Global internet traffic has surged exponentially over the past decades, fueled by the rapid growth in functionality and data flow associated with smartphones, cloud services, and Internet-of-Things devices. The relentless demands of greater bandwidth have placed immense pressure on signal processing systems, especially in data centers, where achieving high-speed, large-bandwidth, and low-power consumption is critical^[1,2]. Currently, data processing relies predominantly on electronic integrated circuits (EICs), one of the cornerstone technological breakthroughs in the 20th century. EICs revolutionized computing by integrating vast numbers of transistors and electronic devices onto a single chip, significantly reducing device sizes and increasing data processing capabilities^[3].

However, EICs, which operate using electrons, face inherent limitations in speed and bandwidth, rendering them unable to meet the ever-increasing demands. In addition, electrical resistance in EICs generates a significant amount of heat, compromising efficiency and consuming a large amount of cooling energy. More importantly, despite remarkable achievements in sub-7 nm node technology over the past decade, complementary metal-oxide-semiconductor (CMOS) technology for the fabrication of EICs is approaching its physical limits as the device size continues to shrink^[4]. This limitation arises from two primary challenges: 1) the severe increases in leakage currents at atomic scales, and 2) the inherent constraints of electronics in interconnects, including the skin effects and impedances of electronic interconnects.

These bottlenecks underscore an urgent need for the development of innovative technologies to overcome the speed, bandwidth, and scalability limitations of conventional EICs, paving the way for the next-generation data processing systems.

Under these circumstances, photonic integrated circuits (PICs) emerge as an optical analog to EICs. Operating with photons instead of electrons, PICs offer unparalleled advantages of high speed (at the speed of light), minimal loss and electromagnetic interference, large bandwidth, low latency, low power consumption and heat generation, and potential security improvements, which are desired in communications, data processing, and computing^[5,6]. With the ability to steer photons in multiple dimensions on the same platform, PICs promise to host more complexity on a single chip, leading to a significant reduction in size, energy, and cost. The transformative potential positions PICs as a cornerstone technology in fields ranging from communications^[5,7] and biomedical devices^[8] to autonomous navigation^[9] and chemical and atmospheric sensing^[10]. As such, PICs are poised to play an irreplaceable role in the next generation of information and communication technology.

Depending on material selection, PICs consist of numerous photonic components integrated on a single chip, either monolithically or heterogeneously^[5]. Unlike EICs, which rely on transistors as their fundamental building blocks, PICs utilize diverse photonic building blocks, such as waveguides^[11,12], multimode-interference-based couplers^[13], Mach–Zehnder interferometers^[14], lasers^[15,16], and optical ring resonators^[17]. Silicon is now the most preferred material platform due to its compatibility with current CMOS fabrication, which is suitable for fabricating passive components like waveguides, multimode interferences, and Mach–Zehnder interferometers. Meanwhile, indium phosphide has been broadly applied in active devices, including laser generation, amplification, control, and detection^[18,19]. The devices based on these platforms are fabricated mainly by conventional nanofabrication techniques, such as photolithography using UV light and CMOS technology. In addition, active devices such as high-speed modulators and photodetectors can be fabricated through CMOS-compatible doping processes and Ge-on-Si epitaxy^[20]. Those techniques are suitable for manufacturing two-dimensional (2D) PICs, where all the photonic components are arranged in a planar configuration, limiting the exploitation of the full three-dimensional (3D) volume for integration.

In comparison, 3D PICs, as shown in Fig. 1, have garnered significant attention for their ability to enhance integration density by efficiently using the full 3D space^[4,21]. Conventional 3D PICs^[22] can be fabricated by repeating the combined processes of the waveguide core layer deposition, lithography, etching, waveguide cladding layer deposition, and planarization, such as chemical and mechanical polishing, to form multilayer structures in 3D, in which low-loss inverse taper waveguides can achieve the interlayer coupling^[23]. Similarly, another type of 3D PIC can be achieved by layer-by-layer stacking of multilayer 2D photonic components to produce a 3D photonic crystal^[24].

