LSST Camera CCDs produced by the manufacturer e2v exhibit strong and novel residual charge images when exposed to bright sources. These manifest in images following bright exposures both in the same pixel areas as the bright source, and in the pixels trailing between the source and the serial register. Both of these pose systematic challenges to the Rubin Observatory Legacy Survey of Space and Time instrument signature removal. The latter trail region is especially impactful as it affects a much larger pixel area in a less well defined position. In our study of this effect at UCDavis, we imaged bright spots to characterize these residual charge effects. We find a strong dependence of the residual charge on the parallel clocking scheme, including the relative levels of the clocking voltages, and the timing of gate phase transition during the parallel transfer. Our study points to independent causes of residual charge in the bright spot region and trail region. We propose potential causes in both regions and suggest methodologies for minimizing residual charge. We consider the trade-offs of these methods including decreasing the camera's full well and dynamic range at the high end. The voltage scheme in the main camera was altered to address this effect accordingly.
]]>The NSF-DOE Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) depends on the ability to detect astronomical sources at faint signal levels over a wide field on the sky [1]. The system will image objects as faint as 24.5 magnitude in a 30 s exposure over 9.6 square degrees, and will inevitably image bright objects above the dynamic range of the camera. LSST will regularly include bright stellar images and bright transient objects like satellites [2] that exceed the saturation point of the Rubin Observatory LSST Camera (LSSTCam) charge-coupled devices (CCDs). This type of exposure has associated sensor effects that must be characterized and mitigated in the survey for the faint science programs to succeed.
One systematic sensor effect common to CCDs is residual images seen in exposures immediately following a bright or overexposed image. These image "ghosts" are a well documented phenomena. The mechanism through which this occurs is fairly straightforward: when CCDs are exposed to a bright object, not all of the electrons produced are captured in the pixel charge packet or read out in that image. This charge instead persists within the CCD after the image is read out. On some time scale which is long relative to a single readout, the charge then thermally leaks back into the pixel's potential well and is read out in subsequent exposures as ghosts of the bright objects [3,4].
More specifically, under some operating conditions it is possible for charge to reach the siliconsilicon oxide (Si-SiO 2 ) interface at the CCD surface where the charge can become trapped. Residual charge images occur when the parallel clocking barrier and collecting voltage levels are configured in such a way that stored charge reaches the surface before it overflows into neighboring pixels in the process called "blooming". The higher collecting voltage level (parallel high), if set high enough, pulls electrons towards the surface. The relative level of the barrier voltage (parallel low) then determines the maximum amount of charge that can be stored in a given pixel before it blooms, which we call the "blooming full well". Increasing the number of electrons in a given pixel will also push charge packets further towards the surface. If the full well is large enough that charge reaches the surface, we call this the "surface full well condition".
In LSSTCam CCDs there are notable residual charge images in CCDs manufactured by e2v (now Teledyne-e2v). Residual images have not been observed to the same degree in CCDs manufactured by -1 -Imaging Technology Laboratory (ITL). The ITL CCDs are also operated at lower nominal parallel clock levels than e2vs which may account for some of this difference in residual charge [5]. Residual charge images were first seen in the main LSSTCam CCD images by Doherty et al. in 2014 [6] in prototype CCDs and later rediscovered during integration and testing at the SLAC National Accelerator Laboratory in the main LSST Camera.
We show that there are multiple mechanisms that enable residual charge between images in LSSTCam: one that allows charge to be trapped at the same pixels as the imaged bright object, and one that allows charge to be trapped during the parallel transfer when charge packets are shifted toward the serial register. These result in distinct manifestations of residual charge images as shown in figure 1. LSSTCam CCDs all have sixteen separate segments each with their own readout amplifier. The images shown in this study are all images from single amplifier segments, not entire CCDs, as the effects discussed are all confined within a given segment. Important residual image features have been labeled with numbers. 1: The bright spot image. The larger halo is caused by out of focus reflections in our system, although these are much too dim to create residual charge in subsequent images. 2: A bright trail read out after the spot in the same image. This trail is caused by charge leaking back into those pixels as they pass through the spot region as they are parallel shifted during readout. 3: The residual image seen in the spot region in subsequent images. As the bright spot was imaged, charge is trapped at the CCD oxide surface. This charge then leaks back into the pixels during subsequent integration causing this residual image. 4: The residual charge trail seen in subsequent images following a bright image. Charge here is trapped at the surface during the parallel transfer.
