SENSEI at SNOLAB: Single-Electron Event Rate and Implications for Dark Matter

We present results from data acquired by the SENSEI experiment at SNOLAB after a major upgrade in May 2023, which includes deploying 16 new sensors and replacing the copper trays that house the CCDs with a new light-tight design. We observe a single-electron event rate of $(1.39 \pm 0.11) \times 10^{-5}$ e$^-$/pix/day, corresponding to $(39.8 \pm 3.1)$ e$^-$/gram/day. This is an order-of-magnitude improvement compared to the previous lowest single-electron rate in a silicon detector and the lowest for any photon detector in the near-infrared-ultraviolet range. We use these data to obtain a 90% confidence level upper bound of $1.53 \times 10^{-5}$ e$^-$/pix/day and to set constraints on sub-GeV dark matter candidates that produce single-electron events. We hypothesize that the data taken at SNOLAB in the previous run, with an older tray design for the sensors, contained a larger rate of single-electron events due to light leaks. We test this hypothesis using data from the SENSEI detector located in the MINOS cavern at Fermilab.

💡 Research Summary

The paper reports the results of the second science run of the SENSEI experiment at SNOLAB, carried out after a major hardware upgrade in May 2023. The upgrade introduced 16 additional Skipper‑CCD sensors (each 2.19 g, 6144 × 1024 pixels, 15 µm × 15 µm) and replaced the original copper trays with a new light‑tight design that eliminates most pathways for external photons to reach the detector. Data were collected between November 2023 and February 2024 using four exposure times (0 h, 2 h, 6 h, 20 h). In total, 101 “commissioning” images and 77 “hidden” images were recorded.

Data processing followed the established SENSEI pipeline with several refinements. Low‑energy three‑pixel clusters, serial‑register hits, and full‑well events were masked out. The “bleeding zone” along the serial register was extended from 50 to 200 pixels to suppress long charge trails from high‑energy interactions. Pixels were grouped into “super‑pixels” by hardware binning of 32 rows, reducing read‑out noise to 0.14 e⁻ per sample and yielding a read‑out time of about 16 minutes per image. After pixel‑level masking, the charge distribution of unmasked super‑pixels was fitted with a double‑Gaussian model to separate the 0‑e⁻ and 1‑e⁻ peaks. Images with anomalously high fit p‑values or noise deviations were discarded via a “noisy‑image” mask, and quadrants with unusually high 1‑e⁻ counts were excluded using a “hot‑image” mask.

To avoid bias, the analysis defined three regions of interest a priori: a “Golden Quadrant” (the quadrant with the lowest background in the commissioning data) and two “Witness Quadrants” that exhibited similar temporal trends. The 1‑e⁻ density was measured as a function of exposure time for each quadrant. Linear fits yielded an exposure‑dependent slope (the true background rate) and an exposure‑independent intercept (intrinsic CCD noise). In the Golden Quadrant the exposure‑dependent rate was measured as
\