Quantum photonics is set to play a crucial role in advancing information technologies, particularly in the fields of quantum computing, communication, and metrology. Quantum photonics systems may outperform classical supercomputers for certain tasks1 and may also be positioned to enable long-range quantum networks,2 for interconnecting multiple quantum devices,3 for implementing error correction,4 for building quantum circuits for computations and simulations,5^,6 and for generating essential resources for universal quantum computation and communication.7 It also plays a crucial role in precision metrology, facilitating multiparameter estimation8 and enhancing sensitivity in applications such as gravitational wave detection.9

Scaling these quantum systems up is essential for enhancing computational power, improving measurement precision, and extending communication distances. However, the inherent complexity and sensitivity of these systems present significant challenges in achieving scalability. Integrated quantum photonics offers a promising path forward, providing compact and scalable platforms that reduce size, weight, and power consumption compared with traditional free-space optics systems. Integrated quantum photonics platforms also offer phase-stable optical paths, which help minimize environmental disruptions. Even though manufacturing imperfections for a quantum chip are unavoidable, they manifest as static errors rather than stochastic noise, meaning that they do not cause photonic states to evolve randomly if modeled precisely. These static errors can be effectively characterized using classical light,10 enabling high precision in system performance.

Photonics is uniquely suited for quantum networks, where the goal is the reliable transmission of quantum states between distant parties. The best-known example is quantum key distribution (QKD), which enables the secure sharing of cryptographic keys.11 To achieve higher secret key rates, QKD systems must incorporate low-loss modulators operating at high clock rates, efficient fiber-to-chip couplers, and high-efficiency detectors. Moreover, future quantum networks aiming to transmit quantum information globally, i.e., a global quantum internet,12 require the distribution of photons over long distances (more than 100 km) among multiple remote nodes, a task hindered by photon loss that scales exponentially with the channel length. Quantum repeaters13 equipped with matter quantum memories are therefore crucial as quantum information cannot be amplified or cloned due to the no-cloning theorem.14 On the other hand, all-optical quantum repeaters,15 which remove the necessity of matter quantum memory, require large entangled resource states,16 i.e., quantum systems that are nonclassically correlated across many particles. These resource states can be built using fast active feedforwarding techniques in linear optics, i.e., multiplexing, which require high-speed and low-loss electro-optic switches.

Quantum metrology,17^,18 with its ability to surpass classical measurement limits, also stands to benefit significantly from integrated photonics as it requires stability against environmental factors, which can interfere with measurement accuracy. Key properties of the material such as thermal stability and high nonlinearity are essential for enhancing measurement precision. Furthermore, many physical processes involve multiple parameters to estimate, making large-scale integration crucial for advancing quantum metrology. In practical applications, integrated quantum metrology could lead to portable, high-precision instruments for fields such as environmental monitoring and biomedical diagnostics19—capabilities limited to specialized laboratories so far.

The ultimate goal for quantum information processing is to achieve universal fault-tolerant photonic quantum computing, a challenge of immense complexity. As progress continues, focusing on intermediate milestones is considered fruitful. Currently, we are in the era of noisy intermediate-scale quantum (NISQ) technologies,20 where the goal is to surpass classical supercomputers in areas such as quantum simulations,21 optimization problems,22 and particular tasks that challenge classical algorithms,23 utilizing experimental capabilities that are currently within reach. However, expanding quantum computing power to tackle more complex NISQ problems requires integrating thousands of optical components on-chip, which introduces potential losses and scaling complexity. In addition, the advancement of large-scale quantum information technologies relies on the periodic, near deterministic, or at least highly predictable generation of single photons as these technologies need many single-photon inputs in different modes simultaneously. A promising approach involves multiplexing to convert heralded photons into on-demand photons, by detecting the idler photon to herald the presence of a signal photon and relocating the signal photons to a predetermined target output mode through fast active feedforwarding.24^–26 Achieving this requires a low-loss medium, a high-quality photon-pair source, and fast, ultralow-loss switches.

In light of these considerations, an ideal integrated quantum platform would therefore feature low-loss propagation and coupling, high nonlinearity, ultra-fast electro-optic modulation, and compatibility with key components such as superconducting nanowire single-photon detectors (SNSPDs), quantum memories, and solid-state emitters, to achieve scalable quantum systems deployed in practical real-world environments. The platform must support precise control over the temporal and spectral profiles of photons to interface effectively with different platforms. It should also enable co-integration with cryogenic electronics, facilitating precise control and synchronization of quantum operations at low temperatures (<4K). Moreover, the platform should support versatile quantum light sources capable of generating Fock states, squeezed light, and entangled photons. Together, these elements contribute to the development of a highly functional, scalable, and versatile quantum photonics platform, suitable for advanced applications in computing, communication, and sensing.

In terms of material choice, silicon (Si) and silicon nitride (SiN) have traditionally dominated integrated quantum photonics due to their well-established fabrication technology. However, emerging lithium niobate (LN)-based platforms (see Fig. 1), with loss metrics competitive with the best of other integrated photonics platforms, offer significant advantages across a broad spectral range. LN’s strong second- and third-order nonlinearity enabling efficient nonlinear processes and its low-loss, ultra-fast modulation characteristics even at cryogenic temperatures through the Pockels effect are the capabilities that Si- and SiN-based platforms currently lack. In addition, LN is an excellent crystalline host for rare-earth ions,30 making it one of the leading candidates for on-chip quantum memories.31 With the lack of single-photon detectors and sources, quantum memory can be potentially mitigated with heterogeneous integration (see Sec. 3).

In this paper, we provide an overview of the state-of-the-art TFLN quantum photonics and analyze its technological maturity and potential future challenges. Unlike previous overview papers on TFLN, focusing on photon-pair multiplexing,32 and more generic reviews of integrated quantum photonics,33^,34 this paper will describe TFLN fabrication, heterogeneous integration with sources and detectors, nonlinear effects in TFLN, and quantum applications of TFLN devices.

The review is structured as follows. In Sec. 2, we consider the general properties of lithium niobate, types of waveguiding structures, and fabrication aspects of LN-integrated photonics and analyze the losses and linear elements performance achieved. Section 3 presents achievements in the heterogeneous integration of missing elements such as single-photon sources, detectors, and quantum memories. Section 4 contains an analysis of nonlinearity-based and system-level quantum applications followed by an outlook and description of challenges to overcome (Sec. 5).

Lithium niobate (LiNbO3, LN) is a dielectric and birefringent material. It has uniaxial crystal orientation, which is defined by the X and Y axes, known as the ordinary axes, and the Z axis, referred to as the extraordinary axis. LN exhibits a crystal structure, characterized by a 3m point group: mirror symmetry at 60-deg intervals and threefold rotational symmetry along the c-axis, aligned with the Z-axis.35 Its noncentrosymmetric and ferroelectric properties are responsible for its remarkable intrinsic characteristics,36 making it an attractive material for various applications, e.g., nonlinear frequency conversion and electro-optic light modulation (Fig. 1). With a high Curie temperature close to its melting point (1210°C and 1257°C, respectively), a broad transmission range (from 350 nm to 5μm), and good nonlinear coefficients (r33=31pm/V, and d33=−25.2pm/V at 1064 nm),37 LN is particularly compelling for photonic applications.38

Since the 1960s, bulk LN materials have been studied for electro-optic modulation and nonlinear optical processes. “Bulk LN” usually refers to a large, unstructured single-crystal material where surface effects are negligible. Single-crystal offers superior optical and electro-optic properties compared with polycrystalline forms rarely used in high-performance applications. Traditional bulk-LN waveguides and photonic devices were fabricated using proton exchange (PE) processes39^–43 or metal in-diffusion, mostly Ti,44^,45 to create low-index-contrast waveguides. The PE method replaces lithium cations (Li+) with protons (H+) by immersing LN in acidic solutions at elevated temperatures (150°C to 400°C) for several hours.39^–43^,46 Variations include annealed proton exchange (APE), reversed proton exchange (RPE), and soft proton exchange (SPE). SPE reduces proton concentration near the surface by adding lithium salts such as lithium benzoate to the acidic solution, achieving lower H–Li substitution ratios (<10%).47^–49 It preserves nonlinear coefficients and domain orientation without crystallographic phase transitions. Ti in-diffusion involves depositing Ti strips on the LN surface followed by thermal annealing, allowing Ti ion diffusion into the crystal and increasing the refractive index to form waveguides.

