Hybrid Integrated QFC Chip Overview
- Hybrid integrated QFC chips are quantum photonic circuits that combine dissimilar material subsystems to achieve nonlinear frequency conversion and on-chip spectral tuning.
- They employ architectures such as thin-film lithium niobate microrings and ridge waveguides to bridge telecom and visible/near-visible wavelengths with high conversion efficiencies.
- Integration strategies like transfer printing and butt-coupling enable precise subsystem assembly and facilitate quantum networking by mitigating spectral mismatches and losses.
A hybrid integrated QFC chip is a quantum photonic integrated circuit in which frequency-domain functionality is realized by combining dissimilar material systems or separately fabricated photonic subsystems on a single platform. In the literature considered here, the term spans two closely related but technically distinct classes of devices: true quantum frequency conversion chips, which translate photons between widely separated wavelength bands through nonlinear optical processes such as sum-frequency generation, and hybrid quantum photonic chips that instead provide in-situ wavelength tuning, spectral matching, or coherent frequency-bin mixing without implementing a nonlinear wavelength-conversion stage (Wang et al., 4 Sep 2025). The distinction is consequential because the underlying physics, device architectures, and application targets differ, even when all of these systems are described as hybrid integrated quantum photonics.
1. Scope and terminological boundaries
True QFC is explicitly a pump-driven frequency-translation process. The thin-film lithium niobate device reported in 2025 connects the telecom band around to the visible band around $630$– using a pump, while the LNOI nanophotonic device reported in 2022 connects telecom-band photons around to the near-visible band around by sum-frequency generation with a pump (Wang et al., 4 Sep 2025, Wang et al., 2022). In both cases, the chip performs wavelength translation through -mediated nonlinear optics and is therefore a QFC device in the strict sense.
By contrast, several hybrid chips that are highly relevant to the same technological trajectory do not perform quantum frequency conversion. The CMOS-silicon/quantum-dot platform with optically driven Cr heating pads demonstrates in-situ thermal wavelength tuning and wavelength matching of quantum-dot single-photon sources, but “does not implement quantum frequency conversion (QFC) in the nonlinear-optics sense)” because there is “no pump-driven frequency mixing,” “no conversion between widely separated wavelengths,” and no or conversion stage (Katsumi et al., 2019). Likewise, the large-scale quantum-dot–lithium-niobate architecture with 20 deterministic single-photon sources provides local strain tuning and on-chip two-photon interference, but “there is no quantum frequency conversion functionality reported here” (Wang et al., 31 Mar 2025). The silicon quantum frequency processor reported in 2026 is again distinct: it monolithically integrates coherent frequency mixing, pump rejection, and programmable spectral control for frequency-bin quantum information processing, rather than a standalone nonlinear wavelength translator (Congia et al., 15 Feb 2026).
A common misconception is therefore to treat any hybrid chip that changes or aligns optical spectra as a QFC chip. The cited work shows that wavelength matching, local tuning, coherent electro-optic bin mixing, and nonlinear frequency translation are adjacent but non-identical functions. A precise usage reserves “QFC chip” for devices that convert photons from one wavelength to another via a nonlinear optical process.
2. Hybrid integration architectures and material systems
The defining architectural feature of a hybrid integrated QFC chip is the co-integration of subsystems that are fabricated separately and then assembled to exploit complementary material advantages. The 2025 electrically pumped QFC chip combines a single-mode distributed feedback laser chip that generates the $630$0 pump with a thin-film lithium niobate photonic chip containing a periodically poled microring resonator. The two chips are edge-coupled or butt-coupled with about $630$1 coupling efficiency, and the laser chip is rotated by $630$2 to match the polarization of the TFLN waveguide mode (Wang et al., 4 Sep 2025). This is hybrid integration in a strict platform-combination sense: pump generation and nonlinear frequency conversion reside on different materials but operate as a single electrically pumped unit.
The nonlinear TFLN circuit itself is built on a $630$3-cut wafer with a $630$4 thick MgO-doped lithium niobate film, a $630$5 buried $630$6 layer, and a silicon substrate. Its core element is a periodically poled lithium niobate microring resonator in a double-pulley add-drop configuration, with a waveguide width of $630$7, radius $630$8, and etch depth $630$9 (Wang et al., 4 Sep 2025). The LNOI nanophotonic QFC chip uses a different geometry: a ridge waveguide on Z-cut lithium niobate thin film bonded to a 0-thick thermally grown 1 layer on silicon, with LN thickness 2, top width 3, ridge height 4, sidewall angle 5, waveguide length 6, and quasi-phase-matching period 7 (Wang et al., 2022).
