Hydra Dark Current II: Keeping the dark current low (2001)

2001 August 31
Roger Smith
(action items in green)

Signal to noise in the Hydra Bench Spectrograph is often limited by the noise floor of the detector, which is critically dependent on dark current. Binned by 2 in one or both dimensions, is commonly used to improve the signal to noise ratio or the read time. One must remember that while this helps overcome electronic noise it doesn't help overcome the dark current: binning the signal also bins the dark current.

We have to concern ourselves with all sources of "unwanted signal". Noise in the absence of any astronomical signal is determined by the summation in quadrature of the noise associated with

  • CCD and electronics
  • Thermal dark current
  • Charge injected into CCD
  • Spurious charge generated by clocking
  • and light leaks.


Definition of Terms

Thermal Dark Current

This is the charge which accumulates per unit time as some electrons in the valence band gain enough energy to overcome the bandgap and jump up into the conduction band. As the detector is cooled the probability that a valence band electron acquires enough energy will drop so that the dark current drops exponentially with temperature. At least this is the case for electrons in a perfect silicon crystal.

In a real device there are imperfections, impurities and crystal dislocations which provide intermediate energy states so that a valence electron can get to the conduction band in two smaller energy steps, increasing the probability of a "thermally generated" conduction band electron. These "hot pixels" produce less dark current with temperature too.

The density of crystal imperfections is greatest at the interface with the oxide layer. Dark current here is suppressed by taking the electrodes negative enough to pull "holes" out of the channel stops (ion implants that create potential barriers between columns). These holes populate the surface of the CCD just under the oxide layer, while the signal charge is repelled to a layer deeper in the CCD. The holes recombine with electrons generated thermally at the surface so that they never reach the storage well.

So, dark current is controlled by lowering the CCD temperature and by maintaining all clock phases negative during integration (MPP mode). The minimum temperature reached by the CCD mount is about 160-165K depending on ambient temperature (radiant load). This should be low enough to ensure good dark current performance. However it was found that dark current always exceeded that expected for the mount temperature measured. The most likely explanation for this is that the CCD is not making good enough thermal contact with the mount and that its temperature is being increased by the radiant load. Fixing this appeared to involve building a new detector mount. Rather than do this the Nitrogen flask is evacuated to promote evaporation/boiling which continues until the remaining liquid solidifies. The temperature drops by 15 degrees in the process and reduces the dark current to an acceptably low level.

Charge injection

Charge injection is the leakage of charge into the image area from the periphery of the CCD during the erase cycle. It can't be erased away since it is smeared across the chip by the erase process. Instead we can clock in reverse at the end of the erase cycle to push it back towards the edge where it originates. I mention it here because it is a serious problem in some of the CCDs in Mosaic which are of the same type, though much less so in this CCD.

Spurious Charge

Spurious charge is the small amount of charge generated by clocking the CCD. The mechanism is impact ionization: the surface just under the oxide separating the electrodes from the depletion region are populated by holes (due to the negative level on the clocks). This is good since these holes recombine with dark current generated as valence band electrons are excited to conduction band more easily by hopping to the intermediate energy levels produced by interface defects. However as these holes flow back into the channel stops during the rising edges of the clocks they have a low but finite probability of delivering energy to valence electrons which once liberated are collected like signal electrons. At the clock voltages needed to achieve full well, one sees about 2 electrons of spurious charge per pixel implying the probability of generating the spurious electron is about 1 in 2000 per shift.

When high gain is sleeted in the setdetector menu, the waveform compiler embeds commands in the new waveform macro which reduces the high levels of the parallel clocks. This reduces the spurious charge to an acceptable level at the expense of full well. When the high gain setting is selected the well capacity is reduced to 15000 e-. At 0.84e-/ADU this represents 18000 ADU, but when binned 2x2 exceeds the ADC range by about 10% so the loss in well capacity will often not be noticed.

Light Leaks

Obviously light leaking into the spectrograph masquerades as dark current and has exactly the same effect. While light leaks are in principal very simple to understand there is an extensive set of tips in the section on Eliminating Light Leaks, which you should read if you detect a problem.

Check for light leaks (etc) before each observing block !....

