Cameras for Cold-Atom Absorption Imaging

November 14, 2026

Laser-cooled and ultracold atoms sit at the center of atomic clocks, atom interferometers, quantum simulators and quantum sensors. National labs, university groups and instrument companies share one measurement that almost every setup depends on: a camera records the shadow an atom cloud casts in a resonant probe beam, and from that shadow come atom number, temperature and density.

This post is for physicists and engineers choosing or replacing the camera on a cold-atom apparatus. It covers what absorption imaging asks of a camera, with Finger Lakes Instrumentation (FLI) cooled cameras as worked examples. For larger detector programs at national labs, see our post on large format cameras for national labs.

What the camera measures

A standard absorption image takes three frames. The first records the probe beam after it passes through the atoms. The second records the probe alone, after the atoms are gone. The third is a dark frame with no probe. For each pixel, the optical density is the natural log of (probe − dark) divided by (atoms − dark). At low probe intensity on resonance, optical density divided by the absorption cross-section gives the column density of atoms. Ketterle, Durfee and Stamper-Kurn's 1999 review of Bose-Einstein condensate experiments is a widely used introduction.

Because every result is a ratio of frames, the camera must respond the same way in every frame, and its offsets must be removed cleanly.

Wavelength, sensor and window

Each atomic species has its own imaging wavelength, from the blue and near-UV for strontium and ytterbium to the red and near-IR for lithium, potassium, rubidium and cesium. Sensor response falls off at both ends of the visible band, so check the published quantum efficiency curve at your exact probe line rather than relying on a headline peak value.

The camera window matters as well. On FLI cameras the F116 window transmits farther into the near-IR and the F101 window farther into the near-UV. Ask for the transmission chart for your wavelength. The window is chosen before a camera enters our clean-room build, so settle it early. Coherent probe light also turns any parallel surfaces in the imaging path into etalons, so the window and the optics around it are part of fringe control.

Pixels at the atoms

What matters is the pixel size projected onto the atom plane, which is the sensor pixel pitch divided by the imaging magnification. In-trap images need fine sampling; time-of-flight images of expanded clouds need a wider field.

The FLI AR533, and the RV533 with the same sensor in a thinner body, use a 3003 × 3003 back-illuminated sensor with 3.76 µm pixels on an 11.3 mm square. The KL400BI uses 2048 × 2048 back-illuminated pixels of 11 µm on a 22.5 mm square, which gives each pixel more area to collect probe light. Our post on pixel size and sampling covers the matching math.

Aurora cameras bin digitally, because their Sony sensors do not support on-sensor binning. Digital binning sums read noise from each pixel in the bin, so it is usually better to set the magnification for the sampling you need than to plan on heavy binning later.

Photons, noise and headroom

Absorption imaging is usually limited by photon shot noise on the probe, so the camera should hold enough electrons per pixel for the probe level you want to use. The AR533 and RV533 list a 50,000 e- full well and 1 e- rolling read noise. The KL400BI lists 1.6 e- read noise and a 16-bit output merged from two 12-bit conversions, which helps when one image holds both a dark cloud center and bright unabsorbed probe.

Dense clouds can absorb almost all the probe in the center, and the optical density there becomes hard to measure. One published approach is to image with a probe well above saturation intensity and correct for it in analysis (Reinaudi et al., Optics Letters, 2007). That raises the photon count per pixel, so full well and linearity both matter. Low read noise matters most for fluorescence imaging of small atom numbers, where the signal per pixel can be just a few photons. For measuring gain, noise and linearity yourself, see our post on camera acceptance testing.

Timing the probe pulse

Probe pulses are short, so the camera exposure has to be timed around them. External trigger in and out is standard on Aurora and Revelation cameras. In rolling shutter readout, rows start and stop exposing at different times, so a short pulse has to fall in the window when every row is integrating. Rolling with global reset, listed for the AR533 and RV533 and for the KL4040FI, starts all rows together. On Aurora cameras in triggered global reset mode, expect a row-reset delay of roughly 300 to 350 µs, and trigger the camera before the probe pulse. In both cases, keep exposures short so room light and stray light add little background.

Decide which device is the timing master, often the experiment's sequencer. Our post on synchronizing multiple cooled cameras covers multi-camera setups, such as imaging along two axes.

Fringes and the reference frame

The most common flaw in absorption images is fringing. Dust and optical surfaces imprint interference patterns on the probe beam, and vibration shifts those patterns between the atom frame and the reference frame. The result is stripes in the optical density map that can look like real structure.

Two habits help. First, keep the delay between the atom and reference frames short, so the beam has less time to move. Ask about the minimum frame-to-frame interval for the readout mode and region of interest you plan to use. Second, collect a library of reference frames and build an optimal reference for each shot in software, an approach described by Ockeloen et al. (Physical Review A, 2010).

Cooling, darks and the apparatus

Exposures in absorption imaging are short, so dark current is small. At the published 0.002 e-/pixel/s at -20 °C for the AR533, a 1 ms exposure collects about 0.000002 e- per pixel. Cooling earns its place another way: a regulated sensor temperature keeps offsets and response steady through long runs of shots.

Aurora and Revelation cool to 50 °C below ambient with air, with liquid cooling as an option. That matters on an optical table, where fan vibration can be unwelcome. Aurora is not offered with a mechanical shutter, so take dark frames with the probe off. Revelation's thin body helps when the camera sits in a crowded space around a vacuum chamber, and every Revelation camera includes our software development kit (SDK). For lens mounts and spacing, see our post on back focus and flanges.

Questions to settle before you choose

  • Which species and probe wavelength will you image, and which window suits it?
  • What pixel size do you need at the atoms, in trap and after time of flight?
  • What probe intensity and pulse length will you use, and how many electrons per pixel does that give?
  • How short must the gap between the atom and reference frames be?
  • Which device is the timing master, and how many cameras share the trigger?
  • How much vibration and heat can the table and enclosure tolerate?

Talk to our sales team

Send our sales team your species, wavelength, imaging magnification and timing scheme, and we will suggest the standard Aurora, Revelation or Kepler configuration that fits. If an instrument needs a modest change to a standard camera, we are glad to discuss it. Request a quote or email sales@flicamera.com.

FLI cameras are Made in USA, and our sensor chambers are assembled in our 1,000+ sq ft ISO Class 7 clean room. See how on our in-house manufacturing page.