The advent of laser nanofabrication, in particular, ultrafast laser nanoprinting (also called direct laser writing), represents a transformative advancement for 3D PIC fabrication (Fig. 1). By employing femtosecond (fs) lasers, this technique enables the direct creation of fully 3D PICs composing 3D waveguides in a single process, significantly simplifying the entire fabrication process and saving cost^[25]. In addition, the inherent flexibility of laser nanofabrication allows for the construction of arbitrary 2D/3D structures, enabling the integration of 2D/3D architectures on a single material platform for hybrid PICs that are inconceivable with other technologies.

Ultrafast laser nanofabrication has emerged over the last decade as a powerful technique for directly fabricating functional photonic devices in transparent dielectric substrates or polymer materials. Utilizing fs laser pulses at wavelengths within the transparent window of the material, this approach enables true 3D fabrication deep inside materials^[26]. The high peak power from the ultrafast laser pulses can trigger nonlinear absorption, including two-photon and multiphoton absorption, in the focal region of a tightly focused ultrafast laser for nanofabrication^[27]. This unique capability allows the focused laser to act as an optical “pen,” capable of “drawing” arbitrary structures with nanoscale precision, surpassing the constraints of conventional nanofabrication techniques like photolithography and CMOS-based processes.

More importantly, different from conventional nanofabrication techniques, which can only change the physical shape of structures via etching (although sometimes doping is required in separate processes using different setups to modify the properties of materials), laser nanofabrication can directly modify material properties through various light–matter interaction mechanisms, such as refractive index modulation, ablation, or polymerization depending on the material property and laser settings, making laser nanofabrication a versatile tool^[26]. Moreover, those mechanisms can be applied in the same material via the control of laser wavelength, pulse width, power, and repetition rate. The unique light–matter interaction mechanisms can be regarded as the fourth dimension of laser nanofabrication, which allows the creation of structures with various properties at different positions. This capability facilitates the realization of advanced functionalities far beyond the reach of conventional nanofabrication technologies, positioning ultrafast laser nanofabrication as a game-changing tool in the development of the next-generation photonic devices.

The great flexibility in material processing is another significant advantage of laser nanofabrication, particularly in the fabrication of devices using newly developed materials. Recently, 2D materials have attracted significant research and application interests in PIC-related fields due to their outstanding optical properties and atomically thin nature (Fig. 1, inset). There are various demonstrations of functional devices for PIC applications that integrate 2D materials with conventional semiconductor PIC devices to achieve functional devices that surpass the performance limitations of traditional approaches^[28–31]. Laser nanofabrication has been demonstrated to be an effective and efficient tool for manufacturing 2D material devices, from material modification and film patterning to functional structure fabrication. Therefore, it is possible to first integrate 2D materials on substrates, then fabricate functional devices within substrates and the 2D materials in a single process via laser nanofabrication to create integrated functional devices. Such an approach paves the way for innovative PIC designs that combine 2D/3D materials and functional devices, opening new horizons in PIC development and manufacturing.

To provide a comprehensive review of laser nanofabrication and its applications in PICs, this review is arranged as follows. In Sec. 2, the 3D capability is first reviewed as the most significant capability of laser nanofabrication. Besides 3D structures, the recent advancement of integrating 2D material into 2D PICs is presented in Sec. 3 to provide various integration strategies, and the state-of-the-art performance enhancement is also reviewed. Functional photonic devices based on 2D materials fabricated via laser nanofabrication are reviewed to demonstrate approaches and requirements of manufacturing 2D materials using lasers. Advanced beam-shaping techniques for improving the quality and the speed of laser nanofabrication are reviewed in Sec. 4. Section 5 reviews current 2D/3D integration strategies, including integrating 3D polymer or silica devices with various 2D materials or silicon devices to form multifunctional 3D architectures. More importantly, we summarize the techniques for integrating 2D materials, emphasizing their potential applications in 2D/3D device fabrication, which could be an important future research direction of PICs. Based on the advancement of 2D/3D PICs, the related applications, including optical communications, quantum computing, biomedicine-optofluidics, sensing, automotive vehicles, and metrology, are reviewed in Sec. 6. Instead of giving detailed information about all the applications, we briefly and broadly describe those applications to cover as many applications as possible to show the transformative potential of PICs. Finally, in Sec. 7, we conclude the review and present the potential development directions of laser nanofabrication in PICs.

