The 3-dimensional (3D) imaging is an important area which can be applied to face detection, gesture recognition, and 3D reconstruction. Many techniques have been reported for 3D imaging using various methods such as time of fight (TOF), stereo vision, and structured light. These methods have limitations such as use of light source, multi-camera, or complex camera system. In this paper, we propose the offset pixel aperture (OPA) technique which is implemented on a single chip so that the depth can be obtained without increasing hardware cost and adding extra light sources. 3 types of pixels including red (R), blue (B), and white (W) pixels were used for OPA technique. The aperture is located on the W pixel, which does not have a color filter. Depth performance can be increased with a higher sensitivity because we use white (W) pixels for OPA with red (R) and blue (B) pixels for imaging. The RB pixels produce a defocused image with blur, while W pixels produce a focused image. The focused image is used as a reference image to extract the depth information for 3D imaging. This image can be compared with the defocused image from RB pixels. Therefore, depth information can be extracted by comparing defocused image with focused image using the depth from defocus (DFD) method. Previously, we proposed the pixel aperture (PA) technique based on the depth from defocus (DFD). The OPA technique is expected to enable a higher depth resolution and range compared to the PA technique. The pixels with a right OPA and a left OPA are used to generate stereo image with a single chip. The pixel structure was designed and simulated. Optical performances of various offset pixel aperture structures were evaluated using optical simulation with finite-difference time-domain (FDTD) method.
In this paper, we propose a pixel averaging current calibration algorithm for reducing fixed pattern noise due to the deviation of bolometer resistance. To reduce fixed pattern noise (FPN), averaging current calibration algorithm by which output current of each bolometer reference pixel is averaged by the averaging current calibration is suggested. The principle of algorithm is that average dark current of reference pixel array is subtracted by a dark current of each active pixel array. After that, the current difference with information of pixel deviation is converted to voltage signal through signal processing. To control the current difference of pixel deviation, a proper calibration current is required. Through this calibration algorithm, nano-ampere order dark currents with small deviations can be obtained. Sensor signal processing is based on a pipeline technique which results in parallel processing leading to very high operation. The proposed calibration algorithm has been implemented by a chip which is consisted of a bolometer active pixel array, a bolometer reference pixel array, average current generators, line memories, buffer memories, current-to-voltage converters (IVCs), a digital-to-analog converters (DACs), and analog-to-digital converters (ADCs). Proposed bolometerresistor pixel array and readout circuit has been simulated and fabricated by 0.35μm standard CMOS process.
A 3dimensional (3D) imaging is an important area which can be applied to face detection, gesture recognition, and 3D reconstruction. In this paper, extraction of depth information for 3D imaging using pixel aperture technique is presented. An active pixel sensor (APS) with in-pixel aperture has been developed for this purpose. In the conventional camera systems using a complementary metal-oxide-semiconductor (CMOS) image sensor, an aperture is located behind the camera lens. However, in our proposed camera system, the aperture implemented by metal layer of CMOS process is located on the White (W) pixel which means a pixel without any color filter on top of the pixel. 4 types of pixels including Red (R), Green (G), Blue (B), and White (W) pixels were used for pixel aperture technique. The RGB pixels produce a defocused image with blur, while W pixels produce a focused image. The focused image is used as a reference image to extract the depth information for 3D imaging. This image can be compared with the defocused image from RGB pixels. Therefore, depth information can be extracted by comparing defocused image with focused image using the depth from defocus (DFD) method. Size of the pixel for 4-tr APS is 2.8 μm × 2.8 μm and the pixel structure was designed and simulated based on 0.11 μm CMOS image sensor (CIS) process. Optical performances of the pixel aperture technique were evaluated using optical simulation with finite-difference time-domain (FDTD) method and electrical performances were evaluated using TCAD.
Recently, CMOS image sensors (CISs) have become more and more complex because they require high-performances such as wide dynamic range, low-noise, high-speed operation, high-resolution and so on. First of all, wide dynamic range (WDR) is the first requirement for high-performance CIS. Several techniques have been proposed to improve the dynamic range. Although logarithmic pixel can achieve wide dynamic range, it leads to a poor signal-to-noise ratio due to small output swings. Furthermore, the fixed pattern noise of logarithmic pixel is significantly greater compared with other CISs. In this paper, we propose an optimized linear-logarithmic pixel. Compared to a conventional 3-transistor active pixel sensor structure, the proposed linear-logarithmic pixel is using a photogate and a cascode MOSFET in addition. The photogate which is surrounding a photodiode carries out change of sensitivity in the linear response and thus increases the dynamic range. The logarithmic response is caused by a cascode MOSFET. Although the dynamic range of the pixel has been improved, output curves of each pixel were not uniform. In general, as the number of devices increases in the pixel, pixel response variation is more pronounced. Hence, we optimized the linear-logarithmic pixel structure to minimize the pixel response variation. We applied a hard reset method and an optimized cascode MOSFET to the proposed pixel for reducing pixel response variation. Unlike the conventional reset operation, a hard reset using a p-type MOSFET fixes the voltage of each pixel to the same voltage. This reduces non-uniformity of the response in the linear response. The optimized cascode MOSFET achieves less variation in the logarithmic response. We have verified that the optimized pixel shows more uniform response than the conventional pixel, by both simulation and experiment.
In this paper, a binary complementary metal oxide semiconductor (CMOS) image sensor with a gate/body-tied (GBT) metal oxide semiconductor field effect transistor (MOSFET)-type photodetector is presented. The sensitivity of the GBT MOSFET-type photodetector, which was fabricated using the standard CMOS 0.35-μm process, is higher than the sensitivity of the p-n junction photodiode, because the output signal of the photodetector is amplified by the MOSFET. A binary image sensor becomes more efficient when using this photodetector. Lower power consumptions and higher speeds of operation are possible, compared to the conventional image sensors using multi-bit analog to digital converters (ADCs). The frame rate of the proposed image sensor is over 2000 frames per second, which is higher than those of the conventional CMOS image sensors. The output signal of an active pixel sensor is applied to a comparator and compared with a reference level. The 1-bit output data of the binary process is determined by this level. To obtain a video signal, the 1-bit output data is stored in the memory and is read out by horizontal scanning. The proposed chip is composed of a GBT pixel array (144 × 100), binary-process circuit, vertical scanner, horizontal scanner, and readout circuit. The operation mode can be selected from between binary mode and multi-bit mode.