The Gaia satellite is a high-precision astrometry, photometry and spectroscopic ESA cornerstone mission, currently scheduled for launch in 2012. Its primary science drivers are the composition, formation and evolution of the Galaxy. Gaia will achieve its unprecedented accuracy requirements with detailed calibration and correction for CCD radiation damage and CCD geometric distortion. In this paper, the third of the series, we present our 3D Silvaco ATLAS model of the Gaia e2v CCD91-72 pixel. We publish e2v's design model predictions for the capacities of one of Gaia's pixel features, the supplementary buried channel (SBC), for the first time. Kohley et al. (2009) measured the SBC capacities of a Gaia CCD to be an order of magnitude smaller than e2v's design. We have found the SBC doping widths that yield these measured SBC capacities. The widths are systematically 2 {\mu}m offset to the nominal widths. These offsets appear to be uncalibrated systematic offsets in e2v photolithography, which could either be due to systematic stitch alignment offsets or lateral ABD shield doping diffusion. The range of SBC capacities were used to derive the worst-case random stitch error between two pixel features within a stitch block to be \pm 0.25 {\mu}m, which cannot explain the systematic offsets. It is beyond the scope of our pixel model to provide the manufacturing reason for the range of SBC capacities, so it does not allow us to predict how representative the tested CCD is. This open question has implications for Gaia's radiation damage and geometric calibration models.

Gaia will only achieve its unprecedented measurement accuracy requirements with detailed calibration and correction for radiation damage. We present our Silvaco 3D engineering software model of the Gaia CCD pixel and two of its applications for Gaia: (1) physically interpreting supplementary buried channel (SBC) capacity measurements (pocket-pumping and first pixel response) in terms of e2v manufacturing doping alignment tolerances; and (2) deriving electron densities within a charge packet as a function of the number of constituent electrons and 3D position within the charge packet as input to microscopic models being developed to simulate radiation damage.

The Gaia satellite is a high-precision astrometry, photometry and spectroscopic ESA cornerstone mission, currently scheduled for launch in 2012. Its primary science drivers are the composition, formation and evolution of the Galaxy. Gaia will achieve its unprecedented positional accuracy requirements with detailed calibration and correction for radiation damage. At L2, protons cause displacement damage in the silicon of CCDs. The resulting traps capture and emit electrons from passing charge packets in the CCD pixel, distorting the image PSF and biasing its centroid. Microscopic models of Gaia's CCDs are being developed to simulate this effect. The key to calculating the probability of an electron being captured by a trap is the 3D electron density within each CCD pixel. However, this has not been physically modelled for the Gaia CCD pixels. In Seabroke, Holland & Cropper (2008), the first paper of this series, we motivated the need for such specialised 3D device modelling and outlined how its future results will fit into Gaia's overall radiation calibration strategy. In this paper, the second of the series, we present our first results using Silvaco's physics-based, engineering software: the ATLAS device simulation framework. Inputting a doping profile, pixel geometry and materials into ATLAS and comparing the results to other simulations reveals that ATLAS has a free parameter, fixed oxide charge, that needs to be calibrated. ATLAS is successfully benchmarked against other simulations and measurements of a test device, identifying how to use it to model Gaia pixels and highlighting the effect of different doping approximations.

A radically new CCD development by Marconi Applied Technologies has enabled substantial internal gain within the CCD before the signal reaches the output amplifier. With reasonably high gain, sub-electron readout noise levels are achieved even at MHz pixel rates. This paper reports a detailed assessment of these devices, including novel methods of measuring their properties when operated at peak mean signal levels well below one electron per pixel. The devices are shown to be photon shot noise limited at essentially all light levels below saturation. Even at the lowest signal levels the charge transfer efficiency is good. The conclusion is that these new devices have radically changed the balance in the perpetual trade-off between readout noise and the speed of readout. They will force a re-evaluation of camera technologies and imaging strategies to enable the maximum benefit to be gained from these high-speed, essentially noiseless readout devices. This new LLLCCD technology, in conjunction with thinning (backside illumination) should provide detectors which will be very close indeed to being theoretically perfect.