Closely related hybrid architectures centered on quantum emitters demonstrate the same design philosophy. On CMOS silicon, InAs/GaAs quantum-dot single-photon sources are fabricated separately in GaAs and then transfer-printed onto glass-cladded silicon waveguides; a separate Cr-on-silicon optical heating pad is subsequently transfer-printed onto the source for local thermal tuning (Katsumi et al., 2019). In the GaAs/TFLN quantum-network platform, self-assembled InGaAs quantum dots in GaAs waveguides are transfer-printed onto X-cut TFLN circuits containing low-loss waveguides, multimode-interference beam splitters, grating couplers, and local electrodes (Wang et al., 31 Mar 2025). The earlier strongly coupled quantum-dot–cavity system on CMOS silicon follows the same heterogeneous logic, placing a GaAs 1D photonic crystal nanobeam cavity containing a single InAs quantum dot above a silicon wire waveguide by transfer printing (Osada et al., 2018).
This broader architectural pattern also appears outside QFC proper. The hybrid QKD transceiver chip combines low-loss silicon nitride interferometric circuitry with indium phosphide electro-optic phase modulators, while the vacuum-state QRNG combines an SOI photonic chip with an InGaAs balanced homodyne detector module and a high-bandwidth transimpedance amplifier in one packaged unit (Dolphin et al., 2023, Bai et al., 2021). These systems do not perform frequency conversion, but they illustrate the general reason hybrid integration is used in quantum photonics: no single material platform simultaneously optimizes low loss, high-speed actuation, single-photon generation, and detection.
3. Frequency conversion, spectral tuning, and coherent frequency control
In true QFC chips, the operative mechanism is three-wave mixing in periodically poled lithium niobate. The TFLN microring device employs quasi-TM fundamental modes for the 8 signal, 9 pump, and 0 sum-frequency output so as to exploit the large 1 nonlinear coefficient. Periodic poling is introduced by an external electric-field poling method and enforces type-0 quasi-phase matching through the azimuthal-mode relation
2
with 3, 4, 5, and 6 for the designed triple resonance (Wang et al., 4 Sep 2025). The same work models the three-mode process with a Hamiltonian of the form
7
and gives the maximum conversion efficiency for the double-pulley add-drop resonator as
8
The LNOI waveguide QFC chip uses the standard periodically poled conversion model
9
where 0 is the conversion efficiency, 1 the maximum efficiency, 2 the normalized conversion efficiency, 3 the pump power at the output facet, and 4 the quasi-phase-matching length (Wang et al., 2022).
The two lithium-niobate implementations differ substantially in pump strategy and noise engineering. The TFLN microring uses a hybrid-integrated DFB laser and achieves electrically pumped operation on chip (Wang et al., 4 Sep 2025). The LNOI nanophotonic waveguide instead uses a long-wavelength 5 pump, intentionally far detuned from the telecom signal, to suppress Raman noise processes that are problematic in earlier telecom-pumped thin-film LN QFC devices; the dominant noise source in the high-efficiency regime is identified as SHG-SPDC, with the total on-chip noise count rate exhibiting an approximate slope of 6 on a log-log plot at 7 (Wang et al., 2022).
Hybrid chips that are not true QFC systems nevertheless implement important spectral-control primitives. In the CMOS silicon/quantum-dot platform, a continuous-wave Ti:sapphire laser at 8 is focused onto a Cr-coated heating pad, the pad absorbs the light, local temperature rises, and the QD emission redshifts; the 9 wavelength is chosen to suppress direct excitation of the QDs (Katsumi et al., 2019). In the GaAs/TFLN network platform, local spectral tuning is produced by the piezoelectric response of X-cut TFLN, which generates a localized anisotropic strain field in the bonded GaAs waveguide. The reported relations are
0
for the induced strain and
1
for the resulting band-gap shift (Wang et al., 31 Mar 2025). The silicon quantum frequency processor provides yet another mechanism: high-speed phase modulators driven at the mode-spacing frequency implement coherent nearest-neighbor frequency conversion between bins, with 2 used for frequency-beamsplitter experiments and 3 for quantum-walk and source-matched experiments (Congia et al., 15 Feb 2026).
