------------------------To evaluate and demonstrate the image quality of the device, we prototyped the AF lens module with F2.4, focal length of 2.87mm and 4 piece plastic lenses. As the CF configuration of the device is different from the Bayer format, a specifically tailored remosaicing block is required for the system. In a prototype, we implemented this function on the FPGA board and delivered the output data in Bayer format. This data is subsequently processed with a general demosaicing block to get the final RGB image. Figure 5(a) shows a sample image of the star-chart[3] taken with the device after full image reconstruction pipeline. Figure 5 (b) is the close-up view of the area which is denoted as yellow line in (a).------------------------------------------------One of the advantage of this device is the density of PDAF pixels and 100% pixels are paired PDs. State-of-the-art devices that have embedded PDAF pixels allow only few percent for the PDAF function pixel in order to maintain the image quality. Furthermore, the 45o rotated Bayer CF SSSSSSSSSSSSSSSSSSSS RRSSSSSSSS = AAAASS � SSSSSSSS LLpixel SSSSSSSS RRpixel ddθθ, +10 0 � SSSSSSSS RRpixel SSSSSSSS LLpixel ddθθ 0