The overall design constraints have been investigated in detail in order to optimise the number and optical design of each viewing direction, the choice of wavelength bands, detection systems, detector sampling strategies, basic angle, metrology system, satellite layout, and orbit (Mérat et al. 1999). The resulting proposed payload design (Fig. 2) consists of:

(a) two astrometric viewing directions. Each of these astrometric instruments comprises an all-reflective three-mirror telescope with an aperture of $1.7\times0.7$ m², the two fields separated by a basic angle of 106 $^\circ$ . Each astrometric field comprises an astrometric sky mapper, the astrometric field proper, and a broad-band photometer. Each sky mapper system provides an on-board capability for star detection and selection, and for the star position and satellite scan-speed measurement. The main focal plane assembly employs CCD technology, with about 250 CCDs and accompanying video chains per focal plane, a pixel size 9 $\mu$ m along scan, TDI (time-delayed integration) operation, and an integration time of $\sim$ 0.9 s per CCD;

(b) an integrated radial velocity spectrometer and photometric instrument, comprising an all-reflective three-mirror telescope of aperture $0.75\times0.70$ m². The field of view is separated into a dedicated sky mapper, the radial velocity spectrometer, and a medium-band photometer. Both instrument focal planes are based on CCD technology operating in TDI mode;

(c) the opto-mechanical-thermal assembly comprising: (i) a single structural torus supporting all mirrors and focal planes, employing SiC for both mirrors and structure. There is a symmetrical configuration for the two astrometric viewing directions, with the spectrometric telescope accommodated within the same structure, between the two astrometric viewing directions; (ii) a deployable Sun shield to avoid direct Sun illumination and rotating shadows on the payload module, combined with the Solar array assembly; (iii) control of the heat injection from the service module into the payload module, and control of the focal plane assembly power dissipation in order to provide an ultra-stable internal thermal environment; (iv) an alignment mechanism on the secondary mirror for each astrometric instrument, with micron-level positional accuracy and 200 $\mu$ m range, to correct for telescope aberration and mirror misalignment at the beginning of life; (v) a permanent monitoring of the basic angle, but without active control on board.

The accuracy goal is to reach a 10 $\mu$ as rms positional accuracy for stars of magnitude V=15 mag. For fainter magnitudes, the accuracy falls to about $20{-}40~\mu$ as at V=17-18 mag, and to $100{-}200~\mu$ as at V=20 mag, entirely due to photon statistics. For V<15 mag, higher accuracy is achieved, but will be limited by systematic effects at about $3{-}4~\mu$ as for V<10-11 mag. Raw data representing the star profile along scan must be sent to ground. An integral objective of the mission is to provide the sixth astrometric parameter, radial velocity, by measuring the Doppler shift of selected spectral lines. Colour information is to be acquired for all observed objects, primarily to allow astrophysical analyses, though calibration of the instrument's chromatic dependence is a key secondary consideration.

The astrometric accuracy can be separated into two independent terms, the random part induced by photo-electron statistics on the localisation process accuracy, and a bias error which is independent of the number of collected photons. The random part decreases in an ideal system as N^-0.5, where N is the number of detected electrons per star; the bias part is independent of N, represents the ultimate capability of the system for bright stars, is limited by payload stability on timescales shorter than those which can be self-calibrated, i.e. shorter than about 5 hours.

GAIA will operate through continuous sky scanning, this mode being optimally suited for a global, survey-type mission with very many targets, and being of proven validity from Hipparcos. The satellite scans the sky according to a pre-defined pattern in which the axis of rotation (perpendicular to the three viewing directions) is kept at a nominally fixed angle $\xi$ from the Sun, describing a precessional motion about the Solar direction at constant speed with respect to the stars. This angle is optimised against satellite Sun shield demands, parallax accuracy, and scanning law. Resulting satellite pointing performances are determined from operational and scientific processing requirements on ground, and are summarised in Table 5.

A mission length of 5 years is adopted for the satellite design lifetime, which starts at launcher separation and includes the transfer phase and all provisions related to system, satellite or ground segment dead time or outage. A lifetime of 6 years has been used for the sizing of all consumables.

