Toward triggered generation of indistinguishable single-photons from MoTe 2 quantum emitters

Single-photon sources operating in telecom band are fundamental for long-distance optical quantum communication and information processing, and this work demonstrates a reproducible and systematic approach for generating near-infrared (1090–1200 nm) quantum emitters in bilayer MoTe2 using deterministic strain and defect engineering.

How it works

The generation of localized quantum emitters (QEs) in bilayer MoTe2 is achieved by combining directional strain engineering with electron-beam-induced defect activation. This process creates stable and spectrally narrow emitters across multiple samples operating in the 1090–1200 nm spectral range. The fabrication involves a highly reflective substrate consisting of a bottom distributed Bragg reflector (DBR) made of GaAs and AlAs layers, capped with an Al2O3 layer to maximize photon extraction efficiency. Bilayer MoTe2 flakes are mechanically exfoliated and then dry-transferred onto the substrate with prepatterned star-shaped nanopillars. During transfer, the bilayer conforms to the underlying threepointed star nanopillars, forming nanowrinkles, which act as localized hosts for polarized quantum emitters via directional strain confinement.

Emitter Properties and Characterization

The resulting quantum emitters exhibit several key characteristics:

1. Strong linear polarization (DOLP > 70%).

2. Sub-nanosecond lifetimes (τ ⩽ 450 ps).

3. High single-photon purity (g(2)(0) < 0.1).

4. Resolution-limited emission (200 µeV linewidth).

Optical characterization reveals that the emission from the strained wrinkle regions shows two sharp peaks attributed to localized quantum emitters (QEA and QES), contrasting with broader emissions from the unstrained regions assigned to delocalized excitonic transitions. Quasi-resonant excitation (1080 nm) yields a narrower emission line (230 µeV) with significantly suppressed background compared to above-band excitation (650 nm, 280 µeV).

Electrical Tuning and Dynamics

Electrostatic biasing enables active Stark tuning over a 3 meV range. This tuning allows for spectral alignment of emission lines between distinct sources. Furthermore, electrical biasing suppresses photon bunching and significantly shortens radiative lifetimes. Under CW quasi-resonant excitation, the linewidth-to-lifetime ratio (R) is reduced to RQR ∼ 154 when using experimental linewidths, and down to RQR ∼ 100 when using transform-limited linewidths. The applied electric field also induces a reduction in the lifetime from 440 ps at 0 V to 240 ps at +15 V, attributed to a bias-induced modulation of the oscillator strength.

Photon Indistinguishability

Two-photon interference measurements reveal Hong-Ou-Mandel (HOM) visibility. The results show an HOM visibility of VHOM ∼ 10% and up to VHOM ∼ 40% with post-selection by temporal filtering, representing the highest reported indistinguishability for any TMD quantum emitters and the first such demonstration in the near-infrared regime. The emitter's coherence time is estimated at τc = 63±12 ps.

Conclusion

The main finding of this study is the demonstration of HOM interference between consecutively emitted photons from a single MoTe2 quantum emitter, achieving a visibility of ∼10% within a 12.5 ns integration window and exceeding ∼30% with sub-200 ps postselection temporal filtering. This represents the first such demonstration for the MoTe2 platform in the near-infrared range. The results establish MoTe2 as a viable platform for tunable, low-noise, high-purity single-photon sources capable of generating indistinguishable photons for photonic quantum technologies at telecom wavelengths.

The gist

Two-photon interference measurements reveal a Hong–Ou–Mandel visibility of VHOM ∼ 10%, and up to VHOM ∼ 40% with post-selection by temporal filtering, representing the highest reported indistinguishability for any TMD quantum emitters and the first such demonstration in the near-infrared regime.

Key Findings Summary

- Deterministic approach using directional strain engineering with electron-beam-induced defect activation.

- Emitters exhibit DOLP up to 70%, sub-nanosecond lifetimes (τ ⩽ 450 ps), and high purity (g(2)(0) < 0.1).

- Stark tuning over a 3 meV range is achieved via electrostatic biasing.

- Ratios of experimental to transform-limited linewidths reach as low as R ∼ 55.

- HOM visibility reaches >30% with post-selection temporal filtering.

**- The first demonstration of photon indistinguishability from a TMD-based quantum emitter in the near-infrared regime.

As a fastidious and diligent researcher, I have analyzed this paper on MoTe2 quantum emitters and identified several high-impact areas where advancements in AI systems could significantly accelerate discovery, characterization, and engineering of quantum materials.

Here are the specific improvements for AI systems based on this research:

) Specific Improvements for AI Systems:

1. [AI System Improvement] Develop a specialized Deep Learning model (e.g., Graph Neural Network or Variational Autoencoder) trained on the structural data (strain maps, defect locations, and resulting PL spectra/polarization profiles from Fig. 1).