The refractive index change reaches a maximum of 0.1 for PE processes (and only for the extraordinary axis) and typically ranges between 0.001 and 0.04 for Ti in-diffusion. More details about PE processes and Ti in-diffusion can be found in previously published reviews.50^,51 This low index contrast between bulk LN and the modified LN area results in weak mode confinement, reduced nonlinear efficiency, and large bending radii (on the order of millimeters), leading to significant footprints.52 Due to these limitations, combined with the challenges of etching LN and the strong competition from other photonic platforms, the LN platform was largely overlooked for several decades.

However, the emergence of thin-film lithium niobate (TFLN) wafers, in the 2000s, has positioned TFLN as one of the most promising candidates for integrated photonic platforms. The relatively high index of LN (no≈2.2 at 1550 nm) allows the formation of high-index contrast waveguides on top of low-index insulator materials, such as glass or sapphire. Significant progress in ion slicing and wafer bonding enabled the fabrication and commercialization of TFLN wafers.53^,54 The “Smart Cut” process, originally developed for silicon-on-insulator (SOI) wafers,55 was successfully adapted for LN.56 This process involves four main steps:57^,58He+ ion implantation, bonding to an insulator layer and substrate, controlled thermal splitting, and polishing. Companies now offer TFLN wafers with customizable LN thicknesses and various substrate materials (Si, sapphire, quartz, fused silica, …), along with optional conductive electrode layers (Au, Pt, Cr, …) integrated above or below the isolation layer. The wafers could also be doped with alkaline earth and rare-earth elements to further enhance their properties.59^–61

The TFLN platform is now widely used for integrating high-performance devices across a broad range of applications, including high-speed electro-optic modulation,62^–68 electro-optic frequency conversion,69 acousto-optic modulators,70^,71 acoustic wave filters,72 optical73^–77 and Kerr frequency comb generation,78^–82 nonlinear wavelength converters,83^–85 entangled photon-pair sources,86 transduction,87^–91 and photonic quantum technologies.32^,33 The terms “LNOI” and “TFLN” are commonly used interchangeably to describe an integrated platform comprising a submicron-thick LN film bonded to a SiO2 layer on a Si substrate. In this work, we chose the term “TFLN” to refer to this specific platform, unless otherwise stated for cases involving thicker films and/or different substrate materials. Table 1 summarizes the most important performance metrics for various implementations of LN, including bulk and TFLN.

Despite its attractive intrinsic properties, lithium niobate has historically faced integration challenges due to difficulties in the etching process. Today, dry etching is the most common method for achieving low-loss LN devices. It is widely used in the semiconductor industry due to its ability to produce anisotropic profiles and provide precise control over etching depth. However, dry etching of lithium niobate has been a longstanding challenge primarily due to redeposition phenomena resulting in rough and nonvertical sidewalls. A detailed discussion of mask selection and various gases used in the dry etching process is provided below.

Device fabrication on TFLN requires an intermediate mask layer for pattern transfer through lithography. For large structures, typically with critical dimensions greater than 1μm, UV resists such as S1828,98 SU-8,99 and SPR-955 CM-0.9100 are used, whereas sub-micrometer features require deep-UV or e-beam lithography with positive (CSAR 62,101 ZEP-520A,102^–105 and PMMA72^,106) or negative (HSQ62^,64 and Ma-N107^,108) resists. Hard masks enhance etching selectivity and depth but require additional patterning steps. Various materials have been investigated including amorphous-silicon,109^,110 silicon oxide,111 nickel,98^,112^,113 diamond-like carbon (DLC),114 silver,112 and chromium96^,115^–118 with its alloys.52^,119 Notable results include DLC-masked etching achieving 4dB/m losses114 and chromium-masked waveguides with 2dB/m propagation losses.120

Fluorine gases (SF6,42^,43CHF3,121 or CF4122) can etch lithium niobate but form a nonvolatile LiF material. LiF redeposition creates rough and sloped sidewalls and reduces etching rates. Proton exchange (PE) minimizes LiF formation by replacing Li+ with H+ but degrades electro-optical properties.52^,123 Enhanced sputtering using Ar gas with SF6,43^,52^,124^,125CH3F,126^,127C3F8 or He gas with C4F8128^,129 or CF442 has been tested. CH3F/Ar mixtures on PE-LN achieved nearly vertical walls.123 Chlorine-based etching using Cl2,130^,131BCl3, or Ar/Cl2 mixtures produced smooth sidewalls with 80 deg angles.110

Nevertheless, the use of pure physical etching with Ar gas has emerged as the most adopted technique for integrating devices into TFLN platforms.62^–64^,72^,96^,101^,103^,104^,107^–109^,114^,115^,132^–134 The influence of the etching parameters has been studied to optimize dry etching.98^,135 Kaufman et al.136 carried out a detailed study to achieve a redeposition-free regime and strongly emphasized the importance of chamber maintenance and conditioning to achieve good repeatability and consistency in the results. Post-etching steps are also used to optimize the device integration. Soaking in basic mixtures or solutions such as (NH4OH:H2O2:H2O) at 70°C98 or KOH136^,137 effectively removes the redeposited nanoparticles induced by the Ar physical etching. Post-etch annealing in O2 or air atmospheres for several hours further reduces optical losses by restoring TFLN optical properties to almost bulk LN levels. It also improves the intrinsic Q-factors by up to fourfold.28^,95^,120^,138^–141 Zhang et al.142 achieved low propagation losses as 2.7dB/m at 1550 nm and microring resonators with quality factors as 107. Zhu et al.95 achieved an intrinsic Q-factor of 29 million for TFLN microresonators corresponding to a record-low propagation loss of <1.3dB/m.

Other fabrication methods have been investigated to integrate devices into the TFLN platform. For instance, wet etching offers reproducible and scalable fabrication processes. The methods using HF and HNO341^,128 with Cr masks and NH4OH, H2O2, and H2O mixture143^,144 with SiO2 masks have been studied. However, this method tends to produce isotropic profiles, leading to under-etching beneath the mask. This creates design limitations as closely spaced trenches may merge if placed too close to each other.

Ion beam-enhanced etching (IBEE) is another used method.145^,146 In this process, after mask etching, the exposed LN surface is irradiated with Ar+ ions of varying energies, followed by wet etching to remove the irradiated LN material while preserving the masked regions. The focused ion beam (FIB) technique has also been applied to integrated photonic structures in LN crystals or into the TFLN platform.57^,147^–152 A high-energy ion beam is focused and aimed at the sample, removing atoms from the surface through sputtering.

Mechanical techniques such as optical grade dicing have also been investigated but are limited to low-scalability fabrication of regular structures with feature sizes larger than 1μm.153^–156 Chemical-mechanical polishing (CMP) has been used for the fabrication of LN waveguides and microring resonators.97^,141^,157^–163 The process involves chromium mask patterning via laser ablation, CMP polishing, wet chromium removal, and optional secondary CMP for surface smoothing. With this method, Gao et al.97 achieved microring resonators with intrinsic Q-factors of 1.08×108, corresponding to 0.34dB/m propagation losses. Post-etching CMP can also be used to remove LiF redeposition and polish sidewalls after fluorine dry etching.164^,165 Wolf et al.165 combined Ar+CHF3 etching with CMP to fabricate microring resonators achieving 4dB/m losses at 980 nm for the TE mode.

