Skipper: Advances in CCD Readout and Algorithms

• Skipper is a polysemous term that primarily defines a CCD design allowing repeated, non-destructive charge measurements to achieve single-electron sensitivity.
• The technique reduces readout noise via averaging multiple samples, following a 1/√N scaling law, which is crucial for low-threshold dark matter and astronomical spectroscopy applications.
• Additional implementations of Skipper include a software technique for quantum annealing and a maximal-matching algorithm in graph processing, highlighting its broader technical impact.

Searching arXiv for papers on “Skipper” to ground the article in the current literature. arxiv_search(query="all:Skipper", max_results=10, sort_by="submittedDate") arxiv_search(query="Skipper", max_results=10, sort_by="submittedDate") Skipper is a polysemous technical term used most prominently for a class of charge-coupled-device readout architectures that permit repeated, non-destructive measurement of the same pixel charge and thereby achieve sub-electron noise, but it also designates a software technique for quantum annealers and a parallel maximal-matching algorithm in graph processing (Drlica-Wagner et al., 2021, Ayanzadeh et al., 2023, Esfahani, 6 Jul 2025). In detector physics, the Skipper-CCD lineage is the dominant usage: it underlies electron counting, low-threshold dark-matter and neutrino searches, astronomical spectroscopy, space-oriented X-ray instrumentation, and recent CMOS hybrids.

1. Skipper-CCD concept and historical placement

A Skipper-CCD is a specialized CCD whose floating-gate output stage allows the same packet of charge to be measured multiple times without destroying it. This distinguishes it from a conventional CCD, where a pixel is effectively measured once and the resulting readout noise sets the limit on whether one can resolve $0,1,2,\ldots$ electrons in a pixel. Modern literature presents this repeated-sampling capability as the defining Skipper innovation, and notes that electron counting was first demonstrated in the design by Stephen Holland at LBNL (Cervantes-Vergara et al., 2022).

The core operating law is the familiar averaging relation

$\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$

where $\sigma_1$ is the single-sample readout noise and $N_{\rm samp}$ is the number of non-destructive samples per pixel (Drlica-Wagner et al., 2021). Because the charge packet is not consumed by a single measurement, the effective noise can be driven well below $1\,e^-$, making discrete charge peaks directly observable. This is the basis of “single-electron sensitivity,” “single-electron resolution,” and, in optical contexts, “single-photon counting” (Lapi et al., 2024).

The principal tradeoff is readout time. Skipper operation improves precision by increasing the number of samples, but total readout time scales proportionally to the number of measurements per pixel and to the number of pixels. This establishes the central systems problem of Skipper instrumentation: how to preserve sub-electron performance without incurring prohibitive acquisition times (Chierchie et al., 2020).

2. Readout theory, noise scaling, and charge quantization

The Skipper readout chain combines CCD charge transfer with repeated digital estimation of the same pixel charge. In the standard formulation, the CCD video signal contains pedestal and signal intervals; the pixel estimate is formed by a dual-slope-integration-like average, and the same charge packet is then measured $N$ times and averaged. When the individual estimates are independent and identically distributed, the resulting standard deviation decreases as $1/\sqrt{N}$ (Chierchie et al., 2020).

This noise reduction has concrete experimental consequences. In a $250~\mu\text{m}$ thick, fully depleted Skipper CCD characterized for astronomy, the reported single-sample noise was $\sigma_1=3.57\,e^-$, improving to $\sigma_{200}\approx0.25\,e^-$ at 200 samples and $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$0 at 400 samples; the same study reports stable single-electron resolution and visible $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$1, $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$2, $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$3, … peaks in histograms of pixel values (Drlica-Wagner et al., 2021). A dedicated Low Threshold Acquisition controller later demonstrated $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$4 at $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$5 with $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$6, placing the residual uncertainty at roughly $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$7 of one electron rms (Cancelo et al., 2020).