In addressing these residual charge sources we explore the parameter space of the parallel clocking scheme, how it affects these residual images. Specifically, we explore how the effect responds to changes -2 -in the parallel transfer voltage levels and timing. We discuss the associated downsides which include excessive leakage current when parallel clock lower voltage is brought too low (less than -6.0 V) and changes to the dynamic range that correspond to changes in the full well level when the parallel clock swing, that is the voltage difference between the high collecting and low barrier voltages, is decreased.
In the main survey, we will always know the precise location of bright stars that will cause residual charge effects. This means that residual images in the central spot region in the following images (labeled '3' in figure 1) will always be in a known location. The effect can then either be removed when processing the images using calibrated data for that sensor, or the spot region can be removed from the survey entirely. The latter option will only affect saturating stars which should number somewhere in the order of 50 million total in the southern sky [7]. Assuming these appear as ∼10 pixel point sources bloomed to 10 times it size, on average we will see 25,000 of them in each LSSTCam image which will cover around 2.5 million pixels. This is only 0.08% of the 3.2 Gigapixel camera.
The trail effects on the other hand (labeled '4' in figure 1) are a more serious issue. If left undressed the trail would greatly impact source extraction and photometry by injecting linear features on the scale of 10 electrons above the sky background. Simply removing the involved pixels from the same and next image in the survey requires removal of every CCD column that includes a saturating source. Not accounting for situations when the bright source columns overlap in the same amplifier segment, this amounts to about 5% of the survey affected by residual image trails from bright stars.
In addition to bright stars, there is also the growing issue of other bright objects which have less well defined locations such as low earth orbit satellites (LEOSats). There are on the order 1 million satellites currently filed for launch during the survey [8]. Satellites which bloom LSSTCam pixels will likely be seen in every image and unlike point sources, will transit 12 entire CCDs [2]. The mitigation and removal of residual charge due to LEOSats is then an important issue that demands attention both in the central bright image region and especially in the trail effects that they will cause which will cover large swathes of the camera focal plane.
We examined this effect at the UC Davis CCD Laboratory which is equipped with the LSST 𝑓 /1.2 Beam Simulator [9] and a science grade LSSTCam e2v CCD250 with the same readout electronics used in the main camera [10].
A standard way to diagnose residual charge images in CCD cameras is to use a flat-field to uniformly overexpose the CCD to a signal level greater than saturation [11]. Then the residual charge can be measured in following dark exposure images as a uniform offset from the nominal readout. This method is a great way to determine the residual charge in the region where an object is imaged (the spot region labeled '1' in figure 1), but it would fail to pick up effects like the residual charge trail we see in LSSTCam e2v CCDs.
We instead use an imaged spot rather than a flat field which allows us to differentiate between the two regions. Our "standard dataset" is as follows. We take a series of 20 bias images and then expose the sensor to a bright spot source around 4 times the full well saturation with nominal settings in a 15 s exposure. We then read out that spot image and immediately follow it with twenty 15 second dark images. We can then look at each subsequent dark to see trends in residual images over time. We repeat this standard dataset 50 times for each set of parameters in order to drive up our signal to noise. We keep exposure times at 15s to mimic the cadence of the LSST survey. This
means that residual charge images and their scaling to specific parameters will act like those seen in the main survey. It also gives sufficient time resolution to examine the decay of residual charge, even if a longer 30s cadence is adopted for LSST.
Images are line-by line overscan subtracted and median bias subtracted. We then average over the 50 standard datasets taking the mean for each pixel. This gives us sub-electron precision. We then take the median value of all pixel in a region of interest (the spot, trail, etc.) which eliminates influence from outlier pixels.
We also report potential downsides to changes made to address the residual charge effect. While changing the clock timing and voltage levels showed no impact on the camera's point spread function (PSF), charge transport efficiency, or most other performance parameters, it does impact the camera's dynamic range via altering the full well level. We therefore report the median of the maximum readout achieved in the coadded bright spot area (labeled '1' in figure 1). This is a stand-in for our full well pixel value which is a measure of our system's dynamic range. This measure of full well is limited by potential variations in gain between different sets of operating parameters.