Fiber-to-chip couplers are essential for interfacing between optical fibers and TFLN photonic circuits. Efficient coupling to TFLN chips relies on two main approaches: edge couplers and grating couplers (GCs).166^,167

The principle of edge couplers involves tapering the waveguide dimensions for spot size conversion (SSC) to expand its mode field diameter (MFD) to match the fiber mode. Edge couplers interface optical fibers with LN waveguides at the chip edge, typically providing a large coupling bandwidth. Coupling losses of 1.32 dB for TE modes and 1.88 dB for TM modes with broad bandwidth by tapering the coupled fiber have been reported.168 However, edge couplers typically require precise alignment and polished facets, making them unsuitable for wafer-scale testing. Detailed design principles of polarization-insensitive edge couplers have been discussed in Ref. 169. Designing efficient edge couplers typically involves a tapering strategy of the TFLN waveguide, either horizontally or vertically. Tapering TFLN waveguide horizontally is typically applied, which could be well defined by a lithography process. Different SSC designs have been studied to achieve efficient coupling with fibers. A TFLN bilayer inverse taper-based SSC reduced coupling losses to 1.7 dB per facet at 1550 nm.170 A tri-layer edge coupler design on the TFLN platform with a SiN assisting waveguide achieved 0.75dB/facet at 1550 nm and 0.64dB/facet at ∼1610nm, though alignment and fabrication deviations remain critical, with ∼0.15dB loss from SiN-to-LN mode conversion.171 Combining silica-based planar lightwave circuits with an LN bilayer taper yielded coupling losses of 0.11 dB (TE) and 0.128 dB (TM) with excellent scalability.172 A bilayer taper structure with narrow tips (<300nm) achieved 0.54 dB (TE) and 0.59 dB (TM) at 1.55μm using an ultra-high aperture fiber.173 To release the alignment sensitivity in the fabrication process, a crossed bilayer taper was proposed and demonstrated, exhibiting low coupling loss of 0.29 and 0.24 dB for TE and TM modes to a high-numerical-aperture fiber (HNAF), respectively, with excellent fabrication tolerance.174 Tapering vertically is typically requiring greyscale lithography.175 Jia et al.176 proposed and demonstrated a novel vertical tapering process controlled by CMP with a chromium mask. A coupling loss of 0.92 dB to HNAF for the TE0 mode was reported, assisted with a SiON waveguide.176 The same group further simplified the design without SiON and with SiO2 cladding, which leads to a coupling loss of 1.43 dB for the TE0 mode with the lensed fiber.177 These innovations demonstrate significant progress in improving coupling efficiency and scalability for TFLN edge couplers.

Grating couplers (GCs), in contrast to edge couplers, enable vertical coupling via diffractive structures, offering the possibility of wafer-scale testing. However, TFLN GCs face challenges from LN’s anisotropy and etching limitations, causing nonvertical sidewalls that restrict pitch size and fill factor. GCs typically provide coupling bandwidths of tens of nanometers, which would restrict their applications, e.g., in wide-band optical communications. TFLN-based GCs are categorized into one-dimensional (1D) and two-dimensional (2D) designs. 1D GCs with optimized periodic structures achieve efficient coupling to fiber. Yet, limitations such as scattering losses, substrate leakage loss, and mode mismatch result in coupling losses between 3 and 10 dB.178^–183 Mitigation strategies include buried metal reflectors to redirect downward-scattered light back into the grating, enhancing efficiency. In addition, techniques such as apodized gratings, angled-etched gratings, and multilayer structures offer improved coupling efficiency.178^,180^,181^,184^–190 For example, an apodized 1D GC on Z-cut TFLN with a gold reflector achieved losses as low as 1.42 dB.184 Furthermore, hybrid structures with high-index materials, such as amorphous silicon, improve grating sharpness, achieving efficiencies up to 50% per coupler with a 1-dB bandwidth of 55 nm.180 In comparison, 2D GCs enable functionalities such as polarization multiplexing but exhibit higher losses, such as −5.13dB for P-polarized light and −7.6dB for S-polarized light, lagging behind optimized 1D designs.191

1D GCs are polarization-sensitive in general. Efforts to reduce polarization sensitivity include metal-based 1D GCs on X-cut TFLN. Using plasmonic modes, these designs align diffractive angles for TE and TM polarized light, reducing polarization-dependent losses to less than 0.69 dB across a 44 nm wavelength range. This approach achieved losses of 3.56 dB for TE modes and 4.08 dB for TM modes. Further improvements, such as cavity-assisted gratings with top metal mirrors, have reduced losses to 0.89 dB,188 though these methods add fabrication complexity.

Optical switching and modulation can be realized easily by the thermo-optic effect. A micro-heater is typically placed either on top or192 aside [see Figs. 2(a) and 2(b)]193^,194 the LN waveguide with sufficient distance to reduce the metallic absorption loss. Applying a voltage or current on the microheater generates heat, which changes the refractive index of the LN waveguide. Due to the slow heat dissipation, microheater-based phase shifters have typical bandwidths of tens of kHz.193 The thermo-optic coefficient of ne and no of lithium niobate is around 3.6×10−5K−1 and 4.2×10−5K−1,195 respectively. A typical power consumption, Pπ (half-wave power consumption), of 44 mW196 and 760 mW,194 is achieved for microheaters designed on top and aside LN waveguides, respectively. To reduce the power consumption, the under-cutting technique can be applied leading to a power consumption improvement by a factor of 20 times.192^,193 However, it also greatly reduces the switching speed to sub-kHz, due to much slower thermal dissipation. To increase the speed of the microheater, the driving signal can be pre-engineered to result in a fast phase rising time of only a sub-microsecond [see Fig. 2(c)].196

If more than GHz bandwidth is required, the Pockels effect of LN can be utilized. A more general description of the Pockels effect and basics of electro-optic modulation can be found in classical books, e.g., written by Boyd.197 The main advantage of the Pockels effect is pure phase modulation with ultra-fast response time (picoseconds and less). For a simple introduction, we would like to consider an X-cut LN phase shifter [see Fig. 2(d)]. To understand the principle, one considers a device where refractive index changes are induced by an external electric field Ez applied along the crystalline Z-axis. The change of the refractive index can be described as Δn=−12ne3r33Ez, where ne is the extraordinary refractive index value and r33=30.9pm/V is the zzz component of the electro-optic tensor of LN. Induced phase shift in this case can be expressed as Δϕ=−12ne3r33Ezk0L=−πVVπ. Here, we expressed our field Ez=Vd in terms of applied voltage V and electrodes gap d and also introduced half-wavelength voltage Vπ. This effect allows pure phase E/O modulation desirable for quantum photonics. The pure-phase modulator can be incorporated in a Mach–Zehnder interferometer (MZI) leading to high-speed and high-fidelity switches with ultralow loss, which is crucial for demultiplexing of, e.g., quantum dot-based single-photon sources,198 multiplexing of nonlinear photon-pair sources,199^,200 and universal quantum computing.201 Furthermore, cascading TFLN switches not only allows for on-chip reconfiguration of the optical propagation mode,202 which would be very useful in on-chip mode-encoded entanglement applications,203 but also allows for quick manipulation of the polarization state of light.204 Advanced in-phase/quadrature (IQ) modulators can be constructed by two Mach–Zehnder modulators (MZMs) [see Fig. 2(e)],205 as is widely used in continuous variable (CV) quantum key distributions (QKD). There are a few key metrics that need to be taken into consideration for evaluating the performance of the LN switches and modulators, including the optical insertion loss, the VπL product, which describes the modulation efficiency of the device, the extinction ratio, which is important for high fidelity preparation and manipulation of quantum states, and the E/O bandwidth, which limits the highest operation rate of the device. To design an effective high-bandwidth modulator, overlap integral and phase-matching conditions between RF and optical modes should be taken into account. In addition, a relative position between electrodes and optical field is important for propagation loss management. A deeper analysis of high-speed TFLN modulators and EO devices and specific of its design can be found in two fundamental reviews from Harvard University.67^,206