Charge quantization is not merely a qualitative effect. In the astronomical characterization study, the gain derived from the spacing of individual-electron peaks was $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$8, while the gain from the photon transfer curve was $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$9, an agreement within $\sigma_1$0 and summarized as better than $\sigma_1$1 (Drlica-Wagner et al., 2021). In a commercial CMOS implementation that embeds a Skipper-like output stage behind a pinned photodiode, the same scaling law was measured up to 3025 samples, yielding $\sigma_1$2 rms in the noise scan and $\sigma_1$3 rms from single-photon histograms (Lapi et al., 2024).

3. Controllers, focal planes, ROI strategies, and CMOS variants

Skipper instrumentation has evolved around the problem of turning repeated non-destructive sensing into deployable systems. A dedicated controller architecture, the Low Threshold Acquisition controller, was introduced as a compact single-board, four-channel platform with low-noise differential video amplifiers, 18-bit 15 Msps ADCs, a Xilinx Artix XC7A200T FPGA, programmable clock and bias generation, and Ethernet-based DAQ; the stated purpose was to remove the electronics bottleneck that had limited extensive use of Skipper-CCDs (Cancelo et al., 2020).

A second systems development addressed the time penalty of uniform high-$\sigma_1$4 sampling. The “smart skipper readout” mode allows the user to specify one or more regions of interest and apply large $\sigma_1$5 only inside those regions, while using far fewer measurements elsewhere. The implementation replaced a fixed state machine with a dedicated microprocessor instantiated in the FPGA and used a pseudo-XML “recipe” to describe arbitrary pixels or rectangular blocks with different measurement counts. Because changing the sequence induces baseline transients, the method also introduced pixel-by-pixel baseline estimation from the median of multiple empty images acquired with the same ROI pattern (Chierchie et al., 2020).

A third development path pursued astronomical deployment and high channel count. A focal plane for the SOAR Integral Field Spectrograph was built as a four-detector mosaic of $\sigma_1$6, $\sigma_1$7-pixel, fully depleted, p-channel Skipper CCDs, read out through 16 amplifiers. The engineering-grade prototype achieved $\sigma_1$8 single-sample noise in the best channel and $\sigma_1$9 from 800 measurements of the same charge in each pixel (Villalpando et al., 2022). On sky, the later SIFS prototype optimized ROI timing to reach $N_{\rm samp}$0 in a $N_{\rm samp}$1 region and $N_{\rm samp}$2 in a $N_{\rm samp}$3 region, each with readout time $N_{\rm samp}$4 min (Villalpando et al., 2024).

Implementation	Core mechanism	Representative result
Low Threshold Acquisition controller	Single-board dedicated Skipper-CCD control and acquisition	$N_{\rm samp}$5 for 5000 pixel measurements (Cancelo et al., 2020)
Smart skipper readout	Variable $N_{\rm samp}$6 by ROI with baseline correction	$N_{\rm samp}$7 letters at $N_{\rm samp}$8 versus $N_{\rm samp}$9 background at $1\,e^-$0 (Chierchie et al., 2020)
SIFS focal plane	Four-detector mosaic with synchronized 16-amplifier readout	$1\,e^-$1 photon-counting ROI in $1\,e^-$2 min (Villalpando et al., 2024)
Skipper-in-CMOS	PPD plus in-pixel Skipper-style CCD output stage	$1\,e^-$3 rms and single-photon counting in 180 nm CIS (Lapi et al., 2024)

4. Scientific applications: astronomy, rare-event searches, and metrology

Astronomy has been one of the principal application domains because Skipper-CCDs preserve standard thick, fully depleted CCD virtues while suppressing readout noise. The $1\,e^-$4 fully depleted device characterized for cosmological applications was operated at $1\,e^-$5 K, read out through four amplifiers, and showed relative QE $1\,e^-$6 from 450 nm to 900 nm, saturation at about $1\,e^-$7, and regional multi-sampling with $1\,e^-$8 single-sample noise reduced to $1\,e^-$9 in a configured subregion (Drlica-Wagner et al., 2021). The SIFS on-sky campaign then demonstrated charge-quantized, photon-counting spectroscopy of HB89 1159+123, and reported a change in read noise from about $N$0 at $N$1 to about $N$2 at $N$3, with the signal-to-noise ratio in one emission-line-galaxy observation improving from roughly $N$4 to $N$5 (Villalpando et al., 2024).