Micro/nano photonic structures are essential for constructing miniaturized and multifunctional PICs. Fs laser nanofabrication is a powerful 3D precision-processing technique that is cost effective, time saving, and maskless^[37–41]. In recent years, the fs laser nanofabrication technique has become a mainstream method for producing versatile 2D/3D photonic structures^[42–52]. This section reviews the functional 3D photonic components and devices fabricated using laser nanofabrication, including waveguides, microelements, and periodic structures. The detailed requirements and the design considerations are provided, as well as the unique capability of laser nanofabrication to fulfill those requirements.

The experimental systems for fs laser nanofabrication share a fundamentally similar design, with minor differences to suit specific applications. In most nanofabrication systems, a microscope objective is usually utilized to focus the fs laser on the surface or inside of a sample (Fig. 2). Using a programmable high-resolution motorized 3D stage, the microscope objective and the sample can move relatively according to a preset route. Therefore, various 2D or 3D micro/nanoscale photonic structures can be produced based on fs-laser-induced multiphoton morphology modifications, refractive index changes, ferroelectric domain inversions, etc^[37].

Several key parameters of fs laser sources must be carefully selected according to the requirements of the fabrication process, including wavelength, pulse width, repetition rate, power, and polarization. Currently, most fs laser nanofabrication systems employ visible and near-infrared (NIR) fs lasers with central wavelengths ranging from $\sim 500\mathrm{nm}$ to $\sim 1\mathrm{\mu }\mathrm{m}$. In addition, the wavelength can be converted on demand using an optical parametric oscillator or an optical parametric amplifier to cover the entire visible wavelength region and extend into the NIR region. In general, for 3D fabrication, it is required that the material does not have linear absorption at the fabrication wavelength so that the laser beam can penetrate the material effectively. Meanwhile, the fabrication resolution/line width ($d$) depends on the combination of the laser wavelength ($\lambda$) and the numerical aperture (NA) of the microscope objective according to the diffraction-limited relationship, which is expressed as $d=\lambda /\text{2NA}$. Therefore, for high-resolution nanofabrication, short-wavelength and high-NA objectives are preferred.

A fs laser typically emits linearly polarized light, which can be directly applied in laser nanofabrication, which, however, may introduce asymmetric intensity distribution in the focal region of a high-NA objective due to the depolarization effect^[53]. A circularly symmetric focal spot in the focal plane can be achieved by converting linear polarization to circular polarization, thus further improving the symmetry of the fabricated structures^[54]. Advanced polarization states (e.g., cylindrical polarizations) can be used to further introduce different intensity and polarization distributions in the focal region, for example, super-resolved^[55] or doughnut-shaped^[56] focal spots with longitudinal or azimuthal polarizations in the focal region.