2. [AI System Improvement] Implement a Reinforcement Learning (RL) agent to optimize the deterministic strain engineering parameters (nanopillar geometry, e-beam dose) required to maximize the degree of linear polarization (DOLP > 70%) and minimize spectral diffusion/linewidth broadening (maximizing R value).

3. [AI System Improvement] Create a predictive model using Bayesian Inference or Gaussian Processes that correlates fabrication parameters with key performance indicators like single-photon purity, radiative lifetime, and Hong-Ou-Mandel visibility (VHOM). This model should specifically predict the optimal operating voltage range for Stark tuning to achieve desired spectral alignment.

4. [AI System Improvement] Implement a sophisticated signal processing pipeline using Convolutional Neural Networks (CNNs) to perform automated time-resolved photoluminescence (TRPL) analysis, distinguishing between fast radiative decay components and slow non-radiative/trapping components (e.g., identifying the 450 ps vs. 2.4 ns decay in QEA).

5. [AI System Improvement] Design a generative AI model to simulate and predict the optical response of MoTe2 structures under various environmental stressors, such as hBN encapsulation (Fig. 3), by learning the relationship between strain/confinement geometry and changes in exciton fine-structure splitting (FSS) and polarization anisotropy (DOLP).

6. [AI System Improvement] Integrate an AI module into the quantum photonic simulation framework to predict the impact of electrical biasing on photon indistinguishability (VHOM) by learning the non-linear relationship between bias voltage, bunching time, and coherence time.

) What the Improved AI System Can Do:

The improved AI system can perform the following advanced tasks:

1. [Materials Discovery] It can autonomously screen millions of theoretical MoTe2 structural configurations (strain profiles, defect placement) to predict which geometry will yield a quantum emitter with high intrinsic polarization and minimal spectral diffusion before any fabrication is attempted.

2. [Process Optimization] The RL agent can dynamically adjust the e-beam defect activation parameters in real-time during fabrication to steer the resulting QEs toward the desired near-infrared emission wavelength (1090–1200 nm) while simultaneously maximizing single-photon purity, significantly reducing experimental trial-and-error.

3. [Predictive Engineering] Researchers could input a target spectral alignment requirement (e.g., matching two distinct emitters) and the AI would instantly suggest the optimal external electric field bias needed to achieve this spectral tuning, drastically speeding up the development of tunable quantum gates or interferometers.

4. [Automated Quality Control] The CNN-based TRPL analysis can automatically assess the quality of a fabricated emitter by classifying its decay profile, immediately flagging samples exhibiting unwanted slow non-radiative traps (e.g., identifying the 2.4 ns component) versus high-quality fast emitters (130–450 ps), allowing for immediate rejection or targeted post-processing.

5. [Design Guidance] The generative AI model can guide the design of next-generation quantum platforms by predicting how encapsulation (hBN vs. SiO2) will affect critical properties like FSS and polarization, helping engineers choose the optimal dielectric environment for a specific application's requirements (e.g., maximizing polarization for entanglement distribution).

6. [Indistinguishability Roadmap] The AI can map the indistinguishability landscape, predicting which combination of excitation power, integration time, and post-selection filter will yield the highest HOM visibility (>30%), effectively creating a guided protocol for achieving benchmark performance in MoTe2 quantum communication.

Single-photon sources operating at telecom wavelengths are fundamental components for long-distance optical quantum communication and information processing. Two-dimensional (2D) transition metal dichalcogenides (TMDs) offer a promising platform for such sources, but their development has been hindered by limited spectral range and poor single-photon indistinguishability. Here, we demonstrate a reproducible and systematic approach for generating near-infrared (1090-1200 nm) quantum emitters in bilayer MoTe 2 using deterministic strain and defect engineering. These emitters exhibit strong linear polarization (DOLP >70%), sub-nanosecond lifetimes (τ about 130-450 ps), high single-photon purity with triggered g(2)(0) values as low as about 0.01 (about 0.16) under p-shell (quasi-resonant) excitation, and resolution-limited emission (about 150 μ eV). Electrostatic biasing enables tuning over a about 3 meV range, suppresses photon bunching, and significantly shortens radiative lifetimes, yielding narrow emission with ratios of experimental to transform-limited linewidths as low as R about55. Most notably, two-photon interference measurements reveal a Hong-Ou-Mandel visibility of V HOM about 7.1 % (3.6 %), and up to V HOM about 60 % (about 40 %) with post-selection by temporal filtering under p-shell (quasi-resonant) excitation. To our knowledge, this presents the highest reported indistinguishability for TMD quantum emitters and the first such demonstration for MoTe 2 platform. These results establish MoTe 2 as a viable platform for tunable, low-noise, high-purity single-photon sources with state-of-the-art indistinguishability for TMD quantum emitters, paving the way for their integration into telecom-compatible quantum photonic technologies.