COSINE-100: Direct Dark Matter Search

Updated 7 October 2025

COSINE-100 is a large-scale dark matter experiment located underground in Korea, using radiopure NaI(Tl) crystals and advanced shielding to minimize background noise.
It employs comprehensive passive and active shielding, pulse-shape discrimination, and machine learning techniques to achieve a low energy threshold of 0.7 keV and strong rejection of noise.
Through extended data collection and rigorous background modeling, COSINE-100 sets stringent constraints on WIMP interactions, challenging the DAMA/LIBRA annual modulation claim.

The COSINE-100 experiment is a large-scale underground direct dark matter search employing ultra-low-background NaI(Tl) scintillation crystals. Located at the Yangyang Underground Laboratory (Y2L) in Korea, its primary scientific goal is to independently test the annual modulation signal in nuclear recoil rates reported by DAMA/LIBRA, utilizing the same detector material but with enhanced background discrimination and environmental monitoring. COSINE-100 combines high-purity sodium iodide crystals, extensive passive and active shielding, comprehensive veto systems, advanced data acquisition and analysis techniques, and rigorous background modeling to enable a conclusive comparison with the DAMA/LIBRA results and to scrutinize backgrounds at the unprecedented sub-keV threshold.

1. Experimental Rationale and Scientific Objectives

COSINE-100 was designed to provide a definitive model-independent test of DAMA/LIBRA’s claimed observation of a 9.3σ annual modulation in the 2–6 keV energy region, conventionally interpreted as evidence for Weakly Interacting Massive Particle (WIMP) dark matter interactions. The experiment aims to either confirm or refute this signal by reproducing the detector target (NaI(Tl)), exposure mass, and energy threshold, while adding state-of-the-art background suppression and veto techniques not present in DAMA/LIBRA. The experiment specifically addresses leading background sources, including internal ^210Pb and ^40K, cosmogenic isotopes such as ^3H, neutron-induced events, and muon-induced signals, employing targeted event tagging and environmental control to minimize systematic uncertainties (Adhikari et al., 2017, Collaboration et al., 2017, Adhikari et al., 2018).

Key objectives are:

Achieving detection thresholds at or below 0.7 keV, matching or extending DAMA/LIBRA’s current sensitivity (Yu et al., 26 Aug 2024).
Demonstrating high rejection efficiency for noise and backgrounds via active vetoes and discriminatory analysis (BDTs, MLP).
Providing robust, time-dependent background models over multi-year exposures for unambiguous modulation signal extraction (Carlin et al., 20 Sep 2024, Collaboration et al., 2021).
Setting stringent constraints on WIMP parameter space, with focus on both spin-independent and spin-dependent interactions, extending sensitivity to sub-GeV masses and leveraging rare interaction mechanisms (such as the Migdal effect) (Yu et al., 23 Jan 2025).

2. Detector Architecture and Shielding

NaI(Tl) Crystal Array

The central detector comprises eight low-radioactivity NaI(Tl) crystals totaling ≈106 kg, produced from multiple powder batches to optimize radiopurity (notably for ^210Pb, ^40K) (Adhikari et al., 2017). Each cylindrical crystal is housed in an oxygen-free copper tube, fitted with quartz windows, and wrapped in PTFE layers to maximize photon collection.

Each end is coupled to a 3-inch Hamamatsu PMT, providing high quantum efficiency (~35%) and dual readout (anode for high-gain, 5th dynode for low-gain) (Adhikari et al., 2017). The light yields reach ~14–15 photoelectrons/keV, generally exceeding historical DAMA performance. Crystals are arranged in a 4×2 array on an acrylic stage, within the innermost shield.