The traditional E/O LN modulators are based on Ti-diffused or PE waveguides, as discussed in Sec. 2. The weak light confinement requires a large distance between the radio frequency (RF) electrodes, leading to low modulation efficiency (typically >10V·cm). A sub-1 V Vπ (half-wave voltage) thus results in a long device of at least 10 cm. Moreover, due to RF signal loss in the long RF electrodes, the bandwidth is typically limited to less than 40 GHz. The TFLN provides high light confinement that eliminates the bottlenecks of the traditional LN modulators and switches in terms of footprint, modulation efficiency, and bandwidth, with a typical modulation efficiency around 2 to 3V·cm. Significant progress has been achieved in high-performance E/O modulation, such as TFLN modulators with CMOS-level driving voltages being reported, relying on the regular coplanar RF waveguide scheme.62 The regular coplanar RF electrode design still suffers from phase mismatch, which is greatly improved by capacity-loaded traveling-wave (TW) electrodes and employing quartz substrate. A 2 cm long EO modulator with a Vπ of only 1.3 V is reported.207 The high-performance LN phase modulators have naturally led to advanced in-phase/quadrature (IQ) modulators. High bandwidths of 67 GHz and driving voltages of 3.1 V205 have been reported. Recently, dual-polarization IQ modulators have also been reported.208 Though those demonstrations are mainly focused on high-capacity optical communications, they can be widely used in CV-QKD systems209 to achieve Gaussian modulation. The modulation efficiency of TFLN modulators still leads to a device length of around 1 cm. To further reduce the footprint, folded MZMs incorporating meander traveling wave electrodes and optical waveguides have been reported,210^,211 with the price of more complicated electrical driving circuits and complex packaging.

TFLN does not provide a native solution for generating and detecting light, but this may be mitigated by the use of hybrid integration techniques.

The first demonstration of quantum dot integration on the TFLN platform was demonstrated212 in 2018 by Aghaeimeibodi et al. The authors prepared a sample of InAs dots in an InP matrix with a mode converter to couple into the TFLN waveguide. Using the “pick-and-place” approach (which today may be referred to as “microtransfer printing”213), an InP nanobeam with a Bragg reflector and a 5μm adiabatic taper was mounted on a TFLN waveguide. The coupling efficiency from the taper to the waveguide was 40.1%, which can be improved by using a longer taper (see the Supplementary Material of Ref. 212) up to 60% (−2.2dB). The total collection efficiency, considering the QD emission efficiency (85%) and grating coupler efficiency (26.7%), was found to be 9%. The experimentally measured collection efficiency of 2.2% differs from the theoretically expected value, which can be explained by fabrication imperfections and residual misalignment between the nanobeam and the waveguide.

This approach was developed further by Wei and Kesse in their work214 in 2021 [see Fig. 3(a)]. They transferred a similar structure with the Bragg reflector, adiabatic taper, and InAs/GaAs QD emitter on the TFLN circuit with tunable directional couplers for path multiplexing of emitted single photons. Although 50μm long taper demonstrated the near-unity transmission (>0.9, corresponding to <−0.46dB loss), the overall source efficiency reached 42%, primarily due to low coupling between QD and GaAs waveguide. These results demonstrate the capability of almost lossless transition of the single photon to the TFLN chip. The main limitation still present is a design for compact structures to couple light from the emitter to the hosting platform waveguide. From our perspective, this issue can be addressed, and the heterogeneous integration approach appears promising—offering better scalability and environmental stability compared with multiple fiber interfaces between the emitter and processing chip while also avoiding the nonuniform behavior of connecting fibers.

In the last couple of years, a group from KTH demonstrated QD emission tuning using the piezoelectric effect in lithium niobate for InAsP QD in an InP nanowire [see Fig. 3(b)],29^,215 thus introducing an additional degree of freedom for single-photon sources.

Alternatively, instead of using QD emitters, White et al.216 demonstrated the integration of a WSe2 monolayer-based single emitter on a Ti-diffused LN coupler.

To integrate the detectors on TFLN, the SNSPD is the best option in terms of detection efficiency and jitter. The description, working principle, and progress on a different material platform can be found in a freshly published SNSPD review.217 The integration of a NbN nanowire on a TFLN waveguide with 46% on-chip detection efficiency and 32 ps timing jitter was shown by Sayem et al.218 [see Fig. 4(a)]. Besides NbN, NbTiN can also be used as material for detectors, and the integration of such nanowire on a TFLN waveguide was reported in Refs. 219 and 222 [see Fig. 4(b)]. Lomonte et al.219 demonstrated NbTiN SNSPD with 27% on-chip detection efficiency and 17 ps jitter. Recently, Prencipe et al.222 demonstrated a wavelength-sensitive SNSPD integrated into TFLN, which potentially opens the path to wavelength detectors. In 2024, Colangelo et al.220 demonstrated 50% on-chip detection efficiency with 82 ps jitter at 1550 nm utilizing the integration of a MoSi nanowire on TFLN [see Fig. 4(c)]. In the same year, He et al.221 demonstrated a heterogeneous integration of an amorphous silicon detector (a-Si) on TFLN [see Fig. 4(d)], utilizing a preceding on-chip upconversion, which, despite low external efficiency, demonstrated potential single-photon detection even at room temperature.

For storage applications, the so-called photon buffers at the single-photon level on a TFLN chip functioning at room temperature were recently demonstrated223 [see Fig. 5(a)]. They exhibit low-loss characteristics, with short-term storage capabilities of up to 1.4 and 1.8 ns and time resolutions of 100 and 200 ps, respectively. Such short, controllable storage media have been exploited to temporally multiplex single-photon sources.24 Moreover, these buffers are expected to enable a variety of novel on-chip photonic applications, including the generation of multiphoton entangled states226 and the multiplexing of probabilistic gates.227 Operating in the picosecond regime is particularly advantageous as it allows multiplexing within the recovery time of commercial SNSPDs, typically on the order of tens of nanoseconds, required for heralding. This enables many probabilistic events, such as a photon-pair generation, to be attempted within that time window. On the other hand, for long-term storage, LN is an outstanding crystalline host for rare-earth ions, positioning it as one of the best candidates for on-chip quantum memories, with coherence times on the order of microsecond for achieving sufficient storage times and high-efficiency storage for multiple temporal modes.30 A recent demonstration reported a compact integrated storage device based on TFLN waveguide doped with thulium rare-earth ions, with a broad storage bandwidth exceeding 100 MHz and optical storage time as long as 250 ns224 [see Fig. 5(b)]. In addition, erbium (Er)-doped TFLN waveguides utilizing a nanophotonic cavity have recently been suggested to pave the way for a quantum memory at telecom-compatible wavelengths225 [see Fig. 5(c)].

The presence of the second- and third-order nonlinear susceptibility combined with high field confinement makes TFLN a very useful material for the integration of various efficient active and nonlinear devices.228 The second-order nonlinearity in lithium niobate appears in terms of d33=−25.2pm/V, d31=−4.6pm/V, and d22=2.1pm/V; all values are given at 1064 nm.37^,229 It is important to keep in mind that these values also have a wavelength dispersion and can vary up to 25%, e.g., the value of d33=−20.2pm/V at 1550 nm measured,230 which differs from the standard value of −25 to 27pm/V37^,229^,231 widely referred to in the literature. Although the high second-order nonlinear coefficient dominates, the third-order nonlinearity of lithium niobate at 1550 nm (n2=1.8×10−19m2/W)206 is also useful, being on a level comparable with that of silicon nitride (n2=2.5×10−19m2/W).

The aforementioned properties give access to a wide set of the nonlinear processes (see Fig. 6), which can be used for frequency conversion [second harmonics generation (SHG), sum-frequency generation (SFG), difference frequency generation (DFG), and Bragg-scattering four-wave mixing (BS-FWM)], pair generation [four-wave mixing (FWM), degenerate FWM (DFWM), and parametric down conversion (PDC)], and squeezed state generation (DFG).

