SPADnet · Fully Networked, Digital Components for Photon-starved Biomedical Imaging Systems

SPADnet aims to develop a new generation of smart, CMOS-based large area networked image sensors for photon-starved biomedical applications, build ring-assembly modules for Positron Emission Tomography (PET) imaging, and carry out performance tests in a PET system evaluation testbed.<br/>While suited to applications offering repetitive measurement techniques, existing sensors are not well adapted to single-shot, rare events often occurring in diagnostic tools based on specific radiation detection, PET, SPECT, gamma cameras, and other minimally-invasive / point of care tools. In addition, the relatively small field-of-view of existing sensors is a limiting factor.<br/>SPADnet's prime objective is to develop a scalable photonic component for large format, rare-event imaging. The core of the component will be a SPAD array implemented in CMOS. Large formats will be achieved by tessellating several tens of dies in abutment style using innovative packaging techniques based on through silicon vias (TSVs).<br/>The ability to stamp the time and position of each photon impingement in a burst event offers a second key advance. The concept of spatial oversampling is introduced, where a single measurement is partitioned into a myriad of sub-measurements, occurring simultaneously. The difference is that in space oversampling many SPADs will detect the same event independently, thus reducing the dead time on average by the number of detectors involved.<br/>The decomposition of the large format imager to a network of independent arrays is key to managing massive data streams. In conventional PMTs or SiPMs, the sensitive device produces a stream of analog electrical pulses. The photonic component proposed in this project on the contrary generates streams of precomputed digital data.<br/>The current state-of-the-art on inter-chip data exchange will be the basis for efficient data communication, in a true network communication style. Data packets will be routed in the network and will be handled on-demand. For example coincidence mapping engines can be used in this context as snoopers on the data bus, thus considerably simplifying systems such as PET.