CYGNO/INITIUM Experiment Overview

• CYGNO/INITIUM is a modular rare-event search program using a gaseous TPC with GEM-based optical readout to reconstruct 3D event topologies and directional signals.
• The experiment’s staged evolution—from prototypes like LEMOn and LIME to the underground demonstrator CYGNO-04—validates low-threshold, high-precision track detection.
• INITIUM’s R&D on negative-ion drift and advanced machine learning techniques enhances event reconstruction by reducing diffusion and improving background discrimination.

Searching arXiv for recent CYGNO/INITIUM papers to ground the article. The CYGNO/INITIUM experiment is a program in directional rare-event detection centered on gaseous time projection chambers with GEM-based amplification and optical readout. In this framework, CYGNO is the main experiment, while INITIUM denotes the broader R&D effort that develops enabling technologies for scalable directional detectors, including negative-ion drift; the program is also situated within the CYGNUS proto-collaboration’s effort toward underground directional observatories for dark matter and neutrino physics (Amaro et al., 2023, Baracchini et al., 2020). Its detector line is based on He:CF$_4$ gas at atmospheric pressure, triple-GEM amplification, sCMOS imaging, and PMT timing, with the explicit aim of reconstructing low-energy nuclear and electron recoils in three dimensions and exploiting their directional information for low-mass dark matter and solar-neutrino measurements (Amaro et al., 2023, Amaro et al., 2022).

1. Program definition and evolution

CYGNO was introduced as a gaseous TPC with optical readout for directional dark matter searches and coherent neutrino scattering measurements, with early design literature describing a 1 m$^3$ module as a prototype toward the 100–1000 m$^3$ CYGNUS network (Baracchini et al., 2019, Baracchini et al., 2020). Later programmatic documents define CYGNO-04 as a 0.4 m$^3$ underground demonstrator, while retaining the longer-term trajectory toward 30 m$^3$ and larger installations (Amaro et al., 2023, Amaro et al., 23 Mar 2026). This suggests a staged evolution in which the architectural principles remained stable while the intermediate demonstrator scale was refined.

Within this sequence, LIME occupies a central position. It is described as the largest prototype built so far, with dimensions matching one basic module of the future CYGNO demonstrator, and its underground operation is explicitly treated as a technological and physics pathfinder for a modular detector composed of repeated LIME-like units (Amaro et al., 2023). In parallel, INITIUM is presented in other CYGNO documents as an ERC-supported R&D path devoted in particular to negative-ion drift in He:CF$_4$:SF$_6$ mixtures at nearly atmospheric pressure, developed “in collaboration and synergy with CYGNO” (Baracchini et al., 2020).

Stage	Stated configuration	Stated role
LEMOn	7 L, 20 × 24 cm$^2$ readout, 20 cm drift	Light-yield, energy-resolution, and first PID studies
LIME	50 L, 33 × 33 cm$^2$ readout, 50 cm drift	Key technological and physics prototype; basic CYGNO module
CYGNO-04	0.4 m$^3$, back-to-back TPC	Underground demonstrator for scalability and background control
CYGNO-30	30 m$^3$0	Long-term physics-competitive directional stage

The scientific scope is correspondingly broad but internally coherent. CYGNO targets low-mass WIMPs, directional discrimination against isotropic backgrounds, sensitivity to both spin-independent and spin-dependent interactions through helium and fluorine nuclei, and, in several papers, solar-neutrino measurements whose directional signatures differ from the expected Cygnus-correlated dark-matter recoil distribution (Amaro et al., 2023, Amaro et al., 23 Mar 2026).

2. Detector architecture and operating principle

The baseline CYGNO detector concept is a gaseous TPC filled with He:CF$^3$1 at 60:40 and operated at atmospheric pressure, with a triple-GEM amplification stage coupled to an optical readout consisting of sCMOS cameras and PMTs (Amaro et al., 2023). Helium supplies a light nuclear target and long low-energy recoil tracks, while CF$^3$2 provides visible scintillation and sensitivity to spin-dependent interactions through fluorine (Amaro et al., 23 Mar 2026). The combination is optimized for rare-event topologies at the few-keV scale rather than for bulk calorimetry alone.

