Self-assembled quantum dots (QDs) play a crucial role in generating quantum states of light, including single photons [1–3], entangled photon pairs [4–6], and cluster states [7–9]. To fully harness QDs’ capabilities, it is essential to control the preparation of low-multi-photon contributed, indistinguishable photons and efficiently extract them from the semiconductor bulk for collection by a free-space lens or optical fiber. The initial hurdle of efficient preparation has been effectively addressed by employing fast pulsed resonant excitation, eliminating dephasing and time jitter, and resulting in near-unity indistinguishable single photons [10]. The subsequent challenge of efficient extraction has been addressed by embedding QDs in Purcell-enhanced photonic cavities like micropillars [11], open cavities [12], circular Bragg gratings [13], and photonic crystal waveguides [14]. The combination of these two strategies produces pure and indistinguishable photons with high extraction efficiency [15–19].

However, a critical obstacle to achieving high end-to-end efficiency lies in preventing the excitation laser pulse from entering the collection path. Typically, separating the pump laser from the signal using pump-probe cross-polarization results in at most a 50% signal reduction with isotropic cavities [10,17]. Recent advancements have alleviated this issue by developing anisotropic cavities, induced by geometry ellipticity [18] or material birefringence [16,20]. In this approach, a circularly polarized quantum emitter is brought into resonance with one of the linearly polarized cavity modes for efficient polarized photon extraction, while the perpendicular one is used for laser exclusion. The state-of-the-art QD-based single-photon source in an open microcavity has achieved high end-to-end efficiency (>71.2%) and high photon indistinguishability (>98.5%) [20]. However, optimizing the pulse width remains crucial to balance excitation efficiency [20] and the risk of QD re-excitation [21]. Meanwhile, polarized cavities’ limitation on variable polarization of single photons hinders the scalability of complex photonic states like cluster states [7].

A straightforward approach to overcome this limitation is resonant excitation of the emitter via non-cavity modes perpendicular to the collection axis, a method used for QDs embedded in planar cavities [22] and waveguides [23,24]. Despite its attractiveness, implementing this method in high-Purcell-enhanced vertical cavities remains challenging due to the isolated structure of micropillars [25,26]. Huber et al. attempted to integrate connected waveguides with micropillars; however, it resulted in fundamental mode leakage and laser scattering at the micropillar-waveguide interface, achieving only a Purcell factor of 2.5 [27]. For open cavities, introducing excitation at another axis on top of the existing two 3D independent adjustments of mirrors is challenging. Here, we propose a feasible solution by introducing a novel design that features a micrometer-scale parabolic lensed-defect between two distributed Bragg reflectors (DBRs). This Gaussian-shaped defect in a Fabry–Perot cavity was initially conceptualized in 2013 [28] and later implemented in QD systems [29,30]. Thanks to its planar microcavity design, it suppresses unwanted pump lasers and achieves filter-free Rabi oscillations with an optimal autocorrelation value of g(2)(0)=0.0322(2), indicating no spectral or polarization filters are used. Moreover, combining resonant excitation with the Purcell effect enables the generation of single photons with a simultaneous high extraction efficiency of 0.87 and indistinguishability of 0.963(4).

To achieve resonant excitation and effectively discriminate the QD signal from stray laser light, we implemented an orthogonal excitation/detection geometry [Fig. 1(a)]. A single-mode fiber, positioned on a three-axis displacement stage, is placed within a few micrometers of the cleaved sample edge. A tunable, mode-locked Ti:Sa pulsed laser, polarized in-plane, is introduced through the fiber to excite the QDs embedded in a 2-λ monolithic microcavity. This microcavity features a parabolic lensed-defect within a Fabry–Perot distributed Bragg reflector (DBR) cavity. Lateral confinement is provided by the parabolic lensed-defect in the central spacer layer, as shown in the cross-section scanning electron microscope (SEM) image in Fig. 1(b). This design results in vertically emitted photons that are collected by an objective and directly coupled into a single-mode fiber without the need for spectral or polarization filtering. The spatial overlap of the QD emission with the cavity mode is achieved through well-established deterministic positioning techniques [31]. Detailed fabrication processes are provided in Appendix A and our previous report [30]. Theoretically, an extraction efficiency of up to 94.9% and a Purcell factor of 40 are achievable [30].