Table 5: Summary of the scanning law and pointing requirements. 0.05 Hz is the maximum frequency that can be identified after measurement post-processing
Parameter Value

Satellite scan axis tilt angle 55 $^\circ$ to the Sun

Scan rate 120 arcsec s^-1

Absolute scan rate error 1.2 arcsec s^-1 (3 $\sigma$ )

Precession rate 0.17 arcsec s^-1

Absolute precession rate error 0.1 arcsec s^-1 (3 $\sigma$ )

Absolute pointing error 5 arcmin (3 $\sigma$ )

Attitude absolute measurement error 0.001 arcsec (1 $\sigma$ )

High-frequency disturbances:

power spectral density at 0.05 Hz $\le1000~\mu$ as² Hz^-1

for f > 0.05 Hz decreasing as f^-2

**Table 5:** Summary of the scanning law and pointing requirements. 0.05 Hz is the maximum frequency that can be identified after measurement post-processing
Parameter	Value
Satellite scan axis tilt angle	55 $^\circ$ to the Sun
Scan rate	120 arcsec s^-1
Absolute scan rate error	1.2 arcsec s^-1 (3 $\sigma$ )
Precession rate	0.17 arcsec s^-1
Absolute precession rate error	0.1 arcsec s^-1 (3 $\sigma$ )
Absolute pointing error	5 arcmin (3 $\sigma$ )
Attitude absolute measurement error	0.001 arcsec (1 $\sigma$ )
High-frequency disturbances:
power spectral density at 0.05 Hz	$\le1000~\mu$ as² Hz^-1
for f > 0.05 Hz	decreasing as f^-2

4.2 Optical design

The astrometric telescopes have a long focal length, necessary for oversampling the individual images. A pixel size of 9 $\mu$ m in the along-scan direction was selected, with the 50 m focal length allowing a 6-pixel sampling of the diffraction image along scan at 600 nm. The resulting optical system is very compact, fitting into a volume 1.8 m high, and within a mechanical structure adapted to the Ariane 5 launcher. Deployable payload elements have been avoided. System optimisation yields a suitable full pupil of $1.7\times0.7$ m² area with a rectangular shape. Optical performances which have been optimised are the image quality, characterised by the wave-front error, and the along-scan distortion, avoiding at the same time a curved focal plane in order to facilitate CCD positioning and mechanical complexity. The optical configuration is derived from a three-mirror anastigmatic design with an intermediate image. The three mirrors have aspheric surfaces with limited high-order terms, and each of them is a part of a rotationally symmetric surface. The aperture shapes are rectangular and decentered, while each mirror is slightly tilted and decentered.

The tolerable optical distortion arises from the requirement that any variation of scale across the field must not cause significant image blurring during TDI operation. The number and size of the CCDs has been determined to match the optical quality locally in the field.

The monochromatic point spread function, $P_\lambda(\xi,\eta)$ , at a specific point in the field, is related to the corresponding wavefront error map w(x,y) in the pupil plane through the diffraction formula. The overall wavefront error of the telescope is the sum of the errors arising from optical design, alignment, and polishing residuals for the three mirrors. The design target of $\lambda/50$ rms over the whole field corresponds to a Strehl ratio of 0.84 at 500 nm. From analysis performed using the optical design software package Code V, alignment errors can be made negligible (wave front error $<\lambda$ /70 rms) provided that the mirrors are positioned with an accuracy of about $\pm1~\mu$ m. A 5 degree-of-freedom compensation mechanism with this accuracy (not considered to be excessively stringent with piezo-type actuators) is therefore implemented on the secondary reflector of each of the astrometric telescopes. This allows optimization of the overall optical quality in orbit as a result of on-ground residual alignment errors, and the recovery of misalignments of the telescope optics which may be induced by launch effects, even if all the mirrors are randomly misaligned by an amplitude $\pm50~\mu$ m in all directions. The required wavefront error measurement will be performed on at least three points of the field of view. In practice, the astrometric performance is not strongly dependent on the actual telescope wavefront error, since the effect of aberrations corresponds to first order in an energy loss in the central diffraction peak, which is the only part of the point spread function used for the star localization.