The nonlinear optical processes require phase matching to be effective. It can be seen from a simple consideration of coupled-mode equations in slowly varying envelope approximation for the SFG process ω3=ω1+ω2, ∂zA1(z)=−iκ1A3(z)A2*(z)exp(−iΔkz),∂zA2(z)=−iκ2A3(z)A1*(z)exp(−iΔkz),∂zA3(z)=−iκ3A1(z)A2(z)exp(iΔkz),where, Ai,i∈{1,2,3} is the envelope function, κi is the wave coupling coefficients, Δk=k3−k2−k1 is the wavenumber mismatch, and exp(iΔkz) term is responsible for phase-matching. The oscillatory nature of this term leads to periodic changes in the energy transfer direction in the case of Δk≠0, significantly reducing the efficiency of the generation. More insights and deeper consideration of second-order nonlinear processes can be found in classical books197 or in fundamental tutorial papers.206^,232

Thus, some measures should be applied to compensate for this mismatch. In integrated photonics, this can be achieved via dispersion and/or nonlinearity engineering. Dispersion engineering helps to design the waveguide cross-section to compensate for the chromatic dispersion of the material with waveguide and modal dispersion.

The spontaneous polarization of lithium niobate can be flipped by applying an external electric field due to its ferroelectric nature. This leads to changing the sign of the second-order nonlinear coefficient, making it possible to obtain quasi-phase matching (QPM).233 In QPM, the phase mismatch Δk=k3−k2−k1−2πmΛ=0 is compensated by a crystal momentum component of the periodic structure with period Λ of polarization poled regions.

These nonlinear effects may find applications in spectral conversion and manipulation of the shape of a wave function at the single-photon level or even for generating quantum light. A broader discussion of nonlinear processes, including other material platforms, can be found in the review of Dutt et al.234

To link different technologies, optimized for different wavelength regions, such as quantum memories based on Rb atoms operating at 780 nm,235 single-photon sources operating at 900 to 980 nm,236 and optical fibers with the lowest losses around 1550 nm,237^,238 it may be highly useful to be able to effectively shift the frequency of the photons. The whole idea of quantum frequency conversion (QFC) is that an incoming single photon can be converted to another frequency range without a change in its statistics. This is challenging as the nonlinear media with the presence of the strong (classical) pump could easily influence the quantum state, if special care is not taken. An early concept for QFC was proposed239 and experimentally observed240 by Kumar and Huang relying on the SFG in periodically poled potassium titanyl phosphate (PPKTP) crystals. Ou237 later did a theoretical study of potential processes for quantum state conversion, including three- and four-wave mixing and Raman and Brillouin scattering. QFC has been seen to preserve entanglement,241 photon number purity, and indistinguishability.242 QFC has been experimentally demonstrated using PPLN waveguides under cryogenic temperatures.243 The most important QFC experiments demonstrated in LN are summarized in Table 2 with a comparison in terms of converter efficiency (including propagation losses) listed chronologically. Some results include coupling losses due to lack of information to subtract it, and these results are listed with a star symbol.

One of the first applications of QFC was for single-photon detection using Si detectors, operating optimally in the near visible range, utilizing a preceding upconversion. This concept has been demonstrated in PPLN crystals,244^,245 RPE waveguides,246 and TFLN258 with high (70% to 90%) internal conversion efficiency and around 40% detection efficiency for 1550 and 1320 nm246 signals. Pelc et al.248 demonstrated 86% internal conversion efficiency and 37% total detection efficiency at 1.55μm, which is considered to be 94% and 45%, respectively. In addition, it was shown that Raman scattering and spontaneous PDC (SPDC) were the two main noise sources.260

For a long time, demonstrations of QFC used upconversion via SFG. However, frequency converters for downconversion based on DFG became interesting alongside the development of nonclassical sources, such as nitrogen vacancy (NV) centers in diamond,238 quantum dots,257 and Rb memory storage,261 to translate generated single photons to the longer wavelength where the optical fiber has minimum losses. One of the challenges for DFG realization is the requirement of a high-power pump in the IR and mid-IR ranges, to provide low noise conversion.

From Table 2, it can be seen that only in one recent paper,258 a TFLN-based converter has been used. The main challenges for end-to-end device performance are the fiber-to-chip loss and propagation loss in the waveguide. This draws special attention to the need for optimized design of fiber-to-chip couplers and fabrication optimization.

In quantum information processing and computing, the indistinguishability of generated and processed photons is crucial for an overall system performance. A key scalability challenge arises from fabrication-induced nonuniformity among quantum emitters. Thus, the ability to change not only central frequency but also the temporal and spectral shape of a single-photon wave package becomes important. Nonlinear conversion processes are one possible approach to achieving this control, contributing to more precise quantum interfaces and improved scalability. Using a PPLN waveguide for frequency conversion, Takesue262 demonstrated the erasure of frequency distinguishability while preserving the Fourier-transform limited characteristics of the photons, confirmed by the observation of a Hong-Ou-Mandel (HOM)263 interference dip detection. The frequency conversion of the photons from two different quantum dots242 was demonstrated using a single-frequency converter, enabling two-photon interference and highlighting the potential to mitigate fabrication-induced variations in quantum dot emission. Donohue et al.264 theoretically showed that engineering of the pump pulse leads to highly effective SFG conversion with the possibility for time-lensing and bandwidth compression. Sielberhorn’s group presented the concept of a quantum pulse gate (QPG),265 using a spectrally engineered pump for SFG to select a particular temporal mode in an upconverted photon using PPLN. The QPG concept was further developed into temporal-mode tomography266 and temporal-mode demultiplexing.267 In addition, the third-order nonlinear effects are suitable for the task of single-photon shaping. Matsuda268 experimentally demonstrated a 0.4 THz single-photon spectral shift using cross-phase modulation (XPM) in dispersion-engineered photonic crystal fiber. All of these techniques can be implemented on the TFLN platform and, combined with electro-optic phase shaping techniques,269^–271 form a versatile toolbox for spectral and temporal shaping.

Lithium niobate may also be used as a source of photon pairs and single heralded photons via SPDC (Fig. 6). In the standard perturbative approach to SPDC within a single-mode waveguide and under low pump power, the generated quantum state can be approximated as |ψ⟩=|0,0⟩s,i+K∫dωsdωif(ωs,ωi)a^s†(ωs)a^i†(ωi)|0,0⟩s,i,where heralding removes the vacuum component |0,0⟩s,i. Here, all relevant factors, including the nonlinear coefficient (e.g., the second-order susceptibility), have been absorbed into the constant K, and a^†(ω) is the photon creation operator at frequency ω. The joint spectral amplitude (JSA), f(ωs,ωi), has a clear physical interpretation: the modulus squared of the normalized JSA, |f(ωs,ωi)|2, defines the joint spectral probability density of the signal and idler photons. The JSA consists of two main components: the pump envelope function and the phase-matching function, the latter arising from momentum conservation. By engineering JSA, especially through the phase-matching conditions, it is possible to reduce spectral correlations, thereby increasing the purity of the state and enhancing interference visibility. For a comprehensive analysis, we refer the reader to a recent review on nonlinear engineering of desired photon characteristics.272

In contrast to the QFC application, photon-pair generation is widely associated with the use of TFLN devices together with Ti-diffused (TD) waveguides and even microresonators.86^,273Table 3 contains the comparison of different realized photon-pair sources in terms of pair generation rate (PGR), coincidence-to-accidental ratio (CAR), second-order autocorrelation at zero delay of a heralded single-photon gh(2)(0), and two-photon interference experiment type and its visibility. The phase matching is limited not merely by QPM but also by phase matching among different spatial waveguide modes at different mixing frequencies (modal phase matching, MPM). Introducing dual-layer TFLN (DLTFLN)283 and layer-poled TFLN (LPTFLN),284 it became possible to achieve nonzero TE₀₀ to TE₀₁ nonlinear mode overlap, improving the efficiency of MPM devices.