The parameter space we explore is configuring the clocking voltages and timings of the CCD readout. Parallel clocking voltage levels are directly related to residual charge [5]. We tested many clocking voltage schemes in our study. The voltages that are relevant to an understanding of our work are the three listed in table 1. The P-Up configuration uses the former nominal voltage configuration, but with both parallel voltages shifted higher. This allows us to exacerbate the residual charge image effects in our sensor which normally shows relatively small residual charge compared to some LSSTCam CCDs. In this way we greatly increase the signal to noise in our measurements. We also list the values used in the nominal settings before this work and the final new configuration which has been adopted for use in the main survey.
Table 1. Relevant voltage configurations used in probing residual charge in LSSTCam e2v CCDs. The nominal voltages are the ones used by the main camera before this work. The new configuration is the final one used in the main camera for the survey which was developed by the LSST Camera team to eliminate residual images by the mechanism described in this study. P-Up is the nominal configuration with the parallel voltages both raised in order to exacerbate the residual charge effect.
Nominal Voltages New Voltages P-Up Voltages Parallel High
We also do a number of tests where we keep most voltages constant at the P-Up or nominal values, but vary only the parallel high, or parallel low voltages. As shown in table 2, we change the parallel -4 -
voltages individually to quickly find direct residual charge voltage dependencies. These tests are useful in determining trends but they are not potential configurations for final camera operations.
Table 2. In order to measure the dependence of residual charge on parallel voltages, we leave all voltages at the nominal values and independently change the parallel high collecting voltage to the values in the first row. We then independently sweep through different parallel low values in the second row with all other values at the nominal. The third row is a voltage sweep where we start with the P-Up configuration and lower the collecting voltage.
We also alter the timing of parallel transfer which has notable impact on the residual charge effect.
In the nominal LSSTCam e2v CCD clocking scheme, when each of the four parallel clocking gates change phase from the low barrier phase to the high collecting phase, there is a short 2660ns period where three of the four gates are held at the collecting voltage at the same time before the next gate changes to the low barrier phase. The gate overlap period accounts for about 23% of the total transfer timing with the nominal readout timing. This standard parallel clocking scheme is shown in figure 2. This scheme gives the charge ample time to transfer, but may be related to the novel trail region -5 -residual charge. We do not know the exact mechanism that would cause this. One possible cause of this would be that, as the three clocks are held high, charge interacts more with the CCD surface. We change the clock timing to decrease the duration of this parallel clock overlap and measure its effect.
There are two useful frames in which to view residual charge. The first is the "single image frame" which is the measured residual charge in a given dark image. Unless otherwise noted, our reported values use this frame and report the value in the first 15s exposure following the bright triggering image. This frame is useful because it is the closest comparison to what we will see in actual LSST on-sky data for a given configuration. The other frame is the "total charge frame" where we instead add all the charges seen in images up to that time. This frame is useful for visualization purposes and for calculating thermal decay time scales for electrons returning from the surface in the silicon.
A plot of residual charge as a function of time for both frames is shown in figure 3. In this figure -6 -
we fit the residual charge with a sum of two exponential decays. There may be multiple electron trap impurities at the CCD surface, and these can have different characteristic decay times. It is important to note that it can take well over a hundred seconds (around ten 15s exposures) for the residual charge to dissipate from the surface to below our detection threshold. This decay rate, unlike the amount of charge trapped, is dependent on the kinetics of silicon at a given temperature and is not affected by changes in the clocking timing or voltages. We can also examine the structure of the residual charge across the affected rows and columns to produce figure 4. One can see that if we look along the affected columns, we can visibly identify four distinct regions which behave differently in figure 4:
(1) The region read out after (to the right of) the spot we see that the residual is not present.
(2) Next there is the spot region. The dotted lines identifying this region in figure 4 are assigned by finding the highest and lowest row where the source reaches saturation level. Towards the top and bottom of this region, there are bloomed pixels which reach the blooming full well condition but they are not actively being exposed to the bright source and may not result in the same residual charge effect. The spot region results therefore may be biased towards these low values. This is not true of the trail region (see below). Values for the spot and trail region are therefore not directly comparable as the algorithm to calculate them varies.
-7 -
(3) The trail region in reported values is selected by taking a window of all the rows below the spot region with a small, 10 pixel buffer on the edges and top to prevent biasing the data with the edges which may not be affected to the same degree. In data analysis, typically both the regions labeled "Trail Region" and "Buildup Near Serial" are combined when discussing the residual trail and readout values. They are separated here to show that there is a general fall off in residual charge farther from the serial register.