Although the optical design only employs mirrors, diffraction effects with residual (achromatic) aberrations induce a small chromatic shift of the diffraction peak. The chromaticity image displacement depends on position in the field, and on the star's spectral energy distribution (colour), but not on its magnitude. One purpose of the broad-band photometric measurements within the main field is to provide colour information on each observed object in the astrometric field to enable this chromaticity bias calibration on ground. Recent developments made on ion beam polishing have shown that polishing errors can be made practically negligible ( $\lambda$ /100 rms obtained on a SiC reflector of about 200 mm diameter). It is therefore likely that the chromatic shift can be reduced below a few tens of $\mu$ as over the whole field, easing calibration requirements. Combination of the satellite Sun shield and internal baffling reduce straylight to negligible levels.

$\begin{figure} \par {\epsfig{file=h2529f3.eps,width=8.8cm} }\end{figure}$

Figure 3: Operating mode for the astrometric field CCDs. The location of the star is known from the astrometric sky mapper, combined with the satellite attitude. A window is selected around the star in order to minimise the resulting read-out noise of the relevant pixels

4.3 Astrometric focal plane

The focal plane contains a set of CCDs operating in TDI (time-delayed integration) mode, scanning at the same velocity as the spacecraft scanning velocity and thus integrating the stellar images until they are transferred to the serial register for read out. Three functions are assigned to the focal plane system: (i) the astrometric sky mapper; (ii) the astrometric field, devoted to the astrometric measurements; (iii) the broad band photometer, which provides broad-band photometric measurements for each object. The same elementary CCD is used for the entire focal plane, with minor differences in the operating modes depending on the assigned functions.

The astrometric sky mapper detects objects entering the field of view, and communicates details of the star transit to the subsequent astrometric and broad-band photometric fields. Three CCD strips provide (sequentially) a detection region for bright stars, a region which is read out completely to detect all objects crossing the field, and a third region which reads out detected objects in a windowed mode, to reduce read-out noise (and hence to improve the signal-to-noise ratio of the detection process), and to confirm objects provisionally detected in the previous CCD strips, in the presence of, e.g., cosmic rays. Simulations have shown that algorithms such as those developed for the analysis of crowded photometric fields (e.g. Irwin 1985) can be adapted to the problem of on-board detection, yielding good detection probabilities to 20 mag, with low spurious detection rates. Passages of stars across the sky mapper yield the instantaneous satellite spin rate, and allow the prediction of the the individual star transits across the main astrometric field with adequate precision for the foreseen windowing mode.

The size of the astrometric field is optimised at system level to achieve the specified accuracy, with a field of 0. $^\circ$ 5 $\times$ 0. $^\circ$ 66. The size of the individual CCD is a compromise between manufacturing yield, distortion, and integration time constraints. The pixel size is a compromise between manufacturing feasibility, detection performances (QE and MTF), and charge-handling capacity: a dimension of 9 $\mu$ m in the along-scan direction provides full sampling of the diffraction image, and a size of 27 $\mu$ m in the across-scan direction is compatible with the size of the dimensions of the point spread function and cross scan image motion. In addition, it provides space for implementation of special features for the CCD (e.g. pixel anti-blooming drain) and provides improved charge-handling capacity. Quantitative calculations have demonstrated that the pixel size, TDI smearing, pixel sampling, and point spread function are all matched to system requirements. The CCDs are slightly rotated in the focal plane and are individually sequenced in order to compensate for the telescope optical distortion. Cross-scan binning of 8 pixels is implemented in the serial register for improvement of the signal-to-noise ratio.

Each individual CCD features specific architecture allowing measurement of stars brighter than the normal saturation limit of about V=11-12 mag: selectable gate phases allow pre-selection of the number of TDI stages to be used within a given CCD array. The resulting astrometric error versus magnitude shows the effect of this discrete selection (Fig. 4).

