PRAGMATICS (Prototype Readiness of Advanced Integrated and Embedded Modules for Adaptive CMOS Optoelectronic Sensors) is a coordinated proof-of-concept initiative aimed at transferring the technological advances developed in the SEMIoTICS project into pre-industrial prototypes. These prototypes integrate intelligent signal processing directly into CMOS image sensors, enabling adaptive, high-performance visual systems for edge computing applications. The project is structured around three subprojects led by IMSE (CSICUniversidad de Sevilla), CiTIUS (Universidade de Santiago de Compostela), and UPCT (Universidad Politécnica de Cartagena). Each team contributes to a distinct innovation line: (1) high-dynamic-range imaging with concurrent auto exposure, (2) event-based and compressive sensing for efficient visual data acquisition, and (3) deep learning hardware accelerators for near- and in sensor AI processing. The IMSE team has developed a patented HDR-AE pixel architecture that adapts exposure at the pixel level using local and global illumination estimates, enabling single-shot HDR imaging without motion artifacts or post-processing. This innovation is being prototyped in a 3D-stacked CMOS sensor to improve fill factor incorpórate functionality while retaining image quality, targeting TRL56. CiTIUS focuses on integrating custom CMOS sensors with FPGAs to create embedded vision systems capable of real-time object detection using both frame-based and event-driven data. These systems will be validated in collaboration with local SMEs in robotics and IoT, aiming to reach TRL6 and explore commercialization pathways, including potential spin-offs. UPCT contributes with the design of compact binary convolutional neural networks (BNNs) optimized for integration with CMOS sensors. Their mixed signal in-memory computing architectures reduce power consumption and silicon area, enabling real-time CNN inference at the sensor edge. A demonstrator will showcase the integration of these accelerators with custom or commercial sensors. The project emphasizes vertical integration of sensing and processing through 3D stacking, enabling efficient, low-latency, and power-aware vision systems. It adopts a lean startup methodology, progressing through prototyping, customer validation, and business planning. Industrial engagement is central, with demonstrators to be showcased at trade fairs and direct collaboration with potential partners.
Project financed PDC2025-165990-C31 by MICIU/AEI /10.13039/501100011033.
