Modelling and simulation of an energy harvesting CMOS image sensor with data compression

Advancements in CMOS technologies continue to bring power consumption improvements in a wide range of applications, including CMOS Image Sensors (CIS). Within the broad field of image sensors, low power optimised CIS have emerged as a unique research area, with state-of-the-art power consumption in the tens of microwatts. This level of power is comparable to the power that can be generated using a few square millimetres of photodiodes biased in the photovoltaic (PV) operating region. This naturally forms the question whether the photodiodes present in the pixel array can serve a dual purpose, and provide the power required to capture images. This type of energy harvesting sensor faces an additional challenge to those already present in the design of a low power CIS, in that the power available from the photovoltaics will be variable. The first part of research in this thesis covers the design and characterisation of photovoltaic testchips in a dedicated image sensor process technology with deep trench isolation (DTI) capability. Photodiodes with four different doping profles were manufactured and measured, focusing on their power generation capabilities. The DTI separating the photodiodes from the substrate allowed for the first time to connect multiple diodes in series to passively generate high voltages on chip without efficiency losses. Under 1klx of sunlight the PV can generate 736nW=mm2. One of the diodes was integrated in a modified 3T pixel, enabling both imaging and energy harvesting operating modes. A small array of these pixels was integrated in a dedicated chip to allow characterisation of PV performance when exposed to real scenes. Based on the measured results the best performing PV is identified, and a simulation model is matched to its performance. The PV model underpins the second part of this thesis, which concerns the modelling and simulation of an energy harvesting CIS. The envisaged imager includes a novel method for image data compression during capture which reduces overall power consumption by lowering the energy needed to output data. The compression is based on a discrete wavelet transform that is calculated in the time domain, during readout. The method enables energy-quality scalable capture, allowing for control over average power consumption at a constant framerate, addressing the variable nature of the PV power supply. The core functional blocks proposed in this thesis are expected to consume between 213nW and 842nW when capturing 1fps, depending on the selected con guration and based on simulations for a specific image. This is increased to between 1.46μW and 6.25μW for the full system, corresponding to a lowest required illumination around 6klx with the most aggressive compression enabled, and 23klx when capturing full quality images.