In experiments, the laser, directed onto the cleaved side facet of the sample, effectively couples into the planar waveguide mode, enabling low-loss transmission [Figs. 1(c) and 1(d)]. Notably, our methodology enables QD excitation at a distance greater than 400 μm from the cleaved edge of the sample. The orthogonal alignment of the resonant excitation laser’s propagation and the QD emissions minimizes upward scattering, eliminating the need for polarization filtering. Importantly, our approach is more versatile than cross-polarization excitation, allowing the QD emission to have any desired polarization for applications like BB84 [32] or specific polarizations for other applications such as boson sampling [33].

To demonstrate our approach, we apply spatially orthogonal resonant excitation to a single-QD-charged exciton (CX) that is deterministically coupled to microcavities with distinct circular and linear polarizations. Figures 2(a) and 2(b) display the photoluminescence (PL) spectra of our microcavity modes under high-power above-band excitation, exhibiting splits into horizontally and vertically polarized (H- and V-polarized) modes at 11.05 GHz and 86.50 GHz in Cavities A and B, respectively. The observed mode splitting is primarily attributed to the residual asymmetric uniaxial strain in the semiconductor material and minor structural asymmetries induced by the dielectric top DBRs, consistent with phenomena observed in open cavities [16,20]. Further adjusting the magnitude of mode-splitting can be achieved by applying strains on the GaAs substrate [34]. The linewidths of the H mode in Cavity A and Cavity B are 0.098 nm and 0.111 nm, corresponding to measured quality factors of 9511 and 8234, respectively. In a birefringent cavity, the spontaneous radiation rate of the exciton’s circularly polarized transition is expected to redistribute into H and V polarizations [18]. Resonance of the QD with the cavity H mode, achieved through minor temperature adjustments, yields predicted polarized spontaneous emission ratios ζH of 0.578 in Cavity A and 0.910 in Cavity B [see insets of Figs. 2(a) and 2(b)].

In the pursuit of creating a highly bright and indistinguishable single-photon source, we conducted pulsed resonance fluorescence for the CX under laser excitation with an 80.1 MHz repetition rate. Time-resolved resonance fluorescence measurements reveal shortened radiative lifetimes for the QDs on resonance with cavities to τon∼134ps [Cavity A, blue line in Fig. 2(c)] and ∼53ps [Cavity B, red line in Fig. 2(c)]. Compared with the average lifetime for the QDs in bulk from the same area [black line in Fig. 2(c)], a Purcell factor Fp of ∼7.5 (19) is achieved in Cavity A (B).

Figure 2(d) presents the detected single-photon flux as a function of driving laser strength, showing a complete Rabi oscillation curve for both microcavities, indicative of coherent two-level system control. Notably, this marks the first observation of pulsed resonance fluorescence for QD-in-microcavity systems under spatially orthogonal excitation, achieved without employing any spectral filter. Under the “π pulse” condition, the avalanche photon detector (APD) records approximately 4.93 Mcps in Cavity A and 6.23 Mcps in Cavity B. Accounting for the setup efficiency and the APD correction factor, an extraction efficiency of 0.63 is obtained for Cavity A and 0.87 for Cavity B. A comprehensive calibration of the system efficiency is detailed in Appendix C. For both circularly and linearly polarized microcavities, spatially orthogonal excitation proves to be an excellent scheme for resonant fluorescence experiments.

We then assess the single-photon characteristics in Cavity B using both spatially orthogonal and polarized-orthogonal excitation strategies [Figs. 3(a) and 3(b)]. In polarized-orthogonal excitation, the QD is excited by a V-polarized laser, and H-polarized single photons are collected. As depicted in Fig. 3(b), only about 4.64×106 photons per second are detected at π pulse. We estimate that the probability of emission into the H-polarized mode, denoted as βH, can be determined to be βH=[Fp/(Fp+1)]×ζH, revealing a value of 0.865 in Cavity B.