$\begin{figure} \par\epsfig{file=h2529f4.ps,width=8.8cm}\end{figure}$

Figure 4: Nominal accuracy performance versus magnitude, for a G2V star. For V< 15 mag the astrometric performance improves because the number of detected photons increases until the detector saturation level is reached ( $V\sim$ 11.6 for G2V star). Brighter than this, the performance is practically independent of magnitude, due to the pre-selection of the number of TDI stages required to avoid saturation

At the apparent magnitude and integration time limits appropriate for GAIA most of the pixel data do not include any useful information. There is a clear trade-off between reading too many pixels, with associated higher read-noise and telemetry costs, and reading too few, with associated lost science costs. This contributes to the choice of on-board real-time detection, with definition of a window around each source which has sufficient signal to be studiable, and determination of the effective sensitivity limit to be that which saturates the telemetry, and which provides a viable lower signal. Combining all these constraints sets the limit near V=20 mag, resulting in an estimated number of somewhat over one billion targets.

The broad-band photometric field provides multi-colour, multi-epoch photometric measurements for each object observed in the astrometric field, for chromatic correction and astrophysical analysis. Four photometric bands are implemented within each instrument.

4.4 Spectrometric instrument

A dedicated telescope, with a rectangular entrance pupil of $0.75\times0.70$ m², feeds both the radial velocity spectrometer and the medium-band photometer: the overall field of view is split into a central $1^\circ\times1^\circ$ devoted to the radial velocity measurements, and two outer $1^\circ\times1^\circ$ regions devoted to medium-band photometry. The telescope is a 3-mirror standard anastigmatic of focal length 4.17 m. The mirror surfaces are coaxial conics. An all-reflective design allows a wide spectral bandwidth for photometry. The image quality at telescope focus allows the use of $10\times10~\mu$ m² pixels within the photometric field, corresponding to a spatial resolution of 0.5 arcsec.

The radial velocity spectrometer acquires spectra of all sufficiently bright sources, and is based on a slit-less spectrograph comprising a collimator, transmission grating plus prism (allowing TDI operation over the entire field of view) and an imager, working at unit magnification. The two lens assemblies (collimating and focusing) are identical, compensating odd aberrations including coma and distortion. The dispersion direction is perpendicular to scan direction. The overall optical layout is shown in Fig. 5. The array covers a field height of 1 $^\circ$ . Each $20\times20~\mu$ m² pixel corresponds to an angular sampling of 1 arcsec and a spectral sampling of approximately 0.075 nm/pixel. The focal plane consists of three CCDs mechanically butted together, each operated in TDI mode with its own sequencing, providing read-noise as low as 3 e^- rms with the use of a dedicated "skipper-type'' multiple non-destructive readout architecture with 4 non-destructive readout samples per pixel.

The requirements for the CCDs are very similar to those of the astrometric field, including the use of TDI, and dedicated sky mapper. The photometric bands (Fig. 1) will require filters to be directly fixed onto the CCD array.

4.5 Science data acquisition and on-board handling

Preliminary investigations have been carried out to identify the minimum set of data to be transmitted to ground to satisfy the scientific mission objectives; to identify some on-board data discrimination compression principles able to provide the targeted data compression ratio; to assess the feasibility and complexity of implementing such compression strategies and related algorithms on board; to assess the resulting compressed data rate at payload output, which are used for the sizing of the solid state memory and communication subsystem; and to derive preliminary mass, size, and power budgets for the on-board processing hardware. For estimating telemetry rates (Table 6), a specific spatial sampling of the CCD data has been assumed. This sampling is not yet optimised and final, but represents a useful first approximation.

The instantaneous data rate will primarily fluctuate with the stellar density in each of the three fields of view, which scale with Galactic latitude. On-board storage will store a full day of observation for downlink at a higher rate during ground-station visibility. Including overhead, the total raw science data rate is roughly a factor 7 higher than the mean (continuous) payload data rate foreseen in the telemetry budget ( $\sim$ 1 Mbit s^-1). Data compression will reduce this discrepancy, but there remains roughly a factor two to be gained either by smarter CCD sampling, or by increasing the link capacity.