LIME realizes this concept in a single-module form: a 50 L active volume, 50 cm drift length, 33 × 33 cm$^3$3 readout plane, triple thin GEMs, one sCMOS camera imaging the full GEM area, and four PMTs located at the corners of the readout plane (Amaro et al., 2023). The camera supplies the high-granularity $^3$4 projection, while PMTs record the time profile of the scintillation light and provide trigger information. In the standard reconstruction picture, the drift coordinate is obtained from

$^3$5

with $^3$6 the electron drift velocity and $^3$7 the measured drift time (Amaro et al., 2023, Amaro et al., 2023).

At the demonstrator scale, CYGNO-04 is specified as a back-to-back TPC with two drift volumes separated by a central cathode; early design descriptions gave two 50 cm drift regions for a total drift length of 1 m and 50 × 80 cm$^3$8 readout area per side (Amaro et al., 2023). More recent programmatic descriptions specify a 0.4 m$^3$9 detector with V-bonded GEMs, three qCMOS cameras per side, and eight PMTs per side (Amaro et al., 23 Mar 2026). The use of V-bonded GEMs is explicitly intended to reduce light reflections and dead areas between segments (Amaro et al., 23 Mar 2026).

A recurrent point in the literature is that CYGNO pursues this architecture at atmospheric pressure rather than in the low-pressure regime common to many other directional gas detectors. The stated rationale is that, with sufficiently granular optical readout, mm-scale tracks at atmospheric pressure remain reconstructible and still retain topology and head–tail information relevant to dark-matter searches (Amaro et al., 23 Mar 2026).

3. Prototype chain, modularity, and underground deployment

The prototype sequence is structured as an incremental scale-up. LEMOn established the optical-TPC response in a 7 L device and was used for light-yield measurements, energy calibration with $^3$0Fe, and first neutron-versus-X-ray particle-identification studies (Amaro et al., 2023). LIME then moved the concept to a 50 L mono-chamber geometry intended to match one future CYGNO module (Amaro et al., 2023).

LIME was installed underground at LNGS in February 2022 (Amaro et al., 2023). A later status paper describes about 27 months of underground data taking organized in five runs, with configurations ranging from no shielding to 10 cm copper plus 40 cm water (Amaro et al., 23 Mar 2026). In that sequence, the trigger rate changed from about 34 Hz in the unshielded run to about 1 Hz with copper plus water shielding, and the campaign accumulated about $^3$1 images (Amaro et al., 23 Mar 2026). These operations were not merely endurance tests: they were used to validate background models, characterize internal and external radioactivity, acquire neutron-source data, and test the gas system, PMT-camera reconstruction chain, and shielding strategy in the exact single-module geometry intended for replication (Amaro et al., 2023, Amaro et al., 23 Mar 2026).

The modular interpretation of LIME is explicit. The detector is described not as a generic test stand but as a basic unit for a larger demonstrator composed of “multiple modules with the same LIME dimensions and characteristics” (Amaro et al., 2023). This modularity connects directly to CYGNO-04, whose role is to demonstrate that the optical-TPC concept scales from the 50 L class to the 0.4 m$^3$2 class while preserving stable drift, gain, and optical performance (Amaro et al., 23 Mar 2026).

CYGNO-04 is sited in LNGS Hall F, with civil works completed and inner detector and vessel expected to be ready in spring 2026 according to the 2026 status report (Amaro et al., 23 Mar 2026). The shielding concept for this stage is also more elaborate than in LIME: a sealed PMMA inner vessel, a 4 cm radiopure copper external vessel, an additional 6 cm of copper from OPERA, and around 100 cm of water (Amaro et al., 23 Mar 2026).

4. Calibration, reconstruction, and measured performance

The core overground characterization of LIME used radioactive X-ray sources. A $^3$3Fe source emitting 5.9 keV X-rays was positioned at different distances along the 50 cm drift, and the measured energy resolution was around 14% across the full drift length (Amaro et al., 2023). The same study reports preliminary multivariate-regression results with energy resolution better than 10% at 5.9 keV, and a threshold of 0.5 keV chosen so that the noise contribution corresponds to a maximum rate of 10 fake $^3$4Fe events per year (Amaro et al., 2023). For multiple X-ray lines between 3.7 keV and 47 keV, the detector response was reported as linear over the full range (Amaro et al., 2023).