Pulse width (PW), another key parameter of a fs laser, decides the peak power ($P_{\mathrm{peak}}$) of a fs laser pulse, which can be expressed as $P_{\mathrm{peak}}=E/\mathrm{PW}$, where $E$ is the energy of a single pulse. The advantage of a fs laser is the extremely high peak power, which can be used to trigger nonlinear effects required for 3D nanofabrication. On the other hand, the repetition rate ($R_r$) mainly governs the heating effect by controlling the interval between pulses. If the interval is longer than the heat dissipation time inside the material (typically at the scale of microseconds), the material will not be heated up, enabling cold manufacturing. Conversely, if the interval is shorter than the heat dissipation time, the material will be heated, introducing a thermal effect. Extensive studies have investigated the effects of low (e.g., 1 kHz) and high (e.g., 100 MHz) repetition rates of fs lasers on the impacts of material processing outcomes. One can flexibly select appropriate repetition rates according to material categories and processing requirements. For example, the repetition rate of 25 kHz can be used to induce refractive index changes in ${\text{LiNbO}}_3$ crystals^[57], whereas a repetition rate of 76 MHz is more suitable for inverting ferroelectric domains of ${\text{LiNbO}}_3$ crystals^[58]. Meanwhile, by knowing the average power ($P$) and $R_r$, the energy of a single pulse can be calculated as $E=P/R_r$, which is also one of the key processing parameters for nanofabrication. In practice, optimizing the pulse energy often provides a quick and effective approach for identifying suitable processing parameters. In the fabrication process, the laser energy dose is also controlled by the exposure time (for dot fabrication) or scanning speed (for line fabrication), which can be expressed as $E=P\times \mathrm{\Delta }t$, where $\mathrm{\Delta }t$ is the dwell time of the laser focal spot. For dot fabrication, $\mathrm{\Delta }t$ is directly controlled by setting the exposure time. In comparison, for line fabrication, $\mathrm{\Delta }t=d/v$, where $d$ is the focal area and $v$ is the traveling speed of the scanning stage.

Optical waveguides are essential elements in integrated photonics, serving as the backbone for numerous photonic devices and systems^[11,12,59]. Fs laser nanofabrication can be applied to fabricate 3D waveguide structures to construct high-performance waveguide-based 3D photonic devices, such as waveguide lasers^[60–62], beam splitters^[63–66], and frequency converters^[67–69]. An optical waveguide is composed of a high-refractive-index region surrounded by low-refractive-index areas. Light waves can be tightly confined in a small modal area by total internal reflection (TIR) in waveguides for non-diffractive propagation. This means that the optical power can be maintained at a long propagation length due to the low propagation loss^[70].

A high-quality waveguide is characterized by low propagation losses and cross-sectional profiles matching the modal area. The propagation losses depend on material properties (e.g., uniformity and linear absorption of materials) and the refractive index contrast of the waveguide that controls light confinement. Thus, to realize lower propagation losses, the material needs to be uniform and have a low absorption at the working wavelength of the waveguide, which minimizes the losses from scattering and absorption. In addition, it requires a large refractive index contrast ($>0.001$) between the waveguide periphery and the bulk material to effectively confine the light inside the waveguide^[71–73]. Therefore, for any given material, enlarging refractive index contrast is an effective approach to decrease the propagation losses of waveguides. In addition, the surface roughness of the waveguide will also affect the propagation loss of the waveguide due to the scattering loss.

Meanwhile, the cross-sectional profile of a waveguide decides the propagation mode of light within the waveguide. Using different nanofabrication techniques, optical waveguides with diverse cross-section shapes can be fabricated, such as rectangular waveguides^[74], ridge waveguides^[75], and circular and elliptical waveguides^[76]. Among them, the circular and elliptical cross-sections [Figs. 3(a) and 3(b)] are two typical shapes of fs laser-fabricated waveguides following the shape of the focal spot of the beam. As shown in Figs. 3(a) and 3(b), a circular waveguide exhibits better light mode confinement compared to a waveguide with an ellipse cross-section. This is also reflected by an increased propagation loss in the ellipse waveguide, as shown in Fig. 3(c). In general, the fundamental mode with a Gaussian profile is preferred to low-loss propagation, as it minimizes the energy dissipation compared to high-order modes. To estimate the mode configuration, a waveguide parameter $V=2\pi \rho {(n_{\text{co}}^2-n_{\text{cl}}^2)}^{1/2}/\lambda$ is defined, where $n_{\text{co}}$ and $n_{\text{cl}}$ are the refractive index values of the waveguide core and cladding layer, respectively, $\lambda$ is the free-space wavelength, and $\rho$ is a typical linear dimension parameter, which corresponds to the core radius of a circular waveguide or the half-width of the planar waveguide^[77]. If $V\gg 1$, the waveguide supports multimode. When $V$ is sufficiently small, a single-mode waveguide is achieved. Therefore, it is convenient to modulate mode profiles of waveguides (e.g., from single-mode to multimode) by changing the morphologies (i.e., areas and shapes) of waveguide cross-sections. This kind of morphology engineering of waveguide cross-sections can tailor single-mode waveguides at specific wavelengths and investigate intriguing physical phenomena and applications based on mode modulation^[77–79].