4. Representative implementations and quantitative benchmarks
The reported devices span nonlinear wavelength conversion, thermal wavelength matching, piezoelectric spectral tuning, and coherent frequency-bin processing. Their metrics are best interpreted by separating these functions rather than collapsing them into a single category.
| Device | Demonstrated function | Representative reported metrics |
|---|---|---|
| TFLN microring with hybrid DFB pump (Wang et al., 4 Sep 2025) | Telecom-to-visible QFC | 4, 5 pump, 6 on-chip QE, 7 noise |
| LNOI ridge-waveguide QFC (Wang et al., 2022) | Telecom-to-near-visible QFC | 8 internal conversion efficiency, 9, 0 on-chip NCR |
| CMOS silicon + QD + Cr heater (Katsumi et al., 2019) | Thermal wavelength tuning and matching | 1 tuning range, 2 at 3, matching at 4 |
| GaAs/TFLN quantum-network chip (Wang et al., 31 Mar 2025) | Local spectral tuning of QD sources | 20 deterministic SPSs, 5 tuning range, 6 TPI visibility |
| Silicon quantum frequency processor (Congia et al., 15 Feb 2026) | On-chip coherent frequency-bin processing | success probability 7, fidelity 8, Bell-state tomography fidelity 9 |
Within true QFC, the TFLN microring emphasizes extreme pump efficiency. It reports an ultra-high normalized conversion efficiency of 0, an ultra-low pump power of 1, and an on-chip quantum efficiency of 2, which the paper states is more than two orders of magnitude lower in pump requirement than traditional straight-waveguide schemes while remaining comparable in on-chip QE (Wang et al., 4 Sep 2025). The LNOI nanophotonic waveguide instead emphasizes the simultaneous achievement of high efficiency and low noise, reporting 3 internal conversion efficiency, 4 on-chip noise count rate, preservation of nonclassical statistics with 5 after conversion, and an upconversion single-photon detector with 6 system detection efficiency and 7 noise count rate (Wang et al., 2022).
The non-QFC hybrid chips quantify different performance frontiers. The CMOS silicon/quantum-dot system achieves a calculated emitter-waveguide coupling efficiency of 8 for a vertical cavity-waveguide distance 9, while the experimentally inferred cavity-waveguide coupling is about 0 from measured 1-factors; wavelength tuning is performed at 2 with a maximum tuning range of 3, and wavelength matching between two dissimilar sources occurs at 4 heating power (Katsumi et al., 2019). The GaAs/TFLN network platform reports a calculated 5-factor of 6 into the fundamental waveguide mode, single-photon purity of 7, a 8 local tuning range for 9 swept from 0 to 1, and on-chip two-photon interference visibility
2
between two spatially separated QD single-photon sources connected by 3-long waveguides (Wang et al., 31 Mar 2025). The silicon quantum frequency processor, finally, integrates a source and processing network on a 4 die and demonstrates tunable frequency beamsplitters with success probabilities exceeding 5, fidelities above 6, correlated and anticorrelated two-photon quantum walks, and on-chip Bell-state tomography with fidelity 7 (Congia et al., 15 Feb 2026).
5. Integration strategies and their role in quantum networking
Hybrid integrated QFC chips are motivated less by packaging convenience than by system-level incompatibility between the optimal material choices for different quantum photonic functions. Lithium niobate offers strong 8 nonlinearity, a wide transparency window, and electro-optic functionality; III–V quantum dots offer deterministic single-photon emission and, in cavity-QED implementations, strong single-emitter nonlinearity; silicon and silicon nitride provide mature low-loss routing and scalable fabrication; InP provides high-speed electro-optic modulation; InGaAs photodiodes provide high-speed detection (Wang et al., 4 Sep 2025, Wang et al., 31 Mar 2025, Osada et al., 2018, Dolphin et al., 2023, Bai et al., 2021).
The assembly methods differ according to the required subsystem. Transfer printing is central in the QD-based silicon and TFLN platforms. The 2019 CMOS-silicon/QD device uses a PDMS transparent rubber stamp for pick-and-place transfer printing and reports pick-up and placement success rates close to 9 with position deviation between nanobeam and waveguide below $630$00 (Katsumi et al., 2019). The 2025 GaAs/TFLN platform also uses a PDMS stamp and reports alignment precision within hundreds of nanometers and $630$01 device yield for the fabricated photonic channels (Wang et al., 31 Mar 2025). The 2018 strongly coupled QD–cavity device similarly relies on a silicon rubber stamp for transfer printing, with alignment precision better than $630$02 and cavity quality factor maintained around $630$03 after transfer printing (Osada et al., 2018).
