A 768x576 Logarithmic Image Sensor with Photodiode in Solar Cell mode

The wide dynamic range and the contrast indexed imaging of logarithmic sensors have a great interest for a large variety of applications from digital photography to automatic machine vision. Still today, the high fixed pattern noise, low sensitivity and image lag remain open in such designs. In this paper, we present a new logarithmic pixel design by using photodiode in solar cell mode. This pixel design has been successfully integrated into a 10um pitch 768x576-pixel CMOS sensor in a standard 0.35um technology and demonstrated an ultra low fixed pattern noise, high sensitivity and absence of image lag. The useful instant dynamic range has been measured superior to 120dB. GENERAL PRINCIPLE AND STRUCTURE OF THE SENSOR All the biological sensory functions follow logarithmic law. The fundamental advantage is the ratio metric response of a logarithmic law sensor. In imaging field, this property results in a contrast indexed image and very wide instant dynamic range. The early work on logarithmic sensors has been based on logarithmic current to voltage converting circuit using mainly MOS in sub-threshold mode [1-7]. Fig. 1 gives a principle schematic of such pixel design. But this kind of pixel designs cannot offer high quality image for general purpose imaging applications despite of years of R&D effort. The high fixed pattern noise, low sensitivity and image lag are the main obstacles. Figure 1. Traditional CMOS logarithmic pixel structure where the linear mode photo current is converted into a logarithmic voltage by a sub threshold mode MOS transistor. Inspired by the knee response in a CCD under blooming state, we have tested a photoreceptor design by using a photodiode under blooming state (solar cell mode) in order to get a logarithmic law signal from the photoreceptor[8]. The cross-talk issue has been resolved by using the P+ diffusion photodiode in a separate Nwell. This structure has given a good logarithmic law signal in function of illumination over more than 120dB dynamic range. But the quantum efficiency was poor and suffered from large image lag too. New pixel design with N-in-Psub photodiode has been developed since. A 160x120-pixel test chip for concept validation has been published [9]. Fig. 2 shows the basic electrical schematic of this newly developed logarithmic pixel. The photodiode working in solar cell mode is made with N diffusion in P-sub. A zeroing NMOS transistor crosses this photodiode and shorts it when it’s on. The negative photovoltaic voltage on the cathode of the N-in-Psub photodiode is translated into a positive one by the in-pixel PMOS voltage follower. A differential readout is realized in order to remove the additive FPN noise in the sensor readout chain as in a classic APS sensor. Figure 2. New logarithmic pixel design by using a photodiode in solar cell mode. The open-circuit voltage of the photodiode is sensed and a short-circuit MOS gives a precise dark reference signal for FPN compensation and also lag suppression. Fig. 3 illustrates the global architecture of this sensor. This sensor has been designed initially for analog surveillance camera, so no AD converter is integrated on-chip. Instead a CCIR/EIA TV encoder is integrated on-chip which permits a direct generation of the composite video signal from the sensor. Thanks to the huge dynamic range of this logarithmic response, no exposure and no AGC are needed. The circuit design is relatively simple. All the control signals are provided by an off-chip controller, either a micro controller or a FPGA. Figure 3. The general structure of this 768x576 logarithmic CMOS sensor with on-chip CCIR/EIA encoding circuit. Pixel lines in the array are sequentially selected by the vertical shift register. RD1 signal loads the image signal from the selected line into a first analog buffer. Then RST signal activates the zeroing transistor which short-circuits the photodiode and at the same time RD2 signal loads the zero voltage signal into a second analog buffer. The photodiode zeroing action gives not only a precise FPN compensation reference signal but also removes the image lag due to the residual charge on the photodiode. The difference between the two readout results gives a FPN compensated image which is output in a differential format. This differential image signal goes back into the CCIR/EIA encoder circuit. The video signal is amplified and mixed with TV synchronization pulse and black level. The composite TV signal will be available on the TVout pin. The sensor is powered with a 3.3V supply. The current drain is about 80mA. Fig. 4 shows some high contrast scenes captured by this sensor. For the characterisation purpose, we have built a CameraLink based camera board operating at 25 images per second. The differential output of the sensor is connected to a differential ADC. The AD conversion range is set at 1V with 12-bit precision. Figure 4. Some high dynamic range scenes captured by the sensor. The main characteristics and performance of this sensor are summarized in Tab. 1. In the following sections, we will focus on the measured results together the theoretical modelling of this pixel design. PHOTOELECTRIC RESPONSE The open-circuit voltage of an illuminated photodiode in its steady state can be simply calculated from Shockley equation: ) ln( s s D I I I q kT V + = λ (1) where S I q T k , , , and λ I represent respectively Boltzmann constant, absolute temperature, elementary charge, junction saturation current and photo current. In our design, the photodiode's open-circuit voltage is established progressively from zero voltage after the zeroing action. The duration of the establishment depends on illumination level and also exposure time. If the voltage is sampled before the fully established state, it can be deviated from the log law given by Shockley equation. By supposing a constant value D C for the junction capacitance, we can get a closed form response for the photodiode in function of photo current (illumination) and exposure time t : s C V t I I s T D I e I I I V V D T s + + = + − ) ( ln λ λ λ , where q kT VT = (2) From this analysis, we can see that the pixel response will be linear at low illumination level or at short exposure time. A very good match has been observed between the theoretical model and the measured data (Fig. 5). So the constant junction capacitance hypothesis is held thanks to the small voltage excursion on the solar cell mode photodiode. Response (logarithmic scale) 0 200 400 600 80