LOOKING BEYOND IMAGES: LOW-POWER SENSOR ARCHITECTURES FOR 2D/3D IMAGING AND VISION
Abstract
TITLE: LOOKING BEYOND IMAGES: LOW-POWER SENSOR ARCHITECTURES FOR 2D/3D IMAGING AND VISIONObjective:The objectives of this project are routed to the merging of sensors and processors intomicro-electronic systems. The focus is on image and vision sensors. Visual stimuli arecrucial for systems to become aware of changes in the environment and react accordingly.Owing to progress in CMOS technologies, image sensors are proliferating and are beingused for an increasing number of applications. However, a big new wave of innovation isforecasted regarding the progressive incorporation of artificial vision into systems and~things~. The challenges of artificial vision intersect many domains, what requires aholistic approach. This project addresses the level where visual signals are acquired,namely the front-end between visual stimuli and processors.Approach:The research activities embrace the proposal of innovative sensor architectures and associated circuit building blocks. They will be demonstrated through dedicated camera modules and sensor-processor systems. SOW:Research topics include:1. Incorporation of scene depth analysis based on CMOS-compatible Single Photon Avalanche Diodes (SPAD). SPAD-based sensors will be explored along several edges:o Analysis of the SPAD physics and optimization of the physical structures formaximum responsivity, minimum noise and improved NIR response.o Design of advanced CMOS pixels combining active quenching and recharge, for low-power and maximum speed and maximum fill factor with a minimum pitch.o Design of readout channels, TDC architectures and strategies for optimalplacement, such as per-pixel, per-column, per-region, etc.o Architecture exploration and design of the circuitry to combine 2D and 3D image acquisition in the same sensor. Very useful for full scene understanding.o Design of memory-efficient, embedded post-processing architectures andcircuits to handle the inherent statistical nature of SPAD measurements.2. Incorporation of parallelization strategies, for signal-adaptive readout and early processing, with a minimum penalty on pixel pitch and fill factor. Parallelization is mandatory to handle the huge amount of image data with maximum throughput and minimum memory and energy budgets. Specific topics for research include:o Sensor architectures tailored for deep learning algorithms. In line with previousachievements of the proposers regarding bio-inspired sensors for image featureextraction and spatial-temporal filtering, including event-based sensors,o Sensor architectures for combining single-shoot, signal-adaptive High Dynamic Range (HDR) image capture with ultra low-noise readout channels. Target is detection at video rate. Time-based signal encoding combined with spatial filtering will be mostly explored for HDR, while multi-sample correlated double sampling will be explored for low-noise.o Optimum mapping of parallelization architectures into vertically-integratedtechnologies. Particularly, a technology provided by Teledyne-AnaFocus will beemployed. This technology combines backside illumination with two differenttypes of vertical connections.o Research activities will also be pursued regarding the mapping of SPAD pixeland sensors on this vertically integrated technology
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 24, 2019
- Source ID
- N000141912156
Entities
People
- Ángel Rodríguez-vázquez
Organizations
- Office of Naval Research
- United States Navy
- University of Seville