5.1 Imaging technologies
Image quality improvements for RGB-stack-type image sensors
We are conducting research on a single-chip color imaging device using organic photoconductive films (organic films) with the goal of realizing compact, high-definition color cameras. The organic films absorb only a specific color of light and convert it to an electrical signal and are transparent to other colors of light, so they can be used to realize an imaging device that separates colors in the device-depth direction, by stacking multiple layers of organic film. In FY2020, we prototyped a three-layer color imaging device with two types of organic film stacked over a CMOS image sensor (CIS), and began developing an active pixel sensor circuit.
For the three-layer color imaging device prototype, we first formed a transparent thin-film transistor (TFT) array for reading the electrical signal from the organic film and an organic film sensitive only to green light (organic film for green) on top of the CIS, which converts red light to an electrical signal, creating a red/green two-layer device. On this device, we then stacked another device that we developed in FY2019, consisting of a transparent TFT array and organic film for blue, formed on a glass substrate, to obtain the three-layer color imaging device (Figure 5-1). The prototype device has QVGA resolution (320×240), pixel pitch of 20 μm, and achieved color video output at a frame rate of 60 Hz (Figure 5-2)(1).
The pixel circuits in TFT arrays that we used previously read currents using a single transistor, which makes them susceptible to external noise. To deal with this, we began developing an active pixel sensor circuit that has an amplifier for each pixel and reads a voltage signal. Realizing this circuit requires use of three TFTs per pixel. We first designed the amplifier circuit and then used circuit simulation to verify that we could drive a QVGA array of pixels at a frame rate of 60 Hz. We then worked on reducing the TFT channel length (2 μm). We developed the TFT fabrication process needed to reduce the size and introduced photolithography using stepper to reduce the channel length to 1 μm. We prototyped the active pixel sensor circuit using this TFT (Figure 5-3), and verified that it was able to read a signal in under 35 μs, which is the time required to read one frame for QVGA resolution driven at a frame rate of 60 Hz(2).
Development of the organic film for blue in this research was conducted in collaboration with Nippon Kayaku Co. Ltd.
Improving SNR on charge-multiplier imaging devices
We are developing a CMOS solid-state imaging device that uses a crystalline selenium photoconductive film that has a charge-multiplier function (multiplier film), with the goal of implementing Highly-sensitive 8K Ultra-High Definition cameras. In FY2019 and earlier, a nickel oxide electron blocking layer was used to prevent electrons from being injected from the pixel electrode in the CMOS signal readout circuit into the crystalline selenium layer, which resulted in spot noise in the captured image, and techniques were developed to improve the image quality when using the multiplier. In FY2020, we studied operation of the charge multiplier in this technology in more detail, further reduced the dark current in the multiplier film, increased the multiplication factor, and reduced the noise in the CMOS signal readout circuit.
To verify operation of the multiplier film, we used the imaging device that we prototyped in FY2019 and studied differences in the charge multiplication facor depending on wavelength. We found that when incident light was blue, the multiplication factor was approximately 1.4, but when it was red, the factor was smaller: approximately 1.2. This can be attributed to the higher absorption rate of blue light in the crystalline selenium, causing the charge to be generated at a shallow position in the layer, while the absorption rate for red light is lower so the charge is generated at a deeper position. This means that the distance the charges travel in the multiplierfilm are different, resulting in a different multiplicatin factor. This suggests that the charge multiplication is due to an avalanche phenomenon in the multiplier film (Figure 5-4)(3). We also found that lattice defects occur due to tellurium atoms, which are intentionally added to prevent membrane peeling, when they diffuse in the crystalline selenium layer, separating bonds between selenium atoms and resulting in dark current. To prevent this, we added chlorine atoms, which appears to suppress dark current(4).
To further increase the multiplication factor, we have continued development of technology to bond a multiplier film formed on a single-crystal sapphire substrate onto the circuit. A combination of crystalline selenium and gallium oxide is used on the sapphire substrate, but with this bonding structure, heat tolerance of the CMOS circuit is not a concern so it can be crystallized at a high temperature of 800 ˚C. This further improves the crystallinity in the multiplier film, achieving a higher multiplication factor. In FY2020, we prototyped a device by depositing each layer including crystallized gallium oxide on a sapphire substrate, pressure bonding it to an amorphous selenium film on a glass substrate, and then applying heat (160 ˚C) to crystallize the selenium (Figure 5-5).
Upon evaluation, we obtained a multiplication factor of approximately 10 when applying 18 V to the film(5). We also applied the nickel oxide layer and addition of chlorine into the selenium to the bonding structure as described earlier, to bond to the CMOS circuits for obtaining the image.
For the CMOS signal readout circuit, we made improvements to developments made in FY2019, such as to the lag characteristics of the circuit that electrically separates the floating capacitance of the multiplier film and the pixel electrode, increasing the charge-voltage conversion gain(6), and we studied circuits to suppress noise produced when the floating diffusion (capacitance that converts charge from the film to voltage) is reset. We also applied these results to the design of the signal readout circuit for a multiplier film.
Parts of this research were done in collaboration with Tokyo University of Science.
Basic technologies for computational photography
We are conducting research on computational photography with the objective of obtaining precise 3D information from objects. In contrast with conventional capture methods in which an image of the object is captured after being formed using a lens, computational photography reconstructs an image of the object computationally, based on light information modulated by interference or a coded aperture. In FY2020, we verified the principles underlying 3D information capture using incoherent digital holography (IDH) and coded imaging that can reconstruct high-resolution images from low-resolution data.
With ordinary holography, highly-coherent laser beam is used for illumination, but IDH uses incoherent light such as natural light or LED light as illumination to form interference fringes, obtain a hologram from the captured interference fringes, and reconstruct the image of the object. A hologram contains amplitude and phase information from the light, so by computing the back-propagation, the shape and 3D information of the object can be obtained. In this case, we built an optical system that separates the light from an object illuminated using LEDs into two components and captures interference fringes by changing the relative phase between the two to obtain a hologram. We then conducted experiments to capture images of 3D objects. (Figure 5-6). From the hologram, we are then able to reconstruct images of the object computationally, with the focal point at any position (Figure 5-7). We also investigated the effects of random noise from the image sensor used to capture the interference fringes on the reconstructed images and found that noise from image capture appears as granular noise in the reconstructed images, and this noise becomes diffused and decreases as the distance to the object increases(7).
For coded imaging, we placed a mask, which reflects part of the light from the object, into the optical system and conducted experiments capturing images of the object, varying the mask pattern or the relative positions of the mask and image sensor. We then reconstructed high resolution images computationally, from multiple captured low-resolution images. The resolution of the images obtained was higher than the total number of pixels in all of the captured low-resolution images used.