4.6 CCD details

CCD detectors form the core of the GAIA payload: their development and manufacture represents one of the key challenges for the programme. In the present study, consideration was given to the requirements on electro-optical behaviour; array size; buttability; pixel size; bright star handling; serial register performance; output amplifiers; power dissipation in the image zone, serial register, and output amplifier; trade-off between QE and MTF; photo-response non-uniformity; dark current; conversion factor and linearity; charge handling capacity per pixel; charge transfer efficiency in the image zone and serial register; minimization of residual images; anti-blooming efficiency; and packaging. The present baseline design is summarised in Table 7.

For astrometric use CCD accuracy depends essentially on the integral of QE $\times$ MTF over the wavelength band. CCD MTF must be optimised in parallel with the QE. The QE and MTF values for the CCD optimised for the astrometric field have been used in the detailed astrometric accuracy analysis. The pixel size, nominally adopted as $9\times27~\mu$ m² for the astrometric field, is an important design parameter. A smaller pixel size would decrease the telescope focal length, as well as the overall size of the astrometric focal plane assembly, and consequently the overall size of the overall payload. However, such devices provide worse performance in a number of other areas. The trade-off between QE, MTF, and charge-handling capacity results in design reference values, and bread-boarding activities are underway to verify these performances in detail.

Table 6: Average stellar flow in the various fields of the astrometric and spectrometric instruments, and the resulting average telemetry rates. A limiting of G=20 mag is assumed for the astrometric instrument (AF and BBP) and for the medium-band photometer (MBP), and G=17 mag for the radial-velocity spectrometer (RVS). It is assumed that each sample represents 16 bits of raw data. The resulting raw data rates are before compression and do not include overhead
Parameter Astro-1 and 2 (per instrument) Spectro