Long-term stability was also quantified in the Frascati campaign: during about one month of operation, LIME ran under stable conditions with a current-spike rate below 2.7 spikes/hour (Amaro et al., 2023). This stability criterion was important because the underground demonstrator concept assumes continuous operation with repeated LIME-like modules. In the underground phase, the 2026 status paper further reports that the $^3$5Fe light yield, after correcting for gas-condition variations, was stable to 5% over about six months (Amaro et al., 23 Mar 2026).

At smaller scale, LEMOn established the optical light-yield benchmark that shaped the broader CYGNO design. With He:CF$^3$6 60:40 and a triple 50 µm GEM stack, the collaboration measured about 650 detected photons per keV and an energy resolution of about 15% at 5.9 keV, fairly constant along the full drift distance; from those measurements, the collaboration inferred an energy threshold of order 1 keV and explicitly set 1 keV$^3$7 as the design goal for later demonstrators (Amaro et al., 2023).

Event reconstruction is based on image clustering and topology extraction. CYGNO developed an adapted density-based clustering algorithm, iDBSCAN, for optical TPC images; in the LEMOn analysis it was shown to provide full signal detection efficiency and very good energy resolution while improving detector background rejection relative to both DBSCAN itself and nearest-neighbor clustering (Baracchini et al., 2020). The same clustering philosophy appears in later LIME analyses, where intensity-based DBSCAN is used on both data and simulation to extract track length, width, and light-density observables (Amaro et al., 2023).

Three-dimensional reconstruction is already demonstrated for high-ionization events. The 2026 LIME status report describes a Bayesian reconstruction method combining camera clusters with PMT timing observables; alpha tracks were reconstructed in 3D, and the resulting length spectrum displayed peaks consistent with internal contaminants from $^3$8Rn and the $^3$9U and $^3$0Th chains (Amaro et al., 23 Mar 2026). For low-energy rare events, the same paper states that measurement of track length, width, light density, and head–tail asymmetries is being used to improve electron-recoil versus nuclear-recoil discrimination (Amaro et al., 23 Mar 2026).

5. Simulation, backgrounds, and directional response

Simulation has been developed as an integral part of the CYGNO/INITIUM program rather than as an auxiliary analysis layer. In LIME, primary energy deposition is simulated with GEANT4 for electronic recoils and SRIM for nuclear recoils; the detector-response chain includes ionization yield, diffusion along the drift region and inside the GEMs, charge amplification, photon yield and collection efficiency, absorption in the gas, and gain saturation (Amaro et al., 2023). For 5.9 keV electron recoils, the simulated light integral and track dimensions showed good agreement with $^3$1Fe data across different GEM gain values, and in the linearity study the preliminary agreement between data and simulation was within 10% (Amaro et al., 2023).

The 2025 light-response modeling study makes explicit why gain saturation matters in CYGNO. Because optical readout requires total gains of order $^3$2–$^3$3, the detector response depends on the spatial density of charge entering GEM holes; a space-charge model for a 2 L prototype reproduced the gain behavior over nearly one order of magnitude with percent-level precision (Amaro et al., 9 May 2025). This provides a calibration framework for predicting how light yield changes with drift-distance-dependent charge density, and it is directly relevant to low-threshold energy reconstruction in larger modules.

Background modeling for underground LIME included intrinsic radioactivity of detector materials together with the natural ambient gamma and neutron flux (Amaro et al., 2023). A principal result was that fiducial cuts, enabled by 3D reconstruction, reduced radioactivity-induced background events by 96% (Amaro et al., 2023). The same study states that shielding design was optimized through dedicated Monte Carlo simulations, with copper used for gamma shielding and water tanks for neutron shielding (Amaro et al., 2023). Before full water shielding, LIME was also intended to perform a spectral measurement of the fast neutron flux underground at LNGS, using nuclear recoils induced by environmental neutrons (Amaro et al., 2023).