In the TFLN QFC chip, hybridization occurs at the pump source rather than the emitter interface. The distributed-feedback laser is butt-coupled to the nonlinear circuit, permitting current-injected pump generation on chip and replacing the external bulky pump laser that limited earlier TFLN QFC scalability (Wang et al., 4 Sep 2025). This architectural shift is directly aligned with quantum-network applications, because telecom-visible interfaces are relevant to single-photon frequency translation, integration with quantum memories and quantum networking nodes, multimode or multinode quantum repeaters, and distributed quantum computing (Wang et al., 4 Sep 2025, Wang et al., 2022).
The QD-based spectral-tuning platforms target a complementary networking problem: source matching. Quantum-dot emitters are position-random and spectrally random, which obstructs interference between independent sources. Transfer printing plus local tuning is presented as a route to integrate preselected emitters at arbitrary locations and then compensate residual spectral mismatch on chip (Katsumi et al., 2019). The large-scale GaAs/TFLN system pushes this logic further by integrating 20 deterministic single-photon channels and demonstrating spectral alignment and Hong–Ou–Mandel-type interference between spatially separated sources. This suggests a division of labor in future networked chips: QFC stages can connect disparate wavelength bands, while hybrid source-tuning stages enforce indistinguishability within a band.
6. Limitations, misconceptions, and technical directions
The principal limitation in current hybrid integrated QFC and frequency-control chips is not the absence of a working physical effect, but the coexistence of multiple non-idealities: coupling loss, internal loss, thermal or spectral crosstalk, and system-level packaging constraints. In the electrically pumped TFLN QFC device, the DFB-to-TFLN butt coupling is about $630$04, and the double-pulley design, while improving visible extraction, adds some coupling loss and lowers intrinsic $630$05 somewhat (Wang et al., 4 Sep 2025). In the LNOI upconversion detector, the system detection efficiency is limited mainly by insertion losses, including $630$06 fiber-to-chip coupling loss and $630$07 filtering and space-fiber coupling loss (Wang et al., 2022). In the silicon quantum frequency processor, the main performance bottleneck is the waveshaper loss of about $630$08–$630$09 per channel, with each carrier-depletion modulator contributing roughly $630$10 and grating couplers about $630$11 (Congia et al., 15 Feb 2026).
The non-QFC spectral-control chips exhibit a different limitation profile. The thermally tuned QD-on-silicon device reaches only $630$12 maximum tuning range, which is attributed to heat dissipation into the glass cladding from the GaAs layer. It also shows thermal crosstalk: tuning QD2 causes a slight shift of QD1, likely due to insufficient thermal separation, with trench-based thermal insulation proposed as a remedy (Katsumi et al., 2019). The same work reports $630$13 under $630$14 heating, confirming antibunching but also indicating background cavity emission from off-resonant QDs (Katsumi et al., 2019). In the earlier strongly coupled QD–cavity device on CMOS silicon, the experimentally inferred waveguide coupling efficiency is only $630$15, though the paper identifies a path to $630$16 if the unloaded cavity $630$17 improves to $630$18 and the design is adjusted to loaded $630$19 at a gap of about $630$20 (Osada et al., 2018).
A second misconception is to equate hybrid integration with monolithic integration. The 2026 silicon quantum frequency processor is explicitly monolithic, integrating source, filter, phase modulators, and pulse shaper on the same silicon photonic chip (Congia et al., 15 Feb 2026). The QFC chips based on TFLN and QD transfer printing are instead hybrid because they assemble separately fabricated subsystems or combine dissimilar platforms. Both strategies are active in the field, and the choice depends on whether the dominant engineering challenge is material incompatibility, packaging of active subsystems, or scaling of passive and programmable photonic networks.
The present literature points toward convergence rather than competition between these approaches. The TFLN QFC chip identifies lower-loss phase modulators, including heterogeneous thin-film lithium niobate on silicon, as a route for improved frequency processors (Congia et al., 15 Feb 2026). The LNOI QFC chip explicitly points to integration with superconducting nanowires, SPADs, and other materials such as Si, NbN, and $630$21 (Wang et al., 2022). The GaAs/TFLN network platform notes that lithium niobate can support frequency conversion in other contexts, even though its demonstrated device does not (Wang et al., 31 Mar 2025). This suggests a plausible next stage in which deterministic single-photon sources, low-loss routing, programmable frequency-bin processing, nonlinear wavelength translation, and on-chip detection coexist within heterogeneous quantum photonic architectures rather than as isolated chip classes.