In summary, the performance parameters of optical waveguides strongly depend on the refractive index contrast (ideally as high as possible) and the geometry (in particular, the cross-section), which can be well controlled by fs laser nanofabrication.

To date, various optical waveguides have been successfully fabricated by fs laser nanofabrication in semiconducting crystals, glasses, and polymers^[80]. These waveguides are primarily created through three mechanisms: polymerization, modification, and modification enhancement. Polymer waveguides are usually produced via polymerization, including two-photon or multiphoton polymerization^[81,82], which create a solid waveguide surrounded by air, resulting in a refractive contrast of $\sim 0.5$. In addition, the surface of the polymer waveguides is quite smooth, therefore featuring a low propagation loss. It should be noted that although polymeric materials are not yet the primary platform for PICs, laser-fabricated polymer waveguides and other 3D optical components play an essential role in 2D/3D device integration within PIC fabrication. One typical application of polymer waveguides is photonic wire bonding (PWB), which can achieve low-loss connections between different 2D/3D photonic chips^[83]. More details about PWBs are described in the subsequent section.

Meanwhile, fs lasers can introduce phase or structural (or both) modifications in transparent materials resulting from laser-induced optical breakdown, in which the optical energy is transferred to the material lattice by ionizing electrons^[85]. This process can create a refractive index change at low intensity or even a void at high intensity. For fs pulses, the timescale of electron excitations is shorter than the electron-phonon scattering time ($\sim 1\mathrm{ps}$). As a result, the fs excitation is shorter than the thermal excitation of any ions, which minimizes the heat diffusion outside the focal region, resulting in high resolution and high precision. The required intensity ($I$) in this process depends nonlinearly on the material’s bandgap, which can be expressed as $I\propto E·{\mathrm{NA}}^2/[\mathrm{PW}·{\lambda }^2·(1-{\mathrm{NA}}^2)$]^[85], where $E$ is the pulse energy, PW is the pulse width, as mentioned before, NA is the numerical aperture of the focusing objective lens, and $\lambda$ is the laser wavelength. One can see that narrower pulses and shorter wavelengths can reduce the required intensity for a given pulse energy and NA. A tightly focused fs laser beam could induce positive ($\mathrm{\Delta }n>0$, i.e., type-I modification) using low-energy fs laser pulses or negative ($\mathrm{\Delta }n<0$, i.e., type-II modification) refractive index changes by high pulse energy in glasses and crystals^[37]. Type-I modification can be applied to directly form a waveguide structure in one step by laser-introduced higher refractive index contrast. Type-I modification [Figs. 4(a) and 4(b)] could be induced in most glass and a few crystals, such as ${\text{LiNbO}}_3$ crystals, ${\text{LiTaO}}_3$ crystals, and Nd: YCOB crystals^[80], which has been used to construct waveguide structures composed of single^[56] or multiple^[51] laser-fabricated lines. Type-I modifications are more easily induced in glass because their disordered, non-equilibrium structure allows for greater flexibility and lower energy barriers for atomic rearrangement^[86]. Additionally, compared to glasses, type-I waveguides in crystals are less stable, especially at high temperatures, thus limiting their practical applications^[87].