Rare-event particle physics has used the same detector property for a different purpose: resolving $N$6, $N$7, $N$8, and $N$9 ionization signals. A new $1/\sqrt{N}$0-gram SENSEI Skipper-CCD operating at $1/\sqrt{N}$1 K in the MINOS cavern measured a post-subtraction single-electron rate

$1/\sqrt{N}$2

observed five single-pixel $1/\sqrt{N}$3 events, no $1/\sqrt{N}$4 or $1/\sqrt{N}$5 clusters, and achieved world-leading sensitivity across broad sub-GeV dark-matter parameter space (Barak et al., 2020). A later SENSEI study decomposed single-electron events previously bundled together as “dark counts” into dark current, amplifier light, and spurious charge, with measured values $1/\sqrt{N}$6 for dark current and $1/\sqrt{N}$7 for spurious charge, while amplifier light was reduced by about two orders of magnitude by changing $1/\sqrt{N}$8 from $1/\sqrt{N}$9 to $250~\mu\text{m}$0 (Barak et al., 2021).

Neutrino experiments have exploited the same threshold advantage. CONNIE installed two Skipper-CCDs next to the Angra-2 reactor and, with $250~\mu\text{m}$1, pushed the analysis threshold for CE$250~\mu\text{m}$2NS-related searches to $250~\mu\text{m}$3, described as the lowest energy threshold among all current CE$250~\mu\text{m}$4NS experiments (Aguilar-Arevalo et al., 2024). Scaling studies for Oscura reported an electron-counting yield of $250~\mu\text{m}$5 for Microchip-fabricated sensors, $250~\mu\text{m}$6 RMS noise with 1225 samples/pix, and a dark current of $250~\mu\text{m}$7 at 140 K, in the context of a 10 kg experimental goal (Cervantes-Vergara et al., 2022).

Metrology has produced a different application: a proof-of-concept quantum-current source based on self-measured and drained charge packets. Using a standard four-amplifier Skipper-CCD, one experiment resolved eleven peaks from 400 to 410 electrons, obtained $250~\mu\text{m}$8 against a Keithley 6517A electrometer with slope correction $250~\mu\text{m}$9, reported about $\sigma_1=3.57\,e^-$0 nA current for an 80 s exposure in a faster mode, and separately observed discrete current steps with mean about $\sigma_1=3.57\,e^-$1 fA (Gamero et al., 11 Feb 2025).

5. Backgrounds, radiation tolerance, and space-oriented X-ray use

The main misconception corrected by recent Skipper-CCD work is that all isolated single-electron events can be treated as undifferentiated “dark counts.” The SENSEI modeling of single-electron events showed instead that exposure-dependent dark current, readout-time-dependent amplifier light, and exposure-independent spurious charge are distinct components with different operational controls (Barak et al., 2021). A later SENSEI study further localized the dominant low-background contribution to the serial register during Skipper readout and introduced a tri-level clocking scheme in which the held-low phase is raised to an intermediate voltage during the readout dwell. Under standard conditions, the serial-register single-electron density was $\sigma_1=3.57\,e^-$2 electrons/pixel/image and fell to $\sigma_1=3.57\,e^-$3 electrons/pixel/image with tri-level clocking, a factor of $\sigma_1=3.57\,e^-$4 reduction (Wu et al., 28 May 2026).

