CCD Radiation Damage Amelioration
Sequencer Manipulations to Reduce CTI
This page reviews some sequencer manipulations appropriate for reducing
radiation damage CTI of CCDs for the ACIS, HETE, Astro-E systems. The examples
tend to be for the CCID-17 using ACIS electronics, but the similarity in
sequencer design should allow general use of the methods. One very important
difference between these instruments is the intended operating temperature
of the CCDs. Currently explored techniques are:
Reduced Parallel Transfer Speed (Image to Frame)
Artificial Charge Injection (Sacrificial Charge)
Trap Filling (Squeegee)
Reduced Parallel Transfer Speed
The typical timed exposure CCD readout involves the bulk transfer of
charge in the image array to the shielded frame store array in preparation
of the readout of the later. This transfer involves 1026 (CCID-17) parallel
row operations and typicaly requires 41 msec (ACIS). Reducing this
speed can be accomplished in too numerous to describe ways, most of
which require but trivial adjustments to the sequencer and need not
involve new SRAM primitives. In the case of the low energy proton damaged
device the
reduction in CTI is quite evident.
These data were made for 9.17 second readout times - somewhat longer
than the nominal 3.5 second ACIS readout rate. This is an unavoidable
consequence of slower clock speeds. Another consequence is the decrease
in collection area due to cosmic ray "blobs". If the smear time
(ratio of transfer time to readout time) of 10% can be tolerated
then a 24 x decrease in clock speed (for ACIS) should decrease the
CTI by 3 for a loss of 6% of collection area due to cosmic rays.
A briefly explored variation of slow readout involves "dancing" the
parallel clocks in a fashion familiar to proponents of "pocket-pumping".
This also improves the CTI - but not really any more than the method
illustrated above. In other words, it doesn't much matter how the
details of the slow transfer are facilitated.
Details of PRAM modifications:
Sacrificial Charge
The concept of sacrificial charge and its influence on CTI will be
familiar to the reader. The techniques described here artificially
place charge at predetermined locations in order to influence radiation
damaged devices. There are 3 convenient sources of charge which may
be employed: 1.) Natural (cosmic particle, x-ray, light leaks and dark
current) 2.) Spurious dark current produced by clocking 3.) Charge
injected via the output node. The later most is by the far the most
wonderful to use but is only convenient for Astro-e sequencer users.
The first 2 methods occur togther in varying degrees due to the
necessary clocking operations on the one hand and the unavoidable
presence of cosmic rays on the other.
Line Charge
The first technique involves placement of a single line (row) of
charge at a predetermined location in the image array prior to
the integration (exposure) period. Natural charge is collected
in the image and frame store immediately after a readout. This
charge is swept into the serial register which is held with all
phases high to allow even distribution of charge. The charge is
then moved backwards through the frame and image array to the
desired location where it produces a line prior to the start
of the next integration (exposure) period.
Here is what the readout looks like. This ccid-17 has been
damaged with 40 MeV protons. (ref)
Notice the line of charge at row 800 (center of damaged area)
and the cosmic rays, the charge from similar being the main
contributors to the charge in the line itself. In general there is
more than enough charge available from cosmic rays to generate a line
of nearly full wells. An extra integration time of 2 seconds is about
right but the actual time needs to be tuned for the background level.
This example comes from the laboratory environment which has a lower
background level than expected on orbit. A severe drawback with this
technique is the loss of integration time during the frame-store
readout. This means the readout time will be longer than the exposure time
or, alternately put, the instrument will not have a 100% on time.
This sequencer modification requires additional SRAM to open the serial
register and to move charge backwards in the parallel direction. Care must
also be exercised in matching various SRAM segments to avoid the case of
all three parallel phases momentarily in the same state. For this reason
the actual number of SRAM segments may increase. When charge is injected
via the output node the creation of the line charge is simplified
but the on time (duty cyle) still suffers. One definite advantage of
output node charge injection is the ease with which multiple lines
can be produced.
The immediate consequence of the line charge is a restoration (partial?)
of the amplitude of subsequent pixels. The degree of this improvement
(for indeed it is such) depends upon the temperature as does the
duration in time for which the effect of the line charge persists.
The parallel transfer rate in this case is the standard 40.00 microsecond
per row. The interpretation of this amusing graph is left as an
exercise for the student.
Charge Sweeping (Squeegee)
The squeegee technique preserves a small number of rows, the charge in these
rows is swept across the image array from top to bottom during the image
to frame transfer, and then swept from bottom to top prior to exposure. Any
charge in these rows is never removed except to fill traps. All the mechanisms
which contribute to the artificial charge injection apply here. This technique
does not impact the readout duty cylce. Experimentation shows that some number
greater than 2 rows is needed, but even 2 rows does something. The success of
this method depends upon long time scale traps (greater than readout time)
which seems to be the case. Here is an example at -113 c:
In this case 48 rows were reserved to collect charge for the squeegee. The
elimination of the long time scale traps has a lot to do with improving the
resolution. It is rather a pity that the reserved rows occur at the center
of the focal plane. It remains to be determined what the minimum number of
required rows is.