In comparison, type-II modification cannot directly support light propagation due to the lower refractive index contrast, whereas their adjacent areas may have a relatively high refractive index, resulting from increased stress in modified regions^[88], as shown in Figs. 4(c) and 4(d). The stress can be caused by localized heating, lattice distortion, volume expansion, and filamentation due to laser irradiation^[89]. This characteristic enables the fabrication of waveguides by strategically placing type-II modifications around the guiding region. Examples include dual-line waveguides^[69], vertical-dual-line waveguides^[57,90], depressed-cladding waveguides^[91], and optical-lattice-like cladding waveguides^[65]. Besides, by combining type-II modification and a laser ablation technique, all-fs-laser-written ridge waveguides, composed of two ablated grooves and a series of bottom type-II modification regions, could also be produced^[92].

In addition to direct laser nanofabrication, wet chemical etching can selectively etch the laser-written area to introduce nanopores/nanochannels inside materials to further enhance the refractive index contrast. For example, Ródenas et al. reported a nanostructured waveguide in yttrium aluminum garnet (YAG) crystals surrounded by several nanopores via this method^[93]. The waveguides fabricated by this method can possess lower propagation losses and better confining ability to light. Similar mechanisms can also be achieved in waveguide fabrication in glass^[52,94]. It should be noted that, in this technique, the laser writing parameters must be precisely controlled and optimized to ensure etching to the desired depth and geometry.

In addition to the simple waveguide demonstration, hybrid waveguide devices have also been achieved using laser nanofabrication. In 2021, Wu et al., for the first time, combined type-I modifications (i.e., single-line waveguides) and type-II modifications (i.e., depressed-cladding waveguides) to fabricate hybrid waveguide structures (i.e., hybrid beam splitters)^[87]. Figure 5(a) shows the schematic of the hybrid beam splitters fabricated by the fs laser in ${\text{LiTaO}}_3$ crystals. Part 1 and part 3 are depressed-cladding waveguides, and part 2 represents single-line waveguides. The hybrid structure leverages the strengths of both modification types: depressed-cladding waveguides reduce propagation losses, while single-line waveguides minimize bending losses. Additionally, the single-line waveguides shown in part 2 could be erased by thermal annealing, and these erased waveguides can be restored through fs laser direct writing, as illustrated in Figs. 5(c) and 5(d). It has been found that hybrid beam splitters possess good performance in the output of programmable optical signals, highly promising for erasable photonic data processors. By fully utilizing the advantages of both type-I and type-II modifications, i.e., the rewritable capability of type-I waveguides and good thermal stability of type-II waveguides, such hybrid waveguide structures pave the way for the fabrication of more advanced reconfigurable optical signal processors. Recently, the same group reported a laser-written waveguide beam splitter in ${\text{LiNbO}}_3$ crystals based on type-II geometry [Fig. 5(g)]^[95]. The device features with a Y-branched configuration with a $40\mathrm{\mu }\mathrm{m}\times 40\mathrm{\mu }\mathrm{m}$ cross-sectioned input arm and two $20\mathrm{\mu }\mathrm{m}\times 20\mathrm{\mu }\mathrm{m}$ output channels. Figure 5(h) shows the optical microscope images of the input faces, output faces, and top views of the fabricated waveguides. These waveguide beam splitters support the light propagation across a broad wavelength range from the visible to mid-infrared, making it highly suitable for multifunctional integrated photonic chips.

Innovative techniques featuring high spatial writing resolution have also been demonstrated. In 2020, Li et al. reported an optical far-field-induced near-field breakdown (O-FIB) method, which could be employed to realize fs laser nanofabrication of solid materials (e.g., dielectric and semiconductor materials)^[96]. Compared with the conventional focused ion beam technique, the most remarkable advantage of O-FIB is that it can achieve far-field nanofabrication with comparable resolution in the atmosphere. The fabrication mechanism of this technique, illustrated in Fig. 5(i), begins with the creation of a nanohole on the sample surface, which acts as a seed. This nanohole is generated by a tightly focused fs laser via multiphoton absorption. The cutting “knife edge” of the nanohole is then sharpened through far-field-regulated near-field enhancement, achieving an extraordinary spatial resolution of sub-20 nm, equivalent to less than $\lambda /40$, where $\lambda$ is the central wavelength of the fs laser. Figure 5(j) shows the relation between scanning speed and line width of nanogrooves, both experimentally and theoretically. Note in Fig. 5(j) that a minimum line width of 18 nm could be obtained with a scanning speed of 8 µm/s.