Layered Passive and Active Shields

Shield/Veto	Material/Design	Function
Plastic Scintillator	EJ-200, 3-cm panels	Muon veto, measures modulating cosmic-ray muons to ~99.9% efficiency (Collaboration et al., 2017)
Lead Castle	20 cm, low-U/Th core	External γ-ray attenuation
Copper Box	3 cm OFC	Suppresses γ/n backgrounds, radiopure interior
Liquid Scintillator	2200 L LAB-based	Active veto for γ/β events, tags internal backgrounds (e.g., 3 keV from ^40K); tagging efficiency 72–75% (Adhikari et al., 2017)
Acrylic Vessel	1-cm wall	Contains LS, forms inner active shield

The whole assembly is situated under ∼700 m rock overburden, reducing the muon flux to ≈3.8×10⁻⁷ cm⁻² s⁻¹ (Adhikari et al., 2017, Collaboration et al., 2017).

3. Signal Acquisition, Vetoes, and Environmental Monitoring

Data Acquisition and Triggering

PMT signals: Digitized at 500 MSPS, 12 bit FADC (full waveform, 8 μs with 2.4 μs pre-trigger window) for crystal channels; charge-sensitive ADCs (SADC, 64 MSPS) for LS/muon triggers (Collaboration et al., 2018).
Crystal trigger: Both PMTs must exceed a ~single-photoelectron threshold within 200 ns. Global triggers synchronize all channels via a central TCB (Collaboration et al., 2018).
LS/muon triggers: Integrated charge thresholds; coincidence (within 400 ns) between PMTs in adjacent panels or LS PMTs for muon confirmation, with efficiency >99.9% and sub-0.3% background(Collaboration et al., 2017).

Environmental Controls

A centralized system measures:

Temperature (ΔT < 0.06 °C, ΔTm_room < 0.09 °C), relative humidity (ΔRH_room ~ 1%), radon (~36.7 Bq/m³), HV stability (±0.1%).
Slow monitoring logs (InfluxDB, Grafana) allow real-time visualization and store redundant copies (local + AWS) (Kim et al., 2021).
Multiple neutron detectors (LS, ^3He tubes) and dust/air particle counters further control systematic backgrounds.

Environmental stability is critical, since the sought annual modulation has amplitude at the percent level; e.g., the light yield decreases at –0.3%/°C (Kim et al., 2021).

4. Background Modeling and Subtraction

COSINE-100 achieves comprehensive background modeling across 0.7–4000 keV (Yu et al., 19 Aug 2024):

Monte Carlo Simulations: Geant4-based, modeling electromagnetic, radiative and full decay chains, and “turning on/off” specific radionuclides and detector materials. Multiple energy calibration points establish nonproportionality corrections—scintillation nonlinearity causes ~20% deviation at 1 keV.
Internal/Self-Consistent Measurements: Alpha backgrounds (from ^210Pb, ^210Po, ^228Th), measured in both α and β/γ channels, are cross-checked by pulse shape criteria and time-correlation. Quenching factors for α decays exhibit two distinct populations (Q₁, Q₂), modeled as energy-dependent linear functions (Adhikari et al., 2023).
External/Material Assay: HPGe and alpha counting data constrain input radioactivity levels in PMTs, quartz, PTFE, cable ties; contribute to fit constraints on e.g. ^40K, ^210Pb.
Cosmogenic Activation: Time-dependent monitoring of isotopes (^125I, ^109Cd, ^3H, ^22Na) using characteristic signatures (e.g., 25.5 keV and 88 keV for ^109Cd, triple coincidence for ^22Na) and semi-analytic production–decay models to extrapolate underground contamination and sea-level production rates (Souza et al., 2019).

In the crucial 2–6 keV region, ^210Pb and ^3H dominate, with total background ≈3.5 dru (counts/kg/day/keV) (Adhikari et al., 2018).

5. Data Analysis Methodology and Performance Metrics

Signal Extraction and Threshold Lowering

Pulse-shape discrimination combines classic parameters (mean-time, tail-to-total ratio) with likelihood-based comparisons to scintillation and noise templates (Adhikari et al., 2020, Yu et al., 26 Aug 2024). Key PSD parameters are further optimized using machine learning:

Boosted Decision Trees (BDT): Trained on calibration and physics data for stage-wise background suppression.
Multi-Layer Perceptron (MLP): Integrates both time-domain and frequency-domain (FFT-based) likelihoods to differentiate multiple noise types down to 0.7 keV (Yu et al., 26 Aug 2024).

