The CCD Nitty-Gritty Ditty

Science Nugget: December 1, 2000

Introduction

This, my latest installment on space hardware issues, talks about the gizmo on board Yohkoh that generates the lovely images of the sun we've been talking about in many of these science nuggets. Let's have a quick glance at the latest full-disk image (at the time of this writing!) to remind ourselves what we're trying to accomplish here, before this nugget gets too thick and everyone bails out and heads to their nearest Dilbert site:

The gizmo in question, for this nugget, is referred to as a Charge-coupled device or, more typically, a CCD detector. In this nugget we'll explore how it is that CCDs work, how they age, what goes wrong with them, and specifically what's the current state of the union for the Yohkoh CCD.

How CCDs work, in three paragraphs or less

CCDs start their lives as large silicon wafers in a semiconductor factory. Through the appropriate use of a selection of extremely toxic chemicals, this wafer ends up with an extensive checkerboard pattern of pixels. The bar napkin illustration is as follows:

This bar napkin was stolen wholesale from one of the big CCD experts in the field, Jim Janesick of Pixelvision, Inc. The pixels are illustrated with analogous buckets, collecting rain (or, in reality, photons). What happens is photons land in each bucket, and knock loose electrons in the CCD via the Photoelectric Effect. Then we have to somehow corral all of these electrons and shuffle them off the CCD in a controlled fashion. Armed with our cattle prod, here's how we do it:

  1. First, the CCD detector is exposed for some amount of time. For a CCD like the one used on Yohkoh, we meter the exposure with a shutter mechanism. This allows some nominal quantity of photons to land on the CCD and kick free some corresponding quantity of electrons. In this illustration, we have buckets to contain the electrons. In a real CCD, special doping within the silicon substrate serves to keep the electrons isolated to individual columns, and special electronics keep the electrons isolated to a specific pixel within a column.
  2. After the shutter is closed, we begin clocking the image off of the CCD. The first clock cycle is to the vertical clock: All the pixel columns are shifted down by one, and the bottom row falls into the horizontal register.
  3. The horizontal register is then clocked off to the output amplifier, or the big graduated rain bucket near the bottom left of our image.
  4. The vertical clocks are again triggered to shift another row of pixels off into the horizontal register.
Through the creative use of traces on the surface of the CCD, the application of a careful selection of voltages to specific areas of each pixel, and the use of special dopants in the silicon wafer, we're able to shift electrons from their chosen pixel off the CCD, with efficiencies that border on magic.

CCDs, radiation and aging

CCDs that are launched into space are subject to a pretty harsh environment. The geometry of the individual pixels is pretty small, on the order of 12 microns in modern CCDs. (This number ranges as low as maybe 3 microns, and ranges as high as 25 microns or larger.) The silicon, the dopants, the interfaces between the gates and the substrate, and the number of charge carriers in the substrate are all carefully controlled during manufacture. It is once this chip is launched into space that we start having difficulties. There are numerous radiation sources that we have to cope with.

The South Atlantic Anomaly

The problem with putting stuff in space is the volume of energetic particles one has to deal with. One major problem we face numerous times a day is the South Atlantic Anomaly, or SAA. That first URL describes this as a sort of pothole in the Earth's magnetic field that collects energetic particles. These energetic particles (10 MeV protons, according to the second URL) wreak havoc in all aspects of space hardware, ranging from the high voltage supplies to the flight computer to the CCD detector. Any one pass through the SAA does little to the CCD, but many passes, over a period of years, will methodically trample on the nice crystalline structure of all that silicon in the chip, degrading performance.

Another concern is the Van Allen Radiation Belts. These belts, located in rings around the North and South poles. Yohkoh doesn't pass through these belts, but other satellites, such as the Chandra X-Ray Observatory and the Transition Region and Coronal Explorer pass through them on a regular basis. Chandra initially had difficulties with the Van Allen belts as the instrument was typically bore-sighted right through these radiation zones, and the energetic protons were focussed directly onto the CCDs. The energy of these protons was sufficient to displace about 12 silicon atoms per incident proton! The charge-transfer efficiency (I'll describe this below) of some of the CCDs was weakened pretty significantly by this process until it was worked out what was going on. TRACE passes through the radiation belts 4 times every 90 minutes or so. The CCDs survive this just fine, but TRACE images incur a pretty heavy penalty during these passes.

Solar Flares

Another source for potential CCD damage would be solar events, discussed in myriad detail in these science nuggets. Recent nuggets demonstrate significant enhancements of the particle flux at Earth.

Incident X-ray radiation

Finally, it's important to note that bit of the electromagnetic spectrum that Yohkoh was designed for: X-Rays. X-rays are also capable of damaging CCD detectors, so we go to special lengths (that I won't discuss here) to try to preserve and protect the CCD for as long as possible.