(4) The region of buildup near the serial region is treated as structure within the trail. The cause of this structure is not well understood. It is a low level effect that is not always detectable in all voltage configurations. The effect is likely related to an inefficiency in transferring charge to the serial region at some voltages.
As would be expected, the general trend is that decreasing the parallel high collecting voltage also decreases the amount of residual charge in images following a bright spot in both the spot and trail regions. This is consistent with the theory that decreasing the collecting voltage decreases the force pulling charge to the surface. Decreasing the parallel swing also decreases the residual charge seen. This is because the full well level is also decreased which allows more charge to bloom before reaching the surface. This effect is shown as a comparison of CCD images when changing the parallel swing via lowering the high collecting voltage in figure 5. This effect holds whether it is the barrier or collecting voltage that is changed, although the effect is of course greater when changing the collecting voltage.
As a direct comparison of how the residual charge scales when we change the parallel high collecting voltage compared to the parallel low barrier voltage, we use the data collected from the two voltage sweeps in table 2 where we vary the collecting and barrier voltages independently and keep all others at the nominal. As shown in figure 6, varying the collecting voltage has a much more pronounced effect on the measured residual charge. This is because while changing either alters the swing, changing the collecting voltage also directly alters how close the electron charge packet in a given pixel is pulled towards the surface.
Decreasing the collecting voltage also has the drawback that it lowers the amount of charge able to be collected in a given pixel. This lowering of the full well level means that the detector's dynamic range is decreased at the bright end. In figure 7 we show the results of collecting data from the P-Up parallel high sweep from table 2. We also report the average maximum pixel readout value as a stand-in for full well level.
Although the dynamic range of LSSTCam will be decreased by lowering the parallel clock swing, the rate at which the maximum pixel readout falls when decreasing the collecting voltage is linear, whereas the residual charge measured decreases exponentially. This is especially pronounced, as evident in figure 7, in the trail region which is the greater threat to the LSST's systematic limits. The loss in dynamic range will also be at the bright end. This is important because the main mission of the Rubin observatory is not to get better images of known bright objects, but rather to have a wide survey of dimmer objects and fainter transient objects than have been seen in previous surveys.
The camera voltage scheme was since changed from the nominal 9.3 V parallel swing to a lower 8.0 V parallel swing by decreasing the high collecting voltage to address this residual charge issue. Other voltages were changed in conjunction to arrive at the configuration called "new" in table 1. This -9 - change resulted in the virtual elimination of residual charge images in the trail region and a large decrease in the spot region. A thorough study still needs to be done using the main LSSTCam sensors to determine the exact quantitative outcomes. We expect the residual charge to scale similarly to that in our UC Davis system which demonstrated around an 80% decrease in the spot region. This is likely sensor dependent and our lab only has access to one e2v CCD.
Altering the parallel clock overlap time does have a strong effect on the residual charge seen in images. However it is not as straightforward as in the case of altering the parallel clocking voltages. Changing the overlap timing affects the residual charge in the spot and trail regions differently. In the spot region, decreasing the parallel overlap increases the residual charge measured, whereas in the trail region, the residual charge decreases as the overlap is decreased. In figure 8 we show these results which point to separate mechanisms that drive residual charge in the spot and trail region. We performed several tests at varying levels of parallel clock timing overlap. As shown in figure 9, the spot residual seems to follow an somewhat exponential pattern while the trail residual exhibits fairly logistic growth. In the trail's case we believe this is because whatever effect is causing the charge to be captured has some amount of overlap time required to fully take effect. After that time is provided the residual charge level plateaus.
The reason why the parallel clock timing overlap effects the residual charge images is not perfectly understood. One hypothesis is that holding three gates high increases the realized potential in the silicon and pulls the charge packet closer to the CCD surface during the transfer, causing charge to be captured during readout in the trail region. However, simulations of the charge packet position within the CCD using the Poisson_CCD package [12] show that charge does not move closer to the CCD surface when three parallel gates are held high. In fact, the charge packet moves further from the CCD surface as there is more area for the electrons to spread out.
This spreading out of charge may allow charges in the readout trail to interact more frequently with a spatially rare (on the scale of the saturated PSF) dilute family of traps along the surface. This would allow charge to interact with these rare traps over a wider area even if the charge packet as a whole is further from the surface.