Trigger timing and energy calibration are precisely managed to account for differences between passively- and self-triggered events.

Summary of Key Quantitative Performance Metrics

Metric	Value/Result
Energy threshold (latest)	0.7 keV (Yu et al., 26 Aug 2024)
Light yield (best crystals)	~14–15 photoelectrons/keV (up to ×2 better than DAMA)
Selection efficiency at 0.7 keV	~20% for nuclear recoils (Yu et al., 26 Aug 2024)
Continuous data-taking fraction	>95% high-quality data
Muon flux (measured)	328 ± 1(stat) ± 10(syst) muons/m²/day (Collaboration et al., 2017)
LS-veto tagging efficiency (for ^40K)	72–75%
Background (2–6 keV, selected runs)	Down to 2 counts/day/kg/keV (post-veto and noise cuts)

6. Results, Exclusion Limits, and Impact

Modulation Searches

COSINE-100 achieves its design goal of directly testing the DAMA/LIBRA annual modulation signal (Adhikari et al., 2017, Collaboration et al., 2019, Collaboration et al., 2021, Carlin et al., 20 Sep 2024):

No significant annual modulation signal is detected after 6.4 years of data (effective exposure 358 kg·years) with a <1% noise contamination at 0.7 keV threshold, challenging DAMA/LIBRA’s claim at >3σ significance in the critical 1–3 keVee range (Carlin et al., 20 Sep 2024).
Results disfavor the hypothesis that the observed DAMA modulation originates from WIMP interactions under the standard halo model; COSINE-100’s exclusion limits cover and surpass the full DAMA/LIBRA 3σ allowed region in both spin-independent and spin-dependent WIMP parameter space (Yu et al., 23 Jan 2025).
Upper limits set by COSINE-100 are enhanced by an order of magnitude over prior results, with additional sensitivity in the sub-GeV WIMP mass range; the inclusion of the Migdal effect further bolsters the low-mass reach (Yu et al., 23 Jan 2025).

Background Understanding

Alpha background analysis confirms two-probability quenching factors for ^210Po; Monte Carlo modeling agrees with total measured alpha rates, and precise activity quantification (e.g., ^210Po in C1: 1.55 mBq/kg [Q₁], 1.38 mBq/kg [Q₂], sum matches 3.20 mBq/kg) further strengthens background reliability (Adhikari et al., 2023).

BSM Sensitivity

Searches for solar axions, dark photons, KK axions, and boosted dark matter set competitive upper limits—e.g., dark photon kinetic mixing parameter ε < 1.61×10⁻¹⁴ (215 eV mass), axion-electron coupling g_ae < 1.61×10⁻¹¹ (<0.2 keV) (Adhikari et al., 2023).

7. Technical Innovations and Future Prospects

COSINE-100’s analysis and operation established several technical benchmarks for rare-event searches:

Full integration of DAQ, trigger, veto, and environmental monitoring with MHz–GHz bandwidths and redundancy (Collaboration et al., 2018, Kim et al., 2021).
Cross-calibrated, nonproportionality-corrected background models (0.7–4000 keV) using internal and external calibration lines and data-driven constraints (Yu et al., 19 Aug 2024).
Consistent, automated multi-year environmental and detector stability monitoring, mitigating systematic uncertainties to below modulation signal amplitude (Kim et al., 2021).

The future COSINE-200 upgrade is planned to feature further background reduction, enhanced light yield (targeting ~22 NPE/keV), improved threshold (possibly ~0.2 keV), and a doubled detector mass (~200 kg) (Adhikari et al., 2023).

COSINE-100’s null result, in conjunction with these technical advances, provides a crucial evidence base for the global direct dark matter search, refuting a dark matter interpretation of the DAMA/LIBRA annual modulation signal under standard assumptions and setting new standards for background understanding, sensitivity, and analytical rigor.