It is most likely that to maintain full sensitivity of the spectrograph it will be necessary to verify that all of these "dark signals" are within normal bounds prior to each observing block. To do this take the following full frame exposures with binning set to 2x2.

  • 3 zeroes with CCD cap on; quadproc (or ccdproc) these to overscan subtract and trim; median filter (imcombine)
  • 600 second "object" exposure with slit cap interlude and the detector cap removed, all lights in dome and surrounding rooms off. Subtract overscan and trim (quadproc). Note that you should not use a "dark" exposure since the camera will not see light leaks when the camera shutter is closed, since the camera has shielding to protect it from stray lights and reduce phosphorescence in the corrector during Nitrogen refills.
  • Subtract the combined zero from the object; use implot and the "s" key to measure dark current, dodging cosmic rays. If you have time and want to increase the sensitivity of this measurement, you can take a sequence of 3 darks; ccdproc and median filter them. With cosmic rays thus removed you can average many lines in implot before measuring the mean using the "s" key or even "fly blind" and imstat the whole frame. If you use imstat, be careful not to include the 20 or so edge columns which can contain line start transients.
  • Look at the difference image with Ximtool, to make sure there are no hot edges or corners.
  • Repeat the object frame(s) with the slit still capped but lights on in the dome, corridor outside the bench spectrograph room, and in the neighboring rooms to ensure that it is safe to take darks when people are working in the dome. Process and review identically.

If the dark signal is abnormally high, even when dome and surrounding room lights are off, check that no abnormal or unauthorized objects have been left in the room which may be phosphorescing. For example the paint on some nitrogen dewars may glow in the dark, too faintly to see by eye but enough to affect the CCD. Unless you find a simple explanation like this, don't waste time guessing about the cause: you should use a cooled CCD and camera lens to image the bench spectrograph room.


What is an acceptable dark current and light leak?

Recall that

Noise_Floor = SQRT{ read_noise2 +
      Xbin*Ybin*(Spurious_charge + [Dark_Current + light_leak]*time) }

Read Noise (in the overscan) is 3.0 e- for the high gain setting. Assuming that the highest binning factor is 2x2, then we require:

Spurious_charge << read_noise2 / (Xbin*Ybin)
  << 3.02 / (2*2)
  << 2 e-/unbinned_pixel

and if the longest exposure time is half and hour, then we require:

            + Light_leaks
<< read_noise2 / (Xbin*Ybin*time)
  << 3.02 / (2*2*0.5)
  << 4.5 e-/unbinned_pixel/hour


Actual performance

  • Spurious charge is only ~ 0.25 e- / unbinned_pixel thanks to the lowered upper voltage rail for the parallel clocks.
  • Dark current < ~0.7 e-/pix/hr, when detector has been protected from bright lights for at least a day.
  • Light leak should be no greater than dark current, and ideally be undetectable. At present (Aug 2001), this is only achieved with the lights in the dome and surrounding rooms off. We are currently working on fixing this both for robustness and the ability to work in the dome while darks are being taken.

When all is working correctly,

Noise_Floor = SQRT{ read_noise2 + Xbin*Ybin*(Spurious_charge + Dark_Current*time) }
  = SQRT{ 3.02 + 2 * 2 *( 0.25 + 0.7 * 0.5 ) }
  = 3.4 e-


How to Eliminate Light Leaks 

This has not been as simple as one might think! Here are some tips....

True leaks:

  • We are susceptible to leaks at such a low level that significant sources will be impossible to see by eye, so intuition is often wrong!
  • The black cloth around the spectrograph attentuates the light but it is not completely opaque.
  • Adhesive tape may dry out and/or come loose, so you can't assume that the previous fix will be permanent.
  • Many types of tape including our favorite high quality black tape is in fact translucent in the visible and/or become transparent beyond the visible range where the CCD is still very sensitive.
  • Metal backed tape is opaque for sure but the adhesive is sometimes not strong enough to overcome the intrinsic stiffness of the tape itself.
  • The metal baffles around the camera stop scattered light and light leaks from reaching the CCD directly but of course light can still enter through the optical path.
  • Use "object" frames, not "dark" frames for testing since you can only see stray light in the optical path when the shutter is open, since the shutter is just in front of the camera corrector.
  • Never leave even the red lights on in the vestibule are or behind the sealed door that leads to the back of the telescope.
  • Always close the door to the corridor.
  • Plug the hole where the cable tray passes from the vestibule to the elevator entrance ("M" floor).
  • There are currently leaks under the access doors, so that the warning light (for the detector cap) in the corridor can be easily seen by the CCD camera even when both the inner and outer door are closed.
  • There is what appears to be a leak at roof level into the dome or mezzanine room above the entrance door.

