We show here a small sample of the pre-flight calibration data that have been
obtained for CUBIC . Most of the data were obtained using a variety of
fluorescent X-ray lines excited by 
Po. A composite spectrum of
these lines, made with data collected using Amp 4 of the CUBIC6 CCD,
was shown in Figure 1 and is repeated here as
Figure 27.
These data were collected at an operating temperature of -83 C, the predicted
operating temperature of CUBIC . All of the data shown here were taken
from chip CUBIC6 using the same clock voltages that will be used in flight.
Similar data were collected from CUBIC7 and a smaller
data set has been taken from the backup CCDs.
Figure 28 shows the 
Fe spectrum (Mn K
and K
),
which is used
as a general calibration line because of its easy availability, and
which will be used on CUBIC for in-flight calibration.
Figure 2, repeated here as Figure 29 shows the FWHM resolution as a function of energy. The data points are taken from the lines shown in Figure 27, while the curves show the theoretical resolution for a silicon photoelectric detector.
The data are consistent with the resolution expected for a readnoise
of 4 e
rms. This is slightly higher than the readnoise measured
in blank images for this device, indicating a small noise contribution
from some other source, but the performance is excellent nevertheless.
The excellent linearity of the CCD detectors is demonstrated by Figure 30, which is based on the data of Figure 27.
The percent nonlinearity is shown in Figure 31, which shows the deviations of the points in Figure 30 from a straight line.
Figure 32 shows a plot of the noise level of the
CUBIC6 CCD as a function of its temperature. The noise at high temperatures is dominated by shot noise in the dark current, which falls off exponentially with T. At low temperatures, the noise is dominated by the read noise of the on-chip amplifier. This is the dominant noise contribution for temperatures below about -70 C. At our operating point of -83 C, we are well below the point where dark current is important for our integration time of 30 seconds.
Figures 33 and 34 show the centroid of
the Mn K
peak at 5.9 keV vs pixel number in the parallel and
serial direction (row number and column number). The small reduction
in gain along the rows and columns moving away from the output amplifier
is due to small charge losses during transfer to the output node. These
are the data used to derive the Charge Transfer Inefficiency (CTI), which
is defined as
where Q is the signal charge (in any units) and 
is the
number of pixels through which the charge has been transferred.
In general, the CTI is different in the serial and parallel registers
due to the different time scales of the serial and parallel clocks.
Figures 35 and 36 show the parallel and serial CTI vs. line energy. Because of the definition of CTI, it is expected to be strongly dependent on the line energy, or the initial charge packet size, Q.
To zeroth order, one might expect CTI to be inversely proportional to Q, but in fact the charge loss per pixel is not necessarily constant, leading to a more complex relationship between CTI and E.
Finally, we present plots of the charge loss, 
, vs energy in
Figures 37 and 38.
Interestingly, the behavior of the charge loss appears to be rather different in the serial and parallel registers. The charge loss in the parallel register appears to be nearly proportional to energy. The charge loss in the serial register, on the other hand, is nearly independent of energy. The one exception is the Fluorine line at 0.677 keV. However, this is probably deceptive. The Fluorine line is produced by a LiF target bombarded by alpha particles. We have discovered that this target crystal exhibits low level optical fluorescence, which provides a ``fat zero'', or low level charge over the entire device. This ``fat zero'' fills traps and reduces the charge loss for the X-ray events in these frames. Therefore, this line does not provide an accurate measurement of the CTI.