To characterize the second-order coherence of the single-photon source, we use a Hanbury Brown and Twiss setup under conditions of I=0.6·ISat by decreasing the excitation power, where ISat represents the intensity of resonant fluorescence at the “π pulse”. The second-order autocorrelation in Fig. 3(c) reveals filter-free g(2)(0) values of 0.0472(2) for spatially orthogonal excitation and 0.0889(7) for polarized-orthogonal excitation, indicating distinct photon antibunching in both cases. Additionally, single-photon coherence is assessed using a Hong–Ou–Mandel (HOM) interferometer, which measures a 12.48 ns time separation between consecutive single-photon emissions, aligning with the laser pulse timing. The photon correlation histograms in Fig. 3(d) for orthogonal and parallel polarizations show raw HOM visibilities of 0.845(2) for spatially orthogonal and 0.731(4) for polarized-orthogonal excitation. Considering imperfect single-photon second-order coherence and an unbalanced beam splitting ratio in the optical setup, we calculate corrected photon indistinguishabilities of 0.966(2) for spatially orthogonal and 0.899(4) for polarized-orthogonal excitation (see Appendix B). This highlights the high coherence of single photons generated using spatially orthogonal excitation in the microcavity which may be attributed to the QDs’ distinct dipole orientations relative to the excitation and collection pathways.

We investigate the origin of the non-vanishing peak at zero-time delay in the QD-in-microcavity system under spatially orthogonal excitation, as shown in Fig. 3(e). The g(2)(0) and HOM visibility (VHOM) values vary with sample temperature. Through slightly increasing the temperature, the wavelength of the QD red shifts while the microcavity remains stable. Tuning the QD away from the cavity resonance leads to increased (decreased) lifetime (Fp and VHOM), indicating that microcavities with high Purcell factors are instrumental in mitigating dephasing. However, the value of g(2)(0) remains at ∼0.048, which we attributed to a slight amount of laser light scattering into the detection channel. This is further supported by the power-dependent behavior of g(2)(0) and HOM visibility in Fig. 3(f), where g(2)(0) increases with higher laser power while HOM visibility remains almost constant. The best autocorrelation value of g(2)(0) is 0.0322(2) at I=0.2·ISat. As shown in Fig. 1(d), the markers on the sample surface and the protruding microcavity cause scattered light, thereby reducing the single-photon second-order coherence. The resulting single-photon source exhibits a simultaneous high extraction efficiency of 0.87, and indistinguishability of 0.963(4) at the condition of “π pulse”.

The planar cavity structure allows the laser used for resonant excitation to effectively couple with the in-plane guided waveguide mode, simultaneously suppressing background scattering. We transition to continuous-wave resonance excitation to observe the Mollow triplet, which is a distinctive feature of resonance fluorescence in a strongly driven and dressed two-level system. Figure 4(a) shows a comprehensive power-dependent PL series from the QD, revealing the emergence of two symmetrically spaced sidebands around the central peak with increasing power. In Fig. 4(b), the extracted Rabi splitting is plotted against the square root of the laser power, confirming the theoretically expected linear relationship and demonstrating a significant Rabi splitting of up to 21.4 GHz. Figure 4(c) demonstrates the linewidth of the Rabi sidebands, exhibiting a clear linear dependence on the squared Rabi frequency. These linear trends strongly suggest phonon-dressed QD Mollow triplet emission within the cavity-quantum electrodynamics (cQED) regime in our sample [35].

In summary, our study introduces a spectral filter-free single-photon source by combining spatially orthogonal resonant excitation with Purcell-enhanced single QDs in a monolithic Fabry–Perot microcavity. This approach effectively separates the excitation laser from the QD emission, eliminating the need for polarization filters and enabling flexible control over the QD’s polarization. These features collectively establish the high-performance attributes of our single-photon source, with an extraction efficiency of 87% and an indistinguishability of 96.3%. The observation of a Mollow triplet with a 21.4 GHz Rabi splitting confirms the presence of a strongly driven and dressed two-level system in our microcavity. To further alleviate residual laser scattering under spatially orthogonal excitation, a surface gold coating is proposed to act as a reflective barrier, reducing laser escape through the surface. Additionally, a thicker cavity or a waveguide-cavity scheme [Fig. 7(a)] could enhance waveguide-fiber coupling efficiency and mitigate beam divergence and edge scattering, thus preserving photon purity. This scheme supports a low-loss waveguide mode [Fig. 7(b)], enabling QD excitation away from the edge and facilitating on-chip pumping. Our work provides a foundation for developing efficient, versatile, and miniaturized quantum light sources, which are easily integrable with fiber optics through direct facet attachment [36] or with photonic integrated circuits via grating coupling [37].