ASM3 AF01-16 AF17 BBP SSM1 MBP RVS

Limiting magnitude, $G_{\rm max}$ [mag] 20 20 17

Average star density, $N_{\rm s}$ [deg^-2] 25000 25000 2900

TDI integration time per CCD, $\tau_1$ [s] 0.86 3.0 30

Field width across scan, $\Phi_y$ [deg] 0.66 1.0 1.0

Star flow through $\Phi_y$ , $f=N_{\rm s}\Phi_y\omega$ [s^-1] 550 833 97

Number of CCDs along scan, $N_{\rm CCD}$ 1 16 1 4 1 14 1

Solid angle of CCDs, $\Omega$ [deg²] 0.019 0.302 0.019 0.077 0.100 1.400 1.000

Number of stars on the CCDs, $N_{\rm s}\Omega$ 473 7568 473 1892 2500 35000 2900

Readout rate, $R=N_{\rm s}\Omega/\tau_1$ [s^-1] 550 8800 550 2200 833 11662 97

Samples per star read out 25 6 30 16 or 10 42 14 930

Samples per star transmitted, n 25 6 30 10 42 8 930

Raw data rate, 16nR [kbit s^-1] 220 845 264 352 560 1494 1443

Raw data rate per instrument [kbit s^-1] 1681 3496

Total raw data rate [kbit s^-1] $2\times1681+3496=6858$

**Table 6:** Average stellar flow in the various fields of the astrometric and spectrometric instruments, and the resulting average telemetry rates. A limiting of G=20 mag is assumed for the astrometric instrument (AF and BBP) and for the medium-band photometer (MBP), and G=17 mag for the radial-velocity spectrometer (RVS). It is assumed that each sample represents 16 bits of raw data. The resulting raw data rates are before compression and do not include overhead
Parameter	Astro-1 and 2 (per instrument)	Spectro
	ASM3	AF01-16	AF17	BBP	SSM1	MBP	RVS
Limiting magnitude, $G_{\rm max}$ [mag]	20	20	17
Average star density, $N_{\rm s}$ [deg^-2]	25000	25000	2900
TDI integration time per CCD, $\tau_1$ [s]	0.86	3.0	30
Field width across scan, $\Phi_y$ [deg]	0.66	1.0	1.0
Star flow through $\Phi_y$ , $f=N_{\rm s}\Phi_y\omega$ [s^-1]	550	833	97
Number of CCDs along scan, $N_{\rm CCD}$	1	16	1	4	1	14	1
Solid angle of CCDs, $\Omega$ [deg²]	0.019	0.302	0.019	0.077	0.100	1.400	1.000
Number of stars on the CCDs, $N_{\rm s}\Omega$	473	7568	473	1892	2500	35000	2900
Readout rate, $R=N_{\rm s}\Omega/\tau_1$ [s^-1]	550	8800	550	2200	833	11662	97
Samples per star read out	25	6	30	16 or 10	42	14	930
Samples per star transmitted, n	25	6	30	10	42	8	930
Raw data rate, 16nR [kbit s^-1]	220	845	264	352	560	1494	1443
Raw data rate per instrument [kbit s^-1]	1681	3496
Total raw data rate [kbit s^-1]	$2\times1681+3496=6858$

A "worst-case'' star density, corresponding to about 2.8 10⁶stars per square degree (about 19-20 mag in Baade's Window) has been used in a detailed analysis of CCD performance. The total noise per sample includes contributions from the CCD read-out noise at the relevant read frequency, analog-to-digital conversion noise, and the video chain analog noise.

All CCDs used for both astrometric and photometric measurements are operated in the drift-scan or TDI (time delay and integration) mode. That is, charge packets are gradually built up while transferred from pixel to pixel at the same rate as the optical image moves across the detector. Centroiding on the digitised output provides a measure of the position of the optical image relative to the electronic transfer along the pixel columns. Demands on the precision of the mean image position are strict: a standard error of 10 $\mu$ as in the final trigonometric parallax translates to a required centroiding precision in the astrometric field of 36 nm, 1/250 of a pixel.

Specific laboratory experiments, using the $13~\mu$ m pixel EEV device CCD42-10 in windowing mode, non-irradiated as well as irradiated at doses of up to 5 10⁹ protons cm^-2, have been conducted in TDI mode, using different illumination levels, and at different CCD operating temperatures. Although not fully representative of the flight configuration, and while not yet fully evaluated, these experiments have demonstrated that the targetted centroiding accuracy appears to be achievable.

A key parameter for achieving a high degree of reproducibility is the Charge Transfer Efficiency (CTE) of the CCD. In the present context it is more convenient to discuss the Charge Transfer Inefficiency (CTI) $\varepsilon=1-\mbox{CTE}$ . Very few charge carriers are actually lost (through recombination) during the transfer process; rather, some carriers are captured by "traps'' and re-emitted at a later time, thus ending up in the "wrong'' charge packet at the output; if short-time constant processes dominate, the main effect observed is that of image smearing. The CTI has an effect both on the photometric measurement (by reducing the total charge remaining within the image) and the astrometric measurement (by shifting charges systematically in one direction). CTI during parallel transfer is particularly critical, since it affects the astrometric measurements in the direction where the highest precision is required, i.e. along the scan. In addition, CTI is worse along-scan due to the lower transfer rate. The magnitude of the problem can be crudely estimated as follows: assume a constant fraction $\varepsilon$ of the charge is left behind while the fraction $1-\varepsilon$ flows into the next pixel. The expected centroid shift is $\simeq N\varepsilon/2$ pixels. The CCDs in the astrometric field of GAIA have N=2780 pixels of size 9 $\mu$ m along the scan. Assuming $\varepsilon = 10^{-5}$ results in a centroid shift of 125 nm or about 500 $\mu$ as.

More careful appraisal of the CTI effects on the astrometric accuracy show the effect after calibration is negligible for the the undamaged (beginning-of-life) CCD, but potentially serious for the degraded performance that may result after significant exposure to particle radiation in orbit. Although most of the CTI effects can be calibrated as part of the normal data analysis, stochastic effects related to the charge losses can never be eliminated by clever processing. Extensive laboratory experiments are underway to quantify the amplitude of these residual effects.