Directional detection is the program’s central discriminator. The 2023 CYGNO overview states that the apparent dark-matter wind should come from the direction of the Cygnus constellation, producing an anisotropic recoil distribution that no known terrestrial background can mimic together with its sidereal modulation (Amaro et al., 2023). In the 2026 status paper, CYGNO is described as optimized for light $^3$4–$^3$5 GeV WIMPs-like particles, with 3D event reconstruction, detailed energy-deposition mapping, and effective topology and head-to-tail discrimination (Amaro et al., 23 Mar 2026). At the current stage, however, not every performance metric has been converted into a final directional sensitivity curve. A preliminary dark-matter sensitivity from a 0.81 kg day LIME exposure was derived under a simplified counting-experiment assumption without angular information; that result was reported as already competitive with contemporary directional searches, while more realistic analyses with $^3$6-dependent thresholds and angular information were still in progress (Amaro et al., 23 Mar 2026).

6. INITIUM-specific R&D, negative-ion drift, and future instrumentation

A defining INITIUM theme is negative-ion drift. Earlier CYGNO documents already identified INITIUM as the R&D line devoted to He:CF$^3$7:SF$^3$8 operation at nearly atmospheric pressure, with the objective of reducing diffusion by replacing electron transport with negative-ion transport (Baracchini et al., 2020). This route was realized experimentally in the first optical observation of negative-ion drift at surface pressure, performed at $^3$9 mbar in a He:CF$^3$0:SF$^3$1 mixture using an optically read out CYGNO/INITIUM TPC (Amaro et al., 6 Mar 2026). The PMT waveform analysis yielded inferred drift velocities corresponding to mobilities of order cm$^3$2 V$^3$3 s$^3$4, and the linear scaling of the mean time extension with drift distance indicated a faster minority carrier population drifting at about 25% higher velocity than the dominant SF$^3$5 species (Amaro et al., 6 Mar 2026). The stated implication is direct: multi-species negative-ion drift at surface pressure opens a concrete path toward large-scale, low-diffusion optical TPCs for rare-event searches (Amaro et al., 6 Mar 2026).

Another major R&D axis concerns light yield. In the study of He:CF$^3$6-based amplification, the addition of a strong electric field below the last GEM plane was found to permit large light-yield increases without degrading the intrinsic characteristics of the amplification stage with respect to regular GEM operation (Amaro et al., 2024). Since CYGNO’s threshold is set by the number of detectable photons per keV, this induction-field strategy is an enabling technology for lower thresholds, improved topology at fixed optical acceptance, and detector-scale optimization (Amaro et al., 2024).

As the apparatus moves from single-camera prototypes to the multi-camera CYGNO-04 demonstrator, data acquisition and online reduction have become core parts of the experiment rather than auxiliary engineering. The 2026 T-DAQ upgrade paper describes a continuous-imaging acquisition mode for CYGNO that reduces camera dead time from about 38% in the earlier frame-based scheme to about 0.03% for $^3$7 ms and $^3$8s, while also introducing extended PMT time tagging and synchronous multi-camera operation without a master camera (Amaro et al., 16 Mar 2026). This is the hardware foundation for CYGNO-04’s multi-camera optical planes.

On top of that hardware layer, the collaboration has developed machine-learning tools for trigger-level data reduction and weakly supervised event classification. A pedestal-trained convolutional autoencoder for ROI extraction retained $^3$9% of reconstructed signal intensity while discarding $_4$0% of the image area, with about 25 ms inference time per frame on a consumer GPU (Amaro et al., 30 Dec 2025). A separate CWoLa-based classifier trained on mixed AmBe and standard datasets used only mixture labels and achieved performance approaching the theoretical ceiling set by the mixture composition, isolating a high-score population with compact, approximately circular morphologies consistent with nuclear recoils (Amaro et al., 28 Jan 2026). These developments are technically significant because the optical TPC’s physics value depends on preserving topology while keeping the data stream manageable.

In aggregate, the CYGNO/INITIUM experiment is best understood as a modular optical-TPC program whose main elements are already experimentally demonstrated: atmospheric-pressure He:CF$_4$1 operation with triple-GEM scintillation read out by sCMOS cameras and PMTs; a 50 L underground module used as the direct precursor of a larger demonstrator; validated simulation and shielding studies; and an INITIUM branch that extends the same architecture to negative-ion drift and diffusion-limited transport. The immediate future is centered on CYGNO-04 as the first multi-camera underground demonstrator of that full stack (Amaro et al., 23 Mar 2026).