The CMOS buried multi-junction (BMJ) detector with multiple outputs has distinct spectral responses that may be exploited for applications such as bio-chemical analysis. We tackle here dark current issue by identifying different components inside the detector structure. The identification methods are based on the observation of bias and temperature dependence, as well as measurements of test detector chip integrating different design variations. Surface thermal generation may become predominant when the detector size shrinks, thus causing dark current degradation. To prevent this effect, we propose a low-sized detector structure with passivation of all its surrounding Si/SiO2 interface areas.
Also for the detector readout, we present a multi-channel charge-amplifier architecture with noise analysis. Effects of noise coming from amplifiers and related to the coupled detector’s dynamic conductances are illuminated. To pick up weak signals, synchronous detection can be implemented. A BDJ (Buried Double Junction) detector chip designed with a switched-phase architectural approach gives a minimum detectable signal of 15μlx@555nm or 1μlx@555nm at 27°C or – 10°C, for an integration time of 3s or 45s respectively.
This paper presents an accurate temporal noise analysis of a new kind of CMOS image sensor for colour design. Operating in the charge storage mode, the noise of this APS is described with a time-varying model. During the reset phase, as the steady state is fast established, classical frequency-domain noise analysis can still be used to determine noise at both sensing nodes. Good agreement is observed between the results obtained by simulation with Cadence CAD tools of a 0.35μm-CMOS test structure and the behaviour predicted by the proposed analytical approach. During the integration phase, as both junctions are floating, the stationary state condition is never fulfilled and the noise analysis must be carried out in the time-domain. Our contribution consists in taking into account the non-linearity of the junction capacitance, which yields more realistic results. Considering only the dominant white noise component, it clearly appears that the junction non-linearity improves the output SNR for both sense nodes at high illumination and/or high integration period.