Table 7: Summary properties of the GAIA CCDs in the two astrometric telescopes (QE = quantum efficiency; MTF = modulation transfer function; CTI = charge transfer inefficiency; RON = read-out noise)
Feature Details

Array size $25\times58$ mm² active area

Pixels per CCD 2150 cols $\times$ 2780 TDI stages

Dead zones top: 0.25 mm; sides: 0.6 mm;

bottom: <5 mm

Pixel size in image zone $9\times27~\mu$ m²

Phases in image zone 4

Pixel size in serial register $27\times27~\mu$ m²

Phases in serial register 4

Device thickness 10-12 $\mu$ m

Si resistivity 20-100 $\Omega$ cm

Buried channel n-type channel

Oxide thickness standard

Anti-blooming shielded at pixel level

Notch channel implanted for all CCDs

Output amplifiers 2 per device, 2-stage

Conversion factor between 3-6 $\mu$ V/e^-

Additional gates 5-10

Power dissipation <560 mW

Non-uniformity <1% rms local

<10% peak-to-peak global

Mean dark current <0.5 e^-s^-1pix^-1 (200 K)

Non-linearity <1 per cent over 0-2 V;

<20 per cent over 2-3.5 V

CTI in image area <10^-5 at beginning-of-life;

$\sim10^{-4}$ after major Solar flare

CTI in serial register <10^-5 at beginning-of-life;

$\sim5.10^{-4}$ after major flare

Quantum efficiency trade with MTF and RON

MTF at Nyquist frequency trade with QE and RON

Read-out noise trade with QE and MTF

**Table 7:** Summary properties of the GAIA CCDs in the two astrometric telescopes (QE = quantum efficiency; MTF = modulation transfer function; CTI = charge transfer inefficiency; RON = read-out noise)
Feature	Details
Array size	$25\times58$ mm² active area
Pixels per CCD	2150 cols $\times$ 2780 TDI stages
Dead zones	top: 0.25 mm; sides: 0.6 mm;
	bottom: <5 mm
Pixel size in image zone	$9\times27~\mu$ m²
Phases in image zone	4
Pixel size in serial register	$27\times27~\mu$ m²
Phases in serial register	4
Device thickness	10-12 $\mu$ m
Si resistivity	20-100 $\Omega$ cm
Buried channel	n-type channel
Oxide thickness	standard
Anti-blooming	shielded at pixel level
Notch channel	implanted for all CCDs
Output amplifiers	2 per device, 2-stage
Conversion factor	between 3-6 $\mu$ V/e^-
Additional gates	5-10
Power dissipation	<560 mW
Non-uniformity	<1% rms local
	<10% peak-to-peak global
Mean dark current	<0.5 e^-s^-1pix^-1 (200 K)
Non-linearity	<1 per cent over 0-2 V;
	<20 per cent over 2-3.5 V
CTI in image area	<10^-5 at beginning-of-life;
	$\sim10^{-4}$ after major Solar flare
CTI in serial register	<10^-5 at beginning-of-life;
	$\sim5.10^{-4}$ after major flare
Quantum efficiency	trade with MTF and RON
MTF at Nyquist frequency	trade with QE and RON
Read-out noise	trade with QE and MTF

4.7 Payload summary

(a) two identical astrometric telescopes:

-	fully-reflective 3-mirror SiC optics,
-	separation of viewing directions: 106 $^\circ$ ,
-	monolithic primary mirrors: $1.7\times0.7$ m²,
-	field of view: 0.32 deg²,
-	focal length: 50 m,
-	wavelength range: 300-1000 nm,
-	4-colour broad-band photometry,
-	operating temperature: $\sim$ 200 K,
-	detectors: CCDs operating in TDI mode,
-	pixel size along-scan: 9 $\mu$ m;

(b) spectrometric instrument:

-	spectrometer for radial velocities,
-	11 colour medium-band photometry,
-	entrance pupil: $0.75\times0.70$ m²,
-	field of view: 4 deg²,
-	focal length: 4.17 m,
-	detectors: CCDs operating in TDI mode.