Table 3 shows that TFLN-based devices are already able to achieve from units86^,243^,279 up to several hundreds280^,285 of GHz pair generation rates. In all papers where gh(2)(0) was measured, its level goes below 0.2, which is at a comparable level demonstrated for silicon photonics.290^,291 The purity and indistinguishability for all demonstrated devices are on a decent level (>90%); however, only one paper278 demonstrated performance akin to the best silicon sources.290^,291

One may also use the nonlinear processes to generate the squeezed states of light, where one of the quadratures achieves noise distribution lower than the shot noise level.292 This property can help to improve the precision of gravitational wave detection,9 frequency metrology,293 resolution of imaging,294 or other applications,293 where the performance is limited by shot noise. The generation of a squeezed state requires phase-sensitive parametric amplification, which can be achieved using nonlinear optical effects.

The generation of squeezed states in lithium niobate waveguides was shown using proton exchanged,295^–297 Ti-diffused,298 thick films,299^,300 and microdisk301 devices. Mondain et al. demonstrated the generation of the squeezed light in a single parametric down-conversion periodically poled waveguide made by SPE297 with a co-integrated interferometer. The generated light has −2.08±0.05dB squeezing and 2.80±0.05dB anti-squeezing levels measured with respect to shot noise level. A total of 5 dB on-chip squeezing298 was generated in a resonant TD PPLN structure with active electrical tuning. The highest squeezing level achieved on a chip, to the best of our knowledge, is −9.7±0.8dB,299 shown in the thick film (5μm) PPLN waveguide and, on the same platform, −6dB squeezing over a 2.5 THz bandwidth.300

TFLN has also shown a capability for squeezed states with −0.56±0.09dB (inferred on-chip −2.6dB) quadrature squeezing over 7 THz.302 This was achieved in a Z-cut 600 nm periodically poled device. Using a silicon nitride strip-loaded waveguide on an X-cut TFLN,303 picosecond pulse squeezing was achieved with a squeezing of −0.33±0.07dB (−1.7±0.4dB on-chip inferred). The most broadband squeezer achieved in TFLN reaches 25 THz304 with −4.9dB squeezing on-chip. In a recent paper, Arge et al.305 demonstrated an MPM TFLN resonator for squeezed state generation with −0.46dB shot noise reduction.

Developing a pure single-photon source is a “holy grail” of quantum photonics. A promising approach to achieve periodic, near-deterministic generation of pure single photons involves active multiplexing of photon-pair sources. The envisioned fully on-chip, hybrid spatial (M sources) and temporal (N modes) multiplexed photon source requires several key components (see Fig. 7):

This source is based on a logarithmic tree structure, where each photon traverses the (lossy) switch ⌈log2(M)⌉ times, followed by temporal multiplexing.24 TFLN offers two significant advantages. (i) Pump photons can be filtered out relatively easily due to their spectral separation from the generated photon pairs. (ii) TFLN modulators are capable of operating at cryogenic temperatures, albeit with a slight increase in modulation voltage compared with room temperature operation.306 On the other hand, cryogenic environments enable the use of superconducting electrodes, which further enhance the performance of LN-based modulators.307

On-chip photon-pair sources can be quasi-phase-matched type-II PPLN waveguides, offering high spectral purity.281 These sources enable the use of high-efficiency polarization beam splitters to separate heralding and heralded photons directly on-chip.204 However, achieving sufficient pump rejection, typically on the order of 100 dB, poses a significant challenge. To address this, Bragg filters adapted to TFLN308 can be used, and the large spectral separation between the pump and photon pairs improves filtering performance. Once filtered, heralding photons can be detected using waveguide-integrated SNSPDs (see Sec. 3.2), at cryogenic temperatures, activating a dedicated heralding logic. A critical step is the implementation of low-latency logic, requiring heralded photons to be delayed for switch activation within the same clock cycle. Most implementations of multiplexing schemes suffer from electronic latencies of around 100 ns.200 Notably, a very recent study demonstrated a cryogenic operation of a dedicated logic, achieving latency as low as 23 ns,309 corresponding to a 4.18 dB loss for TFLN-based delay lines (∼0.182dB/ns for state-of-the-art propagation losses of 1.3dB/m95). Latency-induced loss sets an upper bound on the achievable source efficiency at roughly 0.38, notably below the 2/3 efficiency threshold suggested for loss-tolerant linear optical quantum computing scheme based on the stabilizer tree graph codes.310 Therefore, minimizing latency is critical as both loss and decoherence scale with the length of the delay lines. Achieving latencies on the order of just a few nanoseconds would be necessary to limit losses in TFLN-based delay lines to acceptable levels. Encouragingly, tailored cryogenic control circuits are currently under active development (see Ref. 311 for a comprehensive review). Such advancements enable a new generation of on-chip, scalable, high-quality single-photon sources that surpass the efficiency threshold necessary for practical implementations of linear optical quantum computing. In addition, to perform single-photon multiplexing to the desired output mode, CMOS-compatible modulators are required, and they have already been reported.62 Finally, regarding footprint, low-loss components have been fabricated at the wafer scale on 6-in. TFLN wafers.111 Therefore, it is, in principle, possible to achieve efficient on-chip multiplexing integrated with all necessary components, which can also be volume-fabricated.

The proposed TFLN-based hybrid multiplexed source, using the current state-of-the-art parameters, does not yet reach the efficiency of the leading QD source, which has achieved efficiencies as high as 0.712 at 885 nm.312 However, QD sources face scalability challenges due to poor visibility when interfering with single photons generated by independent quantum dots. Although this issue can be addressed using photon shaping techniques, as discussed in Sec. 4.1.2, scaling QDs remains difficult in practice. As most quantum protocols require multiple indistinguishable single-photon inputs across different modes simultaneously, generating such a stream typically mainly relies on a single QD combined with a demultiplexing strategy,313 which reduces operation rate and introduces additional losses. On the other hand, high-rate nonlinear single-photon sources feature excellent indistinguishability between sources by coherent pumping,314 leading to scalable quantum systems. Furthermore, the emission wavelength of single photons from pair sources can be flexibly engineered through dispersion engineering, enabling generation directly in the telecommunication band, crucial for integration with existing fiber infrastructure.

The generation of remote entanglement between spatially separated quantum nodes is foundational to quantum networks, enabling applications ranging from distributed quantum sensing315 to distributed quantum computing.316 Quantum memories are crucial components in these networks as they can become entangled across distant nodes.255 However, current entanglement rates between quantum memory nodes are below 1 Hz, which limits scalability.317 One approach to increase these rates is to multiplex memory nodes, which requires scalable, large-scale switch networks with low insertion losses to support multiple memory channels while maintaining high efficiency. In addition to efficient multiplexing, interfacing these memory nodes with an optical fiber might demand the ability to shift photon frequencies with high efficiency. Precise frequency manipulation can be achieved via: (i) QFC, discussed in Sec. 4.1.2, or (ii) electro-optic spectral shearing (EOSS).269^–271 EOSS shifts the photon frequency by applying a linearly ramped phase modulation, Δϕ=−Kt, which translates a photon’s initial frequency ω to a new frequency ω+K, allowing frequency shifts on the order of ∼100GHz. Recent advances in TFLN platforms provide a promising solution for multiplexing quantum nodes. These platforms integrate low-insertion-loss couplers (<1dB per facet), frequency shifters (capable of ∼33GHz shifts), and a 2×2 switching network (with extinction ratios exceeding 20 dB) on a single chip, operating from visible to near-infrared wavelengths318 [see Fig. 8(a)]. These advancements are expected to enhance entanglement rates and pave the way for burgeoning quantum networks.