False leaks:

  • Light coming down the fibers during testing must be blocked. It will overwhelm the dark current even with dome lights off. A manually installed metal box (felt lined) has been made to cover the slit assembly. Remember to remove it when you finish!

Light sources to eliminate within the spectrograph:

  • LEDs in Motor Controller (done)
  • LEDs in control box for detector cap (done)
  • LED in new Quicksilver shutter motor ...to be checked.
  • Smoke detector (electronics emit light, not just its LED)
  • Arcon data fiber connectors must be encased and sealed at the Arcon and the cable breakout box. This is probably the weakest link since any work on Arcon will potential disturb this work. Imaging these fibers while live requires that a second working fiber be connected to the Arcon which is taking the pictures. To do this one has to feed the Arcon on the hydra bench from a dummy source, such as the Mosaic Trambox. The Arcon doesn't have to be loaded with code and functioning. It just needs to be powered on and have a live fiber connected.
  • Arcon fiber leakage. Only use black fibers. All other colors leak a lot of light. Check that the fiber cladding is in good shape along its whole length. A damaged cladding can leak copious amounts of light. (I've seen this happen.)
  • There are preflash LEDs in the detector cap which are very useful for checking that the CCD is functioning without needing a working spectrograph. If there is leakage current while they are off, this can appear as excess dark current. The power supply has been modified so the preflash circuit has a shorting switch across the LEDs to shunt the current source to ground as well as a switch in series, to eliminate this effect. Beware: if an unmodified supply is installed the problem will occur again.....one will see dark current increase when the detector is capped!


Many materials absorb light then re-emit it over a long period of time. This can cause a dark current which decays over time and is thus harder to pin down. It is therefore important to not leave any item inside the room unless it is essential and has been proven to be free from phosphorescence. These are some examples which have been addressed:

  • Fluorescent tubes (removed)
  • White paint on fixtures for fluorescent tubes (removed)
  • Smoke detector (Removed)
  • Gray paint on metal power strip (over-painted black)
  • White paint on bulkhead near base of telescope (painted black)
  • Linoleum floor ? .... maybe not, could be reflected light
  • Oscilloscope screen ...of course.
  • White paint on some dewars. (Store dewar and all tools in anteroom)
  • Corrector glass (shield built to minimize illumination by flashlights.) Need to verify that bright comparison lamp or quartz spectra don't cause a remnant signal.

Finding unwanted light sources

If no obvious source is found after consulting the above checklists, don't waste time guessing. Use a cooled CCD camera fitted with a camera lens and integrating for 2-5 minutes to identify light sources. The following setup is very effective:

  • SITe2K camera, e.g. T2K5 or T2K6.
  • Wide angle lens such as Zeiss distagon 40mm f/4, with the custom Hasselblad lens adapter for standard dewar.
  • Or if you happen to be able to borrow a 16mm fisheye, with the custom 35mm lens adapter.
  • Little Gray Box for controlling shutter.
  • Use lab jacks to adjust tilt if necessary. Or one could use the optoliner cart from La Serena as a "tripod".
  • Single channel readout, binned 4x4, 3-5 minute exposure

Sometimes it can be difficult to figure out exactly what you are looking at in the noisy dark images. In this case it helps to take a short exposure without moving the camera leaving a low light source on in the room. Use the smallest aperture and low light such as leaving the door open into the anteroom. Be careful not to be confused by reflections from the metallic surfaces or by the light emitted by the arcon you are using. Its data fibers need to be taped over and it needs to be wrapped in black cloth. To avoid spending all day in the elevator, get an assistant at the keyboard downstairs or setup a laptop on the network to provide a local terminal for controlling Arcon.



The following notes were made during commissioning
and are left here for completeness
though some of the information is duplicated.


