<title>Noise In Charge Injection Device (CID) Image Sensors</title>

Abstract

The Charge Injection Device (CID) is a conductor-insulator-semiconductor structure that employs intracell charge transfer and charge injection to achieve an image sensing function. The significant temporal noise sources are Johnson noise in the distributed resistance of array conductor materials (e.g., polysilicon), in the channel resistance of MOS multiplex selection switches, and in the video amplifier. A source of both temporal noise and fixed pattern noise is array dark current and the spatial variation in this dark current, respectively. There are two factors that lead to very low dark current in CID arrays. First, only the charge storage region, which is smaller than the photosensitive region, is depleted and contributing significant dark current. The second factor is that the interface states that result in surface leakage current are continuously quenched under normal imager operat-ing conditions with the result that only the depleted periphery of each charge storage site contributes surface leakage. Since good quality charge transfer structures are normally surface leakage current dominated, this is an important effect. The major sources of fixed pattern noise are different with the various readout techni-ques. Sequential injection is a destructive readout technique that automatically cancels multiplexer interference and array fixed pattern variations. The fixed pattern noise mea-sured on a random access array operating with sequential injection readout was approximately a factor of 1000 below the saturation signal, lower than the temporal noise. With Pre-injection Readout, multiplexer switching noise couples directly into the video signal bus leading to a high level of fixed pattern noise. Since this component of fixed pattern noise repeats during each line of video, it is readily cancelled with the aid of one line of video storage. Fixed pattern noise after this type of cancellation has been measured as a factor of 250 below saturation. Row Readout operates through the driving of one set of array lines to cause signal charge to transfer to the orthogonal set of array lines with the result that switching noise is greatly suppressed. Raw fixed pattern noise levels of approximately three percent of saturation have been measured. In a low video rate device, fixed pattern noise was suppressed to below one part in 30,000 of saturation with this readout method. The extremely low-loss non-destructive readout capability of the CID structure results in the capability to read imager signals repeatedly for summation in external memory. Since the temporal noise that accompanies the signal sums incoherently, the signal-to-noise ratio improves in proportion to the square root of the number of readout operations. Noise levels below 100 carriers per pixel have resulted with this technique. No intrinsic limitation on the dynamic range improvement that can be obtained with this technique has been identified. External limitations, such as the time available to perform repeated non-destructive readout operations, have prevailed to date.