On the other hand, an all-photonic quantum repeater eliminates the need for matter-based quantum memories but requires large entangled resource states (i.e., graph states),15 which, in principle, can be generated deterministically via quantum emitters320 or by multiplexing probabilistic linear optical entangling gates. To achieve determinism, a feedforward mechanism is necessary, which in turn requires ultra-low loss and low-latency control.

High-fidelity state generation and manipulation may also be used for quantum sensing applications. For quantities such as phase or temperature, which lack direct self-adjoint operators, values can be estimated by measuring related observables, known as parameter estimation. This process is essential in applications such as gravitational wave detection,9 where phase shifts produced by gravitational waves can be estimated through interference. Squeezed states of light are increasingly promising in that sense as they can be bright and enhance sensitivity by reducing the noise level in their squeezed quadrature (see Sec. 4.1.4).17^,293 Recent advancements have demonstrated the integration of squeezed light sources into photonics circuits on the same TFLN chip, achieving nearly complete phase sensor functionality with quantum-enhanced sensitivity319 [see Fig. 8(b)].

It is strongly believed that raw quantum computational power, which surpasses the performance of classical supercomputers on specific tasks, can be demonstrated with modest experimental requirements via rudimentary quantum computers. In computational complexity terms, certain abstract problems that are intractable, if not impossible, for classical computers can be solved by these quantum devices. Photonics offers a candidate for possible demonstration of the so-called quantum computational advantage: boson sampling (BS).321 It is arguably one of the simplest NISQ models in terms of technological requirements as it involves preparing a quantum state with indistinguishable photons, letting it evolve in a multimode optical network through quantum interference, and then detecting the output; see Fig. 9(a) for a conceptual scheme. The distribution of the detection pattern approximates the probability distribution governing the experiment, which is difficult for classical computers to sample. This is because calculating the probability corresponding to an output event requires estimating the permanence of a complex-valued matrix, a task considered notoriously difficult to compute for large matrices. Note that surpassing classical computers depends significantly on the number of occupied input modes (n) and the total number of interferometer modes (m);323 however, increasing either of these is challenging in practice.324 Over the last few years, some plausible demonstrations of quantum advantage with Gaussian states of light325 have been reported.1^,314^,326 However, debates continue regarding these claimed demonstrations as photon losses and partial distinguishability may lead to classically simulable quantum devices.327^,328 In addition to claimed demonstrations of quantum advantage, BS can also be mapped to the simulation of molecular dynamics329^,330 and graph theoretical calculations23 and can be used to enhance classical algorithms.331 A more comprehensive review of BS can be found in Ref. 332.

TFLN photonics is highly effective in generation of quantum states of light and provides precise photon path manipulation due to its high electro-optic coefficients and low-loss properties. These capabilities are expected to pave the way for new demonstrations of quantum advantage. A two-photon quantum interference that occurs at a 2×2 beam splitter represents the most basic scenario and can be further generalized to n-photon quantum interference in an m-mode unitary, called multimode quantum interference. It is crucial for quantum technologies and can be realized using a series of interconnected MZIs. An MZI equipped with a phase shifter in one of its arms acts as a tunable beam splitter, allowing for precise control over photon pathways. This functionality is essential for constructing quantum circuits with complete reconfiguration capabilities. Once the unitary evolution U for a specific quantum protocol is defined, it can be translated into a physical circuit by decomposing U into layers of tunable MZIs and phase shifters. In photonics, two widely used universal decomposition methods are the Reck (triangle) scheme333 and the Clements (rectangular) scheme.334 Recent studies have demonstrated a TFLN-based integrated optical circuit that serves as a fast and reconfigurable four-mode universal linear quantum processor, consisting of a series of MZIs and phase shifters that are programmable through the electro-optic effect. One study followed the Clements scheme and interfaced it with deterministic solid-state single-photon sources based on quantum dots in nanophotonic waveguides at 930 nm,198 as shown in Fig. 9(b), whereas another study implemented the Reck scheme,322 as shown in Fig. 9(c), highlighting the flexibility and scalability of on-chip quantum state manipulation using TFLN technology.

Rudimentary quantum computers help build a foundation of techniques and technologies that will eventually support progress toward the “holy grail” of quantum information science: a universal fault-tolerant quantum computer. Universality refers to a computer capable of performing arbitrary conceivable processes allowed by quantum mechanics, whereas fault tolerance supplements error correction codes with protocols for reliable computation. One of the primary approaches to realizing such a computer relies on discrete variables, i.e., qubits. Qubits are two-level quantum systems, i.e., |ψ⟩=α|0⟩+β|1⟩ where α, β∈C, called probability amplitudes satisfying normalization condition |α|2+|β|2=1. The basis |0⟩ and |1⟩ represent the mutually distinguishable states of the system. Photonic architectures for encoding qubits can be based on various degrees of freedom of the single photon, such as polarization, transverse spatial modes, frequency modes, path, and time-bin degrees of freedom. Among these, path encoding, where a photon occupies different optical paths (waveguides), is central to integrated photonics because it allows for extremely low-error qubit operations using Mach–Zehnder interferometers and phase shifters, as noted above.

TFLN photonics is an emerging candidate for universal optical quantum computation. The standard gate-based model,335 which relies on realizing deterministic gates through ancillary circuitry and photons, faces polynomial yet prohibitive overheads at a practical level. A new paradigm, so-called measurement-based quantum computation (MBQC), pioneered by Raussendorf and Briegel336 changed the way we look at quantum computation by offering a significantly more resource-efficient approach, shifting the focus from the gate-based model to the generation of highly entangled resource states, e.g., universal 2D cluster states.337^,338 In MBQC, computation is driven by performing single-qubit measurements, which are straightforward to implement on photonic qubits, on these entangled states. Depending on the complex correlation shared between qubits, local measurements on one qubit influence its neighbors. The inherent randomness of these outcomes is compensated through adaptive measurements on neighboring qubits via active feedforwarding, enabling the effective execution of a quantum algorithm through precise coordination. A linear optical implementation of MBQC, primarily developed by Browne and Rudolph, significantly reduced the resources required to generate large photonic cluster states339 by introducing the fusion mechanism, two-qubit measurements between qubits from separate entangled states. Subsequent theoretical advancements, incorporating ideas from percolation theory, demonstrated that universal (incomplete) lattice resource states could be produced from 3-qubit GHZ states340 using fusion operations.341^–343 More recently, the fusion-based quantum computing scheme201 has emerged, utilizing many copies of small entangled resource states and fusion operations. However, generating resource states, including 3-qubit GHZ states, with linear optics remains challenging, as all interferometers for two-photon gates are probabilistic. To address this limitation, entangling gates can be multiplexed to deterministically generate resource states.344 High-speed, low-loss optical switches enabled by TFLN technology are expected to herald a new era in the deterministic generation of resource states using linear optics. On the other hand, based primarily on Lindner and Rudolph’s proposal,345 quantum emitters have been shown to be promising candidates for the deterministic generation of resource states, including examples such as QDs346 and single atoms.347 However, the most concerning issue—apart from the increased complexity—is that these methods are vulnerable to stochastic noise as opposed to the linear optical-based methods.

In optical CV MBQC architectures,348 squeezed light is the main ingredient. Ideally, cluster states can be generated by entangling neighboring infinitely squeezed states, which are, however, nonphysical. In addition, homodyne detections, depending on the local oscillator’s phase, drive the computation. In practice, approximate CV cluster states—prepared using finite squeezing rather than ideal infinite squeezing—can be generated deterministically, and homodyne measurements can be effectively implemented. The use of approximate cluster states introduces computational noise resulting from finite squeezing, which can lead to errors in computation. Fortunately, as in other quantum computing platforms, such errors can be mitigated through quantum error correction.