We also operated the CCD in "scan mode" [13,14], which allows us to read out the raw video pulse for each pixel read into our readout electronics. We found that there is no significant change in the signal shape when we change or eliminate the parallel overlap. This indicates that it is not an -11 - issue of giving the charge packed adequate time to transfer. It was also considered that the additional residual charge in the trail region is caused by additional time that the charge packet spends in a given region of silicon. This is certainly not the case since the increase in time due to parallel clock overlap is not proportionally long enough.
There are a few remaining possibilities. It is possible that holding three gates high at the same time is forcing charge that was captured in the spot region out from the surface. That charge then does not all reach the charge packet and some of it is pulled along closer to the surface during readout. It is then captured in the surface in the trail region as the main spot charge packets are clocked towards the serial register. This would also explain why the residual charge in the spot area increases as the trail residual decreases.
Another possibility is that while the charge is not actually moving closer to the CCD surface when three gates are held high, the charge packet is oscillated towards and away from the surface when we change the number of gates held high. In the clocking scheme where only two gates are held high (no parallel overlap), this would not occur and the charge packet would remain at the same distance from the surface. If this is the case, that oscillation may actually cause some charge to oscillate too far and interact with the CCD Si-SO 2 surface.
-12 -As far as a practical measure to remove residual charge persisting in LSSTCam images in the survey, decreasing the parallel clock overlap is not as clearly effective as altering the parallel clock voltage levels. The increase in the spot region residual charge and the uncertainty of the root mechanism mean that there are more clear drawbacks than simply decreasing the dynamic range. There may be more unknown drawbacks if more extensive studies and sensor characterization is done with these modified settings.
LSSTCam e2v CCDs exhibit residual charge effects in images following exposure to bright sources. These manifest not just as ghosts in the same region of the focal plane as the source, but also as a trail between the source and the serial register. This indicates charge is being captured in the CCD and likely on the CCD's oxide surface layer interface not only during integration, but also during the parallel transfer and readout. Both these regions, but especially the trail residual charge images will create a large systematic error in the Rubin Observatory Legacy Survey of Space and Time if not addressed. The trail effect would inject linear contrasting features on the scale of 10 electrons above the sky background which would greatly impact source extraction and photometry.
Through our studies into potential causes of this systematic, we developed a technique for studying residual charge images using bright spot sources rather than flat-field images as is standard for this kind of study [11]. Using this methodology we determined that the residual charge is largely dependent on the levels of the voltages used for parallel transfer clocking, and the timing of the phase changes of gates during the parallel transfer. Specifically in this latter case, there is an overlap period where three gates are held at the high collecting voltage each time the gates change phase. Changing this overlap time affects the residual charge, and it affects the charge in the spot and trail regions differently indicating a new mechanism.
The alteration of parallel transfer clocking voltages has a well understood effect on residual charge images. We can greatly reduce or eliminate the residual charge seen in LSSTCam e2v images by decreasing the parallel transfer clock swing via lowering the high collecting voltage. This comes with the trade-off of decreasing the dynamic range of the instrument at the bright end.
Reducing or eliminating the parallel clock overlap time also significantly reduces the residual charge trail. However this comes not just with the loss of dynamic range of the camera, but also with an increase in residual charge seen in the same pixels as the initial bright source. The exact mechanisms that causes these effects is not understood, although it does point to the parallel clock overlap being a contributing factor in how charge is captured during the parallel transfer and the resulting residual charge trail.
Residual charge images will appear in the LSST at some level. They are very likely going to be below the detection threshold for a single image and they will not be in the same place for multiple images. As LSSTCam goes on-sky it will be necessary to verify that residual charge images are not a significant systematic error. This will require a study of bright source images and the following images in both the spot and trail regions. This has been dealt with in other surveys such as Pan-STARRS' burntool program which removes residual charge signatures using a one-dimensional exponential model [15].
Based on the results of this study it was decided by the LSST Camera team to lower the parallel clocking voltage swing from 9.3 V to 8.0 V via decreasing the high parallel transfer collecting voltage. More extensive studies are needed to characterize residual charge in the main camera. However
preliminary images show that this virtually eliminates residual charge in the trail region and greatly decreases it in the same region as the bright source. In our studies with the f/1.2 beam simulator we saw over an 80% decrease in residual charge in that same region for the same change in voltages.