The O-FIB method offers several compelling advantages. It does not require complicated manipulations, as the nanogrooves can be rapidly written by simply controlling the polarization direction of the incident fs laser. In addition, its capability for large-area nanoprinting surpasses that of conventional techniques. Key benefits such as vacuum-free operation, sub-20 nm spatial resolution, simplicity of polarization manipulation, universality of near-field enhancement, and the stitchless nanofabrication make the O-FIB technique an exceptional tool for industrial production of nanopatterns and nanodevices. Especially this kind of O-FIB technique has great potential in fabricating nano-waveguides on the surfaces of novel thin film materials, e.g., lithium-niobate-on-insulator and 4H-silicon carbide-on-insulator.

The laser-written waveguides can be employed to fabricate various active optical devices, such as on-chip waveguide lasers^[97], optical modulators^[98], wavelength conversion devices^[58], as well as single-photon sources^[99]. More details of these devices and their applications will be discussed in Sec. 6.

Lenses are fundamental elements in optical systems. In particular, microlenses with miniaturized size can be integrated into PIC for coupling and focusing light. While various methods exist for fabricating small lenses^[100], they often face significant limitations, including restricted miniaturization and design flexibility, inability to integrate multiple elements for advanced functions (e.g., aberration correction), and challenges in alignment. In contrast, laser nanofabrication’s high spatial resolution and flexibility enable the precise manufacturing of compound optical elements on demand. Gissibal et al. experimentally demonstrated laser nanofabrication of multi-lens objectives with sizes around 100 µm [Figs. 6(a) and 6(b)]. In addition, the performance of the microlenses is characterized by quantitatively measuring the modulation transfer function and aberrations^[101]. They further fabricated an array of compound microlenses to demonstrate the capability of laser nanofabrication [Fig. 6(c)]. Later on, the same group demonstrated laser nanofabricated eagle eye structures directly attached to a CMOS image sensor to build a miniaturized camera^[102]. Beyond microlenses, laser nanofabrication enables the in-situ creation of various other micro-optical elements, such as micromirrors and couplers, highlighting the unmatched flexibility of laser nanofabrication. Dietrich et al. demonstrated in-situ printing of facet-attached micro-optical elements, which allowed precise alignment and adaptation of vastly different mode profiles, resulting in high coupling efficiencies^[103]. Figure 6(d) illustrates the schematic of these 3D micro-optical elements. This flexibility extends to the fabrication of free-form micro-mirrors and microlens assemblies [Figs. 6(e)–6(j)] for simultaneously controlling the optical beam width, shape, and propagation direction. The same group recently demonstrated a facet-attached microlens on a SiP chip for PIC applications^[104]. This approach could find broad applications in PIC-related research and is poised to find widespread applications in commercial photonic products in the near future.

In addition to individual micro-optical elements, laser nanofabrication has been applied to fabricate microlens arrays (MLAs). Generally, MLAs are a series of miniaturized lenses in 2D arrays, in which the diameters of each lens are from several micrometers to nearly millimeters. MLA can focus incoming light into an array of small light spots with a large field of view, high sensitivity, high signal-to-noise ratio^[105], and even resolution beyond the optical diffraction limit^[106]. Besides, MLAs have the advantages of miniaturization, light weight, and easy integration. Therefore, MLAs have been widely used in various advanced devices, such as biomedical applications^[107], beam shaping and steering^[108], microfluidic sensing^[109], virtual reality systems^[110], optical communications^[111], lab-on-a-chip systems^[112], compound eyes^[113], and digital cameras^[114].