To address scalability challenges, the frequency multiplexing349 and the temporal multiplexing350 schemes have been suggested. Frequency multiplexing schemes encode quantum information in the quadratures of optical frequency combs,351 whereas temporal multiplexing schemes encode information in the quadratures of fields at different time bins.352 In Ref. 352, a 2D cluster state with a cylindrical topology was successfully generated using an experimental setup comprising a pair of bow-tie-shaped squeezers, beam splitters, and optical fiber-based delay lines. Although this scheme enables scaling computation with a fixed number of spatial resources, it also has inherent limitations that constrain the achievable computation depth. Specifically, in temporal encoding schemes, the maximum allowable delay length is limited by propagation losses, which introduce additional computational noise. Beyond a certain delay length, depending on the quantum error correction scheme employed, fault-tolerant computation becomes infeasible.

In Ref. 351, the quantum-wire cluster state in the quantum optical frequency comb is created by pumping optical parametric oscillators with synchronous phase-locked polychromatic beams. A natural next step is to extend this concept to chip-scale implementations using low-loss microresonator platforms such as SiN. As a foundational requirement, squeezed state generation on the SiN platform has been demonstrated.353^,354 Following this, recently, SiN has been used to generate rich-structured CV cluster states across quantum frequency combs using multiple pump lines.355^,356 However, the squeezing levels currently demonstrated on SiN are still below the 12 to 15 dB range considered necessary for fault-tolerant CV quantum computing357^–359 due to the platform’s reliance on inherently weak third-order nonlinearities. Moreover, frequency multiplexed cluster states require polychromatic homodyne detection,360 where each frequency mode is individually addressed. This increases the experimental complexity and overhead, making the approach more challenging in practice in comparison to time-domain approaches.

On the other hand, TFLN photonics offers a compelling alternative to bring approximate CV cluster states closer to their ideal counterparts by combining low propagation losses comparable to SiN with strong second-order nonlinearities (see Sec. 4.1.4). Its support for increased bandwidth may facilitate higher repetition rate setups, which, in turn, would allow more temporal modes to fit within the maximum allowable delay length, ensuring effective quantum error correction. Furthermore, a TFLN-based temporally multiplexed compact CV cluster state generator could further enhance system performance by minimizing coupling losses. Loop-based architectures with integrated switches, as proposed in Ref. 361, could further benefit from TFLN photonics, making it a key technology for the realization of CV MBQC. Furthermore, as CV MBQC is driven by phase-dependent homodyne detection, fully on-chip implementation requires programmable local oscillators where TFLN stands out.362 Taken together, these capabilities position TFLN as a strong candidate for scalable, fault-tolerant CV MBQC.

To further clarify the motivation for using TFLN over SiN, we include Table 4, which summarizes the key differences between the two platforms in the context of large-scale CV MBQC. Although Table 5 provides a material-level comparison, Table 4 highlights the application-specific differences relevant to scalable CV MBQC. Though each material platform alone has its own advantages and disadvantages, lithium niobate is particularly highlighted featuring both high second- and third-order nonlinearity, ultra-low propagation loss, high EO coefficient with pure phase modulation, and high level of squeezing, which are keys for quantum photonic applications. It is also obvious that to achieve a TFLN chip-level quantum photonic system, the heterogeneous integration of QD sources and detectors is a necessary step, as discussed in Sec. 3. Furthermore, the heterogeneous integration of TFLN with silicon nitride platform would lead to a further technological breakthrough leveraging both extremely low loss of silicon nitride and all distinct features of lithium niobate.

Over the past few years, TFLN quantum photonics has emerged as a versatile integrated platform for developing future quantum technologies. Its low-loss, ultra-fast electro-optic switching capabilities, and high nonlinear coefficients are key characteristics that make it ideal for quantum photonic applications. These applications include enhanced quantum metrology, near-deterministic single-photon sources through active feedforwarding, high-secret-key-rate QKD, high-rate boson sampler, and efficient resource generation for universal computation and the quantum internet. The ultimate goal would be to realize fault-tolerant fully on-chip universal quantum processors to minimize coupling losses between different platforms, thereby significantly enhancing efficiency and scalability. However, the integration of quantum light sources, large-scale circuits, and photon detectors on a single chip remains a challenge. This overview lists the current challenges in TFLN quantum photonics and explores potential pathways toward achieving fully on-chip quantum information processing.

There are ongoing studies addressing all of these challenges, and, as discussed in the sections above, significant progress has been made in recent years. Despite these improvements and identified pathways, the realization of fully on-chip, near- and long-term quantum processors demands continued efforts in device fabrication, engineering, and monolithic integration. Nevertheless, based on our analysis and expectations, TFLN is poised to shape the future of integrated quantum photonics with its unique combination of performance, versatility, and scalability.

Acknowledgment. The authors acknowledge the Villum Fonden Young Investigator project (QUANPIC, Ref. 00025298), the NQCP (Novo Nordic Foundation Quantum Computing Programme), and the Danish National Research Foundation (DRNF) Research Centre of Excellence, SPOC (Silicon Photonics for Optical Communications) (Ref. DNRF123).

Fabien Labbé received his PhD from PERSEE, Mines ParisTech (PSL Research University), Valbonne, France, in 2018. He is currently a nano-fabrication engineer at the Technical University of Denmark (DTU), Kongens Lyngby, Denmark. His research interests include the integration of photonic devices into several platforms such as silicon, silicon oxide, silicon nitride, and lithium niobate. His areas of expertise include UV and e-beam lithography, thin-film deposition, dry and wet etching, and morphological characterizations.

Çağın Ekici received his PhD in electronics and communication engineering from the Izmir Institute of Technology, Turkey, in 2021. In 2022, he joined the Silicon Photonics for Optical Communications group at the Technical University of Denmark as a postdoctoral researcher, where he works on deterministic thin-film lithium niobate-based single-photon technology and on-chip analog quantum simulation with photons.

Innokentiy Zhdanov is a postdoctoral researcher at DTU Electro. He received his BSc and MSc degrees in physics from Novosibirsk State University in 2015 and 2017, respectively, and his PhD in optics from the Institute of Automation and Electrometry, SB RAS, Novosibirsk in 2022. He is currently working as a postdoc at the Technical University of Denmark and his current research interests include integrated nonlinear and quantum photonics, especially on lithium niobate platforms.

Alif Laila Muthali received her PhD in 2024 from the Technical University of Denmark (DTU Electro), where she focused on the design, fabrication, and experimental studies of silicon and lithium niobate photonic devices for quantum applications. She is currently a research process scientist at DTU Physics, working on the fabrication of photonic chips on various platforms, including silicon nitride and diamond, for quantum technologies.

Leif Katsuo Oxenløwe received his BSc and MSc degrees in physics and astronomy from the Niels Bohr Institute, University of Copenhagen, in 1996 and 1998, respectively. He received his PhD in 2002 from DTU, and since 2009, he has been a professor of Photonic Communication Technologies. He is the group leader of the High-Speed Optical Communications group at the Department of Electrical and Photonics Engineering (Electro), at the Technical University of Denmark (DTU), and he is the leader of the Centre of Excellence SPOC (Silicon Photonics for Optical Communications). His research interests are focused on silicon photonics for optical processing and high-speed optical communication.

Yunhong Ding received his BS and PhD degrees in electronic science and technology from Huazhong University of Science and Technology, Wuhan, China, in 2006 and 2011, respectively. He is currently a senior researcher/associate professor at the Department of Electrical and Photonics Engineering, Technical University of Denmark (DTU Electro). In 2011, he joined the Department of Photonics Engineering (DTU Fotonik) as a Postdoc. He has been the PI or Co-PI of H.C. Ørsteds Grant, DFF FTP Grant, Sapere Aude Forskertalent, and EU QuantERA Project. Since 2017, he has been a senior researcher/associate professor and was granted the Villum Young Investigator in 2019. His research interests include integrated quantum information processing, quantum simulation, quantum communication, and optical communications. He is also the founder of the foundry SiPhotonIC ApS, Denmark.