Radiation hardness has become central because Skipper-CCDs are being considered for long-duration space use. Preliminary irradiation at the Northwestern Medicine Proton Center with 217-MeV protons showed that p-channel Skipper amplifiers continued to function after doses exceeding six years at Sun-Earth L2, with noise still following $\sigma_1=3.57\,e^-$5 up to at least 324 samples and reaching $\sigma_1=3.57\,e^-$6–$\sigma_1=3.57\,e^-$7 rms/pix on working amplifiers (Roach et al., 2024). A more X-ray-specific study for the DarkNESS mission irradiated a thick, fully depleted p-channel Skipper-CCD quadrant at a fluence of $\sigma_1=3.57\,e^-$8; the Mn $\sigma_1=3.57\,e^-$9 FWHM changed from $\sigma_{200}\approx0.25\,e^-$0 eV in a non-irradiated reference quadrant to $\sigma_{200}\approx0.25\,e^-$1 eV in the beam-exposed full quadrant, while a worst-case 3-year Sun-synchronous-orbit end-of-life prediction gave $\sigma_{200}\approx0.25\,e^-$2 eV under the conservative linear model (Alpine et al., 2 Feb 2026).

Space X-ray instrumentation imposes an additional requirement: suppressing optical and near-infrared loading without sacrificing keV X-ray efficiency. Thin aluminum layers deposited by e-beam evaporation have been tested as integrated optical blockers. Aluminum thicknesses of 50 and 100 nm provided $\sigma_{200}\approx0.25\,e^-$3 and $\sigma_{200}\approx0.25\,e^-$4 suppression, respectively, across most of the 650–1000 nm range, while showing no apparent efficiency loss for 5.9 and 6.4 keV X-rays from $\sigma_{200}\approx0.25\,e^-$5Fe (Botti et al., 31 Dec 2025). This strategy feeds directly into DarkNESS, a 6U CubeSat that will use four thick, fully depleted p-channel Skipper-CCDs, $\sigma_{200}\approx0.25\,e^-$6–$\sigma_{200}\approx0.25\,e^-$7 keV X-ray sensitivity, and space-qualified Low Threshold Acquisition electronics adapted to the PC104 CubeSat form factor, with launch opportunity reported through Firefly Aerospace’s DREAM 2.0 program (Alpine et al., 22 May 2025).

6. Unrelated homonyms: quantum annealing and maximal matching

Outside detector physics, “Skipper” also denotes a software technique for quantum annealers. In that context, the problem is minor-embedding overhead: logical “program qubits” are represented by chains of physical qubits, and long dominant chains both reduce effective capacity and increase fragility. The method called Skipper skips dominant chains, replaces the removed program qubit with its two possible readout values $\sigma_{200}\approx0.25\,e^-$8 and $\sigma_{200}\approx0.25\,e^-$9, and solves the resulting subproblems. On a 5761-qubit D-Wave Advantage system, this enabled up to $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$00 larger problems with an average increase of $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$01 when up to 11 chains were skipped, and reduced the energy residual by up to $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$02 with average improvement $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$03 when trimming up to five chains; the greedy variant Skipper-G reduced the run burden from up to 2048 quantum executables to at most 23 for eleven cuts (Ayanzadeh et al., 2023).

A second unrelated use appears in shared-memory graph processing. There, Skipper is a deterministic incremental asynchronous maximal-matching algorithm that processes both endpoints of an edge together, uses only one byte per vertex to encode the states Accessible, Reserved, and Matched, and makes a single pass over the edge set. The evaluation covered real-world and synthetic graphs up to 161.1 billion edges on AMD Zen2, AMD Zen3, and Intel Xeon Platinum systems. Across those datasets, the algorithm processed only $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$04 of edges on average, delivered a $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$05 average speedup in the abstract, and achieved an output size that was $\sigma_N = \frac{\sigma_1}{\sqrt{N_{\rm samp}}},$06 of the Lim-Chung algorithm on average (Esfahani, 6 Jul 2025).

In contemporary technical usage, these meanings remain distinct. “Skipper” most often denotes the floating-gate, repeated-sampling CCD architecture and its derivatives, while separate literatures in quantum optimization and graph algorithms use the same name for techniques built around selective skipping of dominant overheads.