[Introduction]In recent years, high-performance intelligent CMOS image sensor chip, as the core component of visual information acquisition, has increasingly become a national major scientific and technological research project (such as manned spaceflight, large aircraft, high-resolution earth observation, etc.) aerospace and space exploration important foundation and technical support in the field. Image sensors based on CMOS technology have become the mainstream devices for visible light imaging due to their monolithic integration, random reading of image information, low power consumption, small size, and low cost.
The three-dimensional integrated CMOS image sensor chip architecture lays the foundation for the construction of an intelligent imaging system. The intelligent bionic image sensor that simulates the mechanism of biological visual perception effectively solves the challenges of perception and processing capabilities brought about by the increasing demands of resolution and frame rate. At the same time, it also has the advantages of large dynamic range, low data redundancy and low power consumption.
According to Memes Consulting, Professor Xu Jiangtao and Assistant Engineer Cheng Silu from the School of Microelectronics of Tianjin University recently published a review article on “High Frame Rate CMOS Image Sensors for Hyperspectral Imaging” in the journal “Space Return and Remote Sensing”. . Professor Xu Jiangtao is mainly engaged in the research of CMOS image sensor chips and systems, power management circuits, digital image signal processing circuits, etc.
This research analyzes the imaging mechanism of intelligent CMOS image sensor, summarizes the types of intelligent CMOS image sensor chips, systematically reviews and summarizes the key technologies of intelligent imaging and identification, and looks forward to the possible development direction of future research in this field. Provide a reference for seizing the commanding heights of intelligent CMOS image sensor technology.
The metal layer and photodiode positions of the back-illuminated (BSI) CMOS image sensor and the front-illuminated (FSI) CMOS image sensor are just reversed. In the back-illuminated structure, the light directly illuminates the photodiode layer, which greatly improves the quantum efficiency and effectively enhances Imaging quality in low-light environments, and at the same time, it can suppress the problem of sensitivity drop caused by changes in light incident angle. The back-illuminated structure has promoted a series of major breakthroughs in CMOS image sensor technology, of which the most milestone is the stacked CMOS image sensor, which has undergone a series of developments: the first generation of traditional back-illuminated CMOS image sensors, the second The first generation of 3D stacked backside illuminated CMOS image sensors using through-silicon via (TSV) technology, and the third generation of 3D stacked backside illuminated CMOS image sensors using Cu-Cu connection technology.
Development History of Stacked CMOS Image Sensors
Imitating the asynchronous and independent perception mode of the human visual system, the visual system constructed with pulsed electrical signals and a large number of parallel and hierarchical processing mechanisms can achieve an efficient and stable working mechanism like human vision. Based on this biomimetic perception Image sensors with processing methods are called smart image sensors. Intelligent CMOS image sensor chips mainly include: single-chip fusion intelligent image sensor, new intelligent bionic image sensor – dynamic vision sensor, new intelligent vision chip – pulsed image sensor and new intelligent vision chip – bionic compound eye chip.
Schematic diagram of monolithic integrated image sensor chip and monolithic fusion intelligent CMOS image sensor chip
The efficient realization of intelligent image sensors is inseparable from the reading and processing of intelligent images. The traditional intelligent vision system combines the functions of image sensing and processing. With the development of the three-dimensional stacking integration process, the image sensor and the image processor can be integrated into the same chip. In addition to traditional silicon-based fusion methods, researchers are currently studying pixel-side fusion computing based on new materials, such as the realization of reconfigurable photosensitive arrays based on two-dimensional semiconductor material WSe2-a.
Realization of Reconfigurable Photosensitive Array Based on WSe2-a
The research of intelligent CMOS image sensor technology has achieved staged development, but there are still many problems and challenges beyond the performance of biological vision systems. The single-chip integrated pixel perception and calculation processing circuit greatly improves the integration of the system, but it is still the method of “perception first, calculation later”, and the amount of image data and calculation has not changed. The dynamic vision sensor adds a subtraction operation to the pixels, and adopts an event-driven method to greatly reduce the amount of data, but the simple subtraction operation also brings about the loss of some effective information in the image, and the application field is limited. Two-dimensional semiconductor materials realize the integration of photoelectric sensing and computing processing, but their fabrication process is not compatible with standard CMOS processes.
The follow-up research on intelligent CMOS image sensor technology should focus on the following two aspects: (1) research on intelligent image sensor technology based on CMOS technology with integrated sensing and computing; (2) research on intelligent image information processing and compression coding technology; (3) multiple Research on intelligent image perception and processing technology of sensor information fusion.
This project was supported by the National Natural Science Foundation of China (61774110).