Dark Current

One of the prime methods we have for exploring the nature of the SXT CCD is to study "dark current" images. Dark current is spurious charge that builds up in each pixel on the CCD during exposure. This is a function that is dependent on the substrate, the type of CCD, the mode it's operated in, the temperature it's operated at, how badly it has degraded over time, and a host of other things. (A quantified version of this can be found here). First, let's have a look at the data. Here's two images from the Yohkoh dark current database. For the sake of the more intense data aficionadoes, the specific images are: 
   0 22-DEC-91  21:21:05  QT/H  Open /Al.1  Half Dark C  23  2668.0  512x512
   1 12-SEP-00  17:32:02  QT/H  Open /AlMg  Half Dark C  23  2668.0  512x512
What I've selected here are two random dark current images: one from shortly after launch, and one from a couple of months ago as of this writing. The resolution and exposure times are the same, and while the filters are different, that's OK since the shutter is closed during the exposure and there shouldn't be any light getting to the CCD. In both cases, the CCD temperature was -21 C. Without further ado, the images:
22-Dec-199112-Sep-2000

If one places any creedence in a statistical sample of two images, things have indeed changed in the intervening 9 years since launch. For a real study, I'd use the entire database, but for a short study like this, two images serve our purposes just fine. These images are displayed with the same scaling, so it's immediately apparent that the more recent dark image has greater dark current. We can also look for other things like hot pixels and streaks in the images. If the 1991 image were scaled more aggressively, we'd also see streaks in it as well. There are also streaks associated with hot pixels. The two kinds I can think of are:

  1. Shorted pixels: these continuously leak charge. They give rise to a bright column of pixels running the entire length of the CCD. Maybe there's a few here; I think I see one column that runs through the first dash in the "12-Sep-00" label in the rightmost image.
  2. Pixels with bad Charge-Tranfer Efficiency, or CTE. These pixels have a tough time moving charge from one pixel to the next, and so they tend to leak their charge slowly over time rather than dumping it all at once. They show themselves as a bright pixel with a long decreasing trail behind (er, to the top) of them. I don't see any obvious candidates in these images.

Another thing we can note, especially in the 12-Sep-2000 image, is a "ramp" in the dark current. This is because we're still accumulating dark current as the image is being transferred off of the CCD. Thus, at the end of, say, a nominal 10-second exposure, we then have to shift the entire image off the bottom of the CCD. So the first few rows have comparatively little dark current, but the rows of pixels near the top of the image spend an additional 8 or so seconds waiting to be clocked off the CCD, and thus collect more spurious electrons. Hence: the dark current ramp.

So, that's some basic stuff we can garner from staring at the images. Let's look at the statistics now. First, a histogram of the CCD dark current for both frames:

22-Dec-199112-Sep-2000

OK, now we're getting somewhere. It's apparent that the dark current has about doubled over time, and there's a lot more spread in it now. (I think this spread can partially be attributed to that "ramp" we discussed earlier.) There's also a bunch more hot pixels, so there's just a lot more of everything.

Finally, let's look at a slightly more complicated plot that tells all about CCDs:

22-Dec-199112-Sep-2000

This plot is a standard tool that CCD folks use to evaluate devices, though normally this is done under much more controlled conditions using, say, an Iron 55 X-ray source. (Discussions of why that's cool is beyond the scope of this nugget, which is already beyond the scope of a normal SXT nugget!)

So what are we looking at? The X-axis is the column number across the chip. The Y-axis is recorded CCD signal. Then we plot the value of each pixel on the CCD: each column on this plot has 512 points plotted on it. We can see first-hand the distribution across the chip, and the distribution of pixels. So what can we see here?

First, the spread in dark current is evident. Also note the uneven baseline in signal in the 12-Sep-2000 baseline: humps in the dark current exist at rows 100 and 400, approximately. These very likely correspond approximately to the location of the solar limb on the CCD, and have impacted the dark current there. (I think.) Also note the relatively small dark current for a few pixels at the far right of both images: This is the charge that is stored up in the horizontal register when the image is collected. The spread in that distribution is the read noise for the CCD. It's increased over time. My back-of the envelope calculation gives a read noise in 1991 of about 3 DN. In 2000, I get a read noise of about 7.2 DN. (I got this by taking the standard deviation of rows 500-512. I'm not convinced I know what I'm talking about here.)

Some concluding thoughts

We've really only scratched the surface with this basic analysis, but in the spirit of keeping this as simple as possible, this is a good stopping point. While at first glance it seems alarming that the dark current has increased so dramatically, it's important to realize that the dark current is also very consistent from image to image, and thus dark current corrections generally work extremely well even with a degraded CCD. It's even possible to do science with a warm CCD, as we've previously discussed during a CCD bake-out.

December 1, 2000

Brian Handy <handy@isass0.solar.isas.ac.jp>