ROHMAN Shahriar

Abstract: The need to develop increasingly efficient image sensors for consumer applications is driving microelectronics manufacturers to try to add additional functionalities to existing image sensors, such as extending the dynamic range of sensors, sensitivity in new spectral ranges (UV, NIR-SWIR), or detecting the contours of moving objects. In recent years, notable progress has been made on this latter type of sensor, also known as neuromorphic sensors, due to the use of denser technological nodes, as well as the development of hybrid bonding techniques for back-illuminated sensor technologies, allowing for the reduction of pixel size and, more recently, the integration of event-detection pixels associated with standard pixels capable of recording conventional color images using a Bayer grid. The applications of these new cameras are numerous: in the automotive sector, the ability of cameras to avoid dazzling in complex environments and to quickly identify elements within a scene can provide increased assistance for decision-making in autonomous vehicles. Similarly, in the field of security, the use of event-based cameras can enable the implementation of intelligent and rapid wake-up functions, associated with low power consumption and the preservation of anonymity. For virtual reality applications, the combination of fast image capture and contour detection is sought to improve real-time spatial localization performance and reduce the latency of user gesture detection. Finally, for mobile applications, the addition of fast contour detection functionalities can reduce motion blur during image capture and enable slow-motion video functions. In this context, STMicroelectronics, through this thesis in collaboration with CEA and the TIMA laboratory, aims to study different architectural options that could eventually lead to the development of its own image sensor technology combining conventional RGB imaging and event-based image capture, with the constraint of maintaining state-of-the-art RGB imaging performance in terms of resolution. To achieve this, the thesis will rely on the expertise of CEA and TIMA, which have been developing event-based sensor architectures for several years, particularly leveraging expertise in asynchronous circuit design. Several pixel options will be evaluated in this thesis, including the use of pixel architectures combining electron and hole readout to create pixels offering quasi-simultaneous APS and event-based operation, as well as the design of logarithmic transduction stages using silicon bipolar transistors to reduce pixel response variability, thereby improving the dynamic response range of sensors, reducing latency, and decreasing the number of false events detected by event-based sensors. Based on this work combining both analog design and physical simulations, pixel architectures already developed in previous theses will be reused to provide pixel arrays and proceed with the electrical characterization of the sensors once fabricated.