Multi camera, high data rate SPAD imaging for computation-intensive applications

Single-photon avalanche diode (SPAD) detection technologies are increasingly important across various scientific domains, due to their ability to capture high-resolution temporal data. However, they face critical challenges in {\em in-vivo} applications, such as afterpulsing artifacts that degrade signal fidelity, a lack of synchronization between multiple cameras that limits temporal alignment and data coherence, and the generation of extremely high data volumes that strain readout and processing capabilities. This thesis addresses these limitations by developing high-speed, synchronized, and computation-efficient camera architectures. A high-speed interface enables full readout of a $512 \times 512$ binary frame imager at up to 88 kfps. %, with continuous streaming at 50 kfps. A dual-camera system achieves pixel-to-pixel timing alignment error better than 120 ps and streams continuously at 25 kfps. The same synchronization method is applied to a dual TCSPC camera system ($32 \times 32$ resolution), reaching 50 Mcps per camera; it was integrated into a 3D quantum microscope with a 220-ps (FWHM) coincidence peak. The TCSPC gating mechanism is characterized, showing reduced afterpulsing via a hard gating technique, enabling photon collection up to 3 ns earlier. Building on these high-data-rate systems, two advanced camera architectures are introduced: an interleaved-gate design for FLIM optimization that mitigates photobleaching effects and preserves fluorescence response shape, and a GRU-based design for real-time fixed-point neural network execution using SIMD topology, supporting parallel processing of eight pixels. In this implementation, scheduling is performed manually, while instruction generation is automated. Finally, a novel SPAD imager is developed for {\em in-vivo} applications, featuring $64 \times 108$ pixels, 50 ps TCSPC resolution, and the 16 pixels groups' comb organization, where each group shares two TDCs following a Winner-Take-All algorithm. The imager is more efficient than its predecessors, thanks to the capability of better managing photon timestamps over extended periods of time. This is of paramount importance in quantum imaging applications. The image thus achieves a count rate of 901 Mcps, or an average of 141 kcounts per pixel, one of the highest ever achieved in a SPAD camera. These contributions collectively advance the capabilities of single-photon imaging systems in terms of speed, synchronization, and intelligent data processing.