Advanced techniques for SPAD-based CMOS d-ToF systems

The possibility to enable spatial perception to electronic devices gave rise to a number of important development results in a wide range of fields, from consumer and entertainment applications to industrial environments, automotive and aerospace. Among the many techniques which can be used to measure the three-dimensional (3D) information of the observed scene, the unique features offered by direct time-of-flight (d-ToF) with single photon avalanche diodes (SPADs) integrated into a standard CMOS process result in a high interest for development from both researchers and market stakeholders. Despite the net advantages of SPAD-based CMOS d-ToF systems over other techniques, still many challenges have to be addressed. The first performance-limiting factor is represented by the presence of uncorrelated background light, which poses a physical limit to the maximum achievable measurement range. Another problem of concern, especially for scenarios where many similar systems are expected to operate together, is represented by the mutual system-to-system interference, especially for industrial and automotive scenarios where the need to guarantee safety of operations is a pillar. Each application, with its own set of requirements, leads to a different set of design challenges. However, given the statistical nature of photons, the common denominator for such systems is the necessity to operate on a statistical basis, i.e., to run a number of repeated acquisitions over which the time-of-flight (ToF) information is extracted. The gold standard to manage a possibly huge amount of data is to compress them into a histogram memory, which represents the statistical distribution of the arrival time of photons collected during the acquisition. Considering the increased interest for long-range systems capable of both high imaging and ranging resolutions, the amount of data to be handled reaches alarming levels. In this thesis, we propose an in-depth investigation of the aforesaid limitations. The problem of background light has been extensively studied over the years, and already a wide set of techniques which can mitigate the problem are proposed. However, the trend was to investigate or propose single solutions, with a lack of knowledge regarding how different implementations behave on different scenarios. For such reason, our effort in this view focused on the comparison of existing techniques against each other, highlighting each pros and cons and suggesting the possibility to combine them to increase the performance. Regarding the problem of mutual system interference, we propose the first per-pixel implementation of an active interference-rejection technique, with measurement results from a chip designed on purpose. To advance the state-of-the-art in the direction of reducing the amount of data generated by such systems, we provide for the first time a methodology to completely avoid the construction of a resource-consuming histogram of timestamps. Many of the results found in our investigations are based on preliminary investigations with Monte Carlo simulations, while the most important achievements in terms of interference rejection capability and data reduction are supported by measurements obtained with real sensors.