  • The lens cap and camera baffle/cover are necessary to avoid phosphorescence or image remnance effects in the CCD or its window. Values of ~5e-/pix/hr from the detector (window?) are easily achieved by exposing the detector to a flashlight briefly. I have not tested this aggressively since I didn't want to compromise the observer.
  • The detector cap plus baffle are sufficient to make it safe to adjust the spectrograph using flashlight for illumination without risk of compromising the observing. The curtain gives an extra margin of safety for cryogen refills.
  • However the Schmidt corrector (or surrounding cable?) is mildly phosphorescent. If hit with a flashlight will increase dark current to 3-10 e-/pix/hr decaying over an hour or so to negligible levels. The plan is to divide the camera baffle into two overlapping parts, one serving to protect the corrector from accidental illumination. [Done]
  • The test has not been performed to see if doing flats is enough to cause significant phosphorescence by the corrector. I suspect this wont be a problem given based on my flashlight test: a series of images alternating between darks and flats should be done to test for this.
  • The LEDs in the filter "windmill" which illuminate the slit were shown to turn off fully now that a relay had been installed.
  • A conventional dark frame is an exposure with the shutter closed. Since the shutter is immediately in front of the spectrograph camera, the dark frame does not reveal light leaking into the optical path. A cap is needed over the slit to permit measurement of dark current plus light leaks. [Manually installed cap has been made] (Even when the dome lights are off, a significant amount of light comes down the fibers.)


The room

  • Detector limited dark current is achieved whether or not the dome lights are on provided that the baffle is installed and provided that shutter is closed. (Detector uncapped and schmidt corrector not exposed to light for hours).
  • If the slit is wrapped in Aluminum foil and black cloth, one can achieve detector limited dark current even with the shutter open, but only with done lights off. When dome lights are on the "dark current" increases about 10 fold. Hence there is a leak into the spectrograph room.
  • Removing the baffle over the spectrograph camera increases "dark current". This test was done before I had determined that light leaked into the room from the dome so it is not clear whether the light blocked by the baffle was coming from the dome or the Arcon.
  • The black tape which seals up the doors and various feedthroughs from the dome is actually translucent. The room needs to be imaged with Arcon plus camera lens needs to be used to image the room to see whether the black tape needs to be replaced or augmented.


  • The communication fibers produce a huge amount of light (~850nm) where they enter the Arcon front panel. A rough metal box with holes for the cables to pass though was made, but this requires large amounts of black tape and black cloth to seal it. A more precisely made cover is needed to provide an adequate seal while being easily de-mounted to provide access for maintenance.
  • There are 6 LEDs inside the detector cap, controlled by the preflash parameter in the setdetector menu. These allow one to verify CCD performance. Since it is also useful to be able to check dark current with the cap on, the fact that there is leakage current through the LEDs when off is a problem. A transistor needs to be added to the LED drive circuit in the Arcon power supply to shunt the current past the LEDs when off. [In progress] This is preferable to a relay since the LED pulse must be very short.
  • The white plastic diffuser in front of these LEDs was removed since it was found to phosphoresce (tens of e-/pix/hr) with seemingly infinite decay time. This effect is clearly distinct to the LED leakage current problem as it is present when the LED cable was unplugged.
  • The dark current is seen to rise after the Nitrogen is refilled. This is because the addition of liquid Nitrogen to the pumped (solid) Nitrogen raises the temperature of the heat sink for the detector, for an hour or so, until the pumping (and consequent accelerated boil off) cools and re-solidified the Nitrogen. It should be noted that dark current is so critical in this application that the CCD is being operated at the lowest temperature achievable (about 150K): CCD temperature is not actively stabilized. The extra 16 K or so of cooling proved by pumped Nitrogen, is possibly due to poor thermal contact between the CCD and its supporting plate due to differential contraction between the Al plate and the stainless steel bolts which pass through it. Spring washers may be difficult ot retrofit due to the cramped space, and may be insufficient. Since the pumping and refill process are working well it has been decided to continue with this solution. Therefore, vacuum lines need to be routed to the large mechanical pump located in the corridor behind the telescope, instead of using the small pump currently located on the cass elevator platform directly under the telescope. [Done]
  • Not related to darks but important: I need to figure out why the combination of gain=1 and Xbin=2 causes a strong pattern in the images.