Breaking In a Different CCD Camera for the JEOL JEM 1010

image of replacement cameraThe Gatan Ultrascan 890 on the JEOL JEM 1010 (with a 794/20 controller box) started to produce horribly noisy images in late January, 2016. After lengthy trouble shooting and numerous false leads, we determined that the controller box (and not the actual camera) was at fault. Unfortunately, this particular camera is somewhat of an odd beast (a 4k x 4k CCD but run by a 794/20 controller box that is almost always described as the controller for a 2k x 2k MegaScan camera). Gatan only produced about 80 such cameras (so there are essentially no "spare parts" sitting around), and Gatan also no longer supports camera models this old (>10 years). With help from people in the cryoEM community, Gatan and JEOL (especially Wim Hagen at EMBL, Joe Mulqueen with Gatan and Matthew Foil with JEOL), we were able to obtain an older model replacement camera (a MegaScan 794) and to install it on the 1010.

The replacement camera has a 1k x 1k CCD with 24 μm pixels (compared to the original 4k x 4k CCD with 15 μm pixels) and while the field of view of the replacement camera is smaller than that of the original, it is still about 40% of the original camera's field of view because of the larger pixels. Images showing a field of view comparable to the UltraScan 890 but recorded with this replacement camera should be acquired at half the magnification used with the original camera.


After the camera was installed, we realized (again after some false starts) that we needed to adjust the way we acquired gain references. The target for the original camera had been around 10,000 counts/pixel, but when we did this with the new camera, we saw the following horrible results when recording images of a blank field as a function of electron dose:

blank field recorded at 84.2 pA/cm^2 and a 1 s exposure blank field recorded at 70.5 pA/cm^2 and a 1 s exposure blank field recorded at 55.7 pA/cm^2 and a 1 s exposure

While the image on the left may be close to the saturation level of the replacement camera's sensor, the other two images are not saturated (and even lower electron doses look like the image on the right). Not only do the images show horrible artifacts, the artifacts show an odd contrast reversal (the large black regions turn white) as the dose is lowered. We even determined that careful choice of electron exposure could come close to eliminating these artifacts:

blank field recorded at 50.2 pA/cm^2 and a 1.0 s exposure blank field recorded at 50.2 pA/cm^2 and a 1.1 s exposure blank field recorded at 50.2 pA/cm^2 and a 1.2 s exposure
blank field recorded at 50.2 pA/cm^2 and a 1.3 s exposure blank field recorded at 50.2 pA/cm^2 and a 1.4 s exposure blank field recorded at 50.2 pA/cm^2 and a 1.5 s exposure

The image recorded using an exposure time of 1.4 s is not perfect, but the fact that we could get an image that appeared this good gave us hope that we could figure out the problem.

We knew that the black (or white) artifacts in these images were not bad pixels in the CCD since an image with no illumination did not show them. Except for the observation that we could not make the largest of the artifacts vanish with new gain references, these artifacts behave essentially as dirt on the sensor would (and we looked into ways to clean the sensor). We further played with different electron doses and exposure times, and convinced ourselves that the problem was simply the number of electrons that the CCD was trying to record. It finally occurred to us that the black/white contrast reversal in the artifacts was happening right around an electron dose that matched the target of 10,000 counts/pixel set for the gain reference.

It also occured to us that the gain reference was being recorded at an electron dose that might be too close to the saturation point of the CCD (this 14 bit dynamic range camera should saturate at 16,000, but actual images seem to have fewer than ~14,000 counts/pixel). Indeed, for the UltraScan 4000 on the JEOL JEM 3200FS, a target of 10,000 is < 20% of the level of saturation, while that same target for the 1010's replacement camera is ~70% of saturation.

When we lowered the target for the gain reference, the images looked significantly better: we lost the horrible black artifacts and obtained good images regardless of the electron dose (except when the sensor is saturated). We tried several different gain reference targets, and finally settled on a target value of 7000.


Calibration of magnifications: Once we had solved the problem wtih the gain references and the overall appearance of images collected using the replacement camera, we needed to calibrate the different magnifications. Calibrating a camera with DigitalMicrograph is done by measuring known distances with some sort of calibration grid and turning those known distances into pixel sizes and the pixel size into an apparent magnification. For the lower mags, spacings between the lines in a replica diffraction grating (a "waffle grid") and/or the diameter of latex beads (if any are present) can be used. For higher mags, the atomic spacings in various crystals (catalase, asbestos, Au, etc.) can be used (and these measurements are actually done in Fourier space).

For the lower mags, there must be at least one diffraction grating line present in an image used for calibration (and significantly more is much better), and "lower mags" reallly means "magnifications low enough to have 1 or 2 lines in an image. On the JEOL JEM 3200FS (with 15 μm pixels and a 4k x 4k field of view), one can use images up to ~25,000x for this purpose. For the higher magnifications, the spacings from polycrystalline Au/Pd on the waffle grids becomes visible at 100,000x and higher. However, for this replacement camera, with larger pixels (24 μm) and a 1k x 1k field of view, the highest magnification that can be used with the waffle grid's diffraction grating lines is ~12,000x, and we were unable to see any atomic spacings at high magnifications where we attempted to record calibration images. Because of these constraints, the replacement camera was calibrated using images acquired in normal Mag mode, starting at 600x and ending at 12,000x. The magnifications and pixel sizes from this calibration are collected here.

While acquiring images to calibrate the new camera, we noticed that there was a significant difference in magnification for the two magnifications (600x and 1000x) that appear in both the normal Mag mode magnifications and in the Low Mag mode magnifications. For example, the images below are both nominally recorded at 1000x: the image at the left was acquired at the 1000x magnification in the normal Mag mode settings (the third available mag) while the image at the right was recorded at the highest possible Low Mag mode magnification, also nominally 1000x. These images are clearly nowhere near the same magnification, and based on measuring the length of 20 "waffle squares," the image at the left is ~1.27x higher magnification than the image on the right.

image of waffle grid at nominal magnification of 1000x in normal Mag mode image of waffle grid at nominal magnification of 1000x in Low Mag mode

We will have a JEOL service engineer who works on the 1010 investigate this issue. At this time, we do not know whether the Low Mag mode magnifications are all slightly off, or if some of the lower normal Mag magnifications are incorrect.


Finally, because the replacement CCD contains many fewer pixels than the original CCD (220 vs 224), the readout time for the replacement camera is significantly faster: The original camera required ~60 s while the replacement camera only requires ~10 s. This makes it easier to do things like focus and stigmate using the Camera View modes in DigitalMicrograph, and generally speeds data collection.

We have both a detailed description of how to use this replacement camera and the Gatan technical document for the family of MSC cameras (including this MSC 794).