MLAs can be fabricated by various micro/nanofabrication technologies, including hot embossing^[116], ion exchange technique^[117], thermal reflow^[118], photolithography^[119], electron-beam lithography^[120], laser nanofabrication^[121], inkjet technology^[122], self-assemble technology^[123], and other mixed methods^[124]. Among them, the laser nanofabrication method has attracted much attention due to its numerous advantages, such as being maskless, low cost, one-step fabrication process, high speed, and scalability. More importantly, laser nanofabrication offers exceptional flexibility, allowing for the precise fabrication of MLAs on 3D surfaces with varying sizes and NAs of each microlens. Furthermore, the versatility enables the fabrication of MLAs on any materials and substrates^[124–129] beyond the limitation of the conventional photolithography methods. For example, Wu et al. employed a voxel-modulation laser nanofabrication method^[115] based on the two-photon-polymerization (TPP) mechanism to create a 3D artificial compound eye as schematically shown in Figs. 6(k) and 6(l). The resulting hexagonal compound eye features high qualities of a fill factor of 100%, a large NA of 0.4, a surface roughness of 2.5 nm, a high resolution of $\sim 1.7\mathrm{\mu }\mathrm{m}$, and a distortion-free field of view of 90°. Besides, customized filling factors and arbitrary off-axis MLAs can be fabricated with this one-step method. These advantages make laser nanofabrication an outstanding technique for industrial-scale production of MLAs. As simpler, more cost-effective, and highly efficient methods continue to be developed, laser nanofabrication is poised to play a critical role in the manufacturing of micro-optical elements for PIC applications^[130].

Periodic structures are critical elements for PIC applications, serving for two primary purposes: 1) coupling light into PIC structures and 2) confining optical energy within 2D/3D structures to create waveguides or cavities. The inherent flexibility of laser nanofabrication has enabled the creation of arbitrarily designed gratings, demonstrating its capability for advanced photonic designs.

One example showing the flexibility of laser nanofabrication is the fabrication of novel waveguide Bragg gratings (WBGs), which integrate waveguides and gratings to periodically modulate refractive indices within waveguide cores along the light propagation direction^[131–133]. The grating periods of WBGs could be tailored to reflect particular wavelengths around the Bragg wavelength (${\lambda }_0=2n_{\mathrm{eff}}\mathrm{\Lambda }$, where $n_{\mathrm{eff}}$ is the effective refractive index and $\mathrm{\Lambda }$ is the period) while transmitting lights at other wavelengths. The laser nanofabrication is eminently suitable to process WBGs because it allows flexible control of the periods and configurations of the gratings during the nanofabrication process.

In 2006, using laser nanofabrication, Marshall et al. first successfully fabricated both optical waveguides and grating structures in fused silica, realizing the fabrication of WBGs^[134]. In addition, Kroesen et al. demonstrated the laser fabrication and characterization of WBGs in ${\text{LiNbO}}_3$ crystals (Fig. 7)^[135]. The schematic of the experimental setup for WBG fabrication is shown in Fig. 7(a). Figure 7(b) is the mode profile at 1550 nm in a depressed-cladding waveguide along extraordinary polarization. The top view of WBGs is shown in Fig. 7(c), and an end-face microscope image of WBGs is shown in Fig. 7(d). The WBGs present low-loss, symmetric mode distribution, and narrowband reflection in the C-band. The ability to fabricate both grating structures and optical waveguides underscores the significant potential of fs laser nanofabrication in producing versatile WBGs. This capability is poised to play an increasingly pivotal role in advancing PIC technologies by enabling precise and customizable periodic structures.

Another method to fabricate nanometer period gratings is to use the fs-laser-induced periodic surface structures (LIPSSs), which can fast fabricate large-area nanoscale gratings on the surface of various materials, especially the subwavelength grating structures, on the surface of various materials^[136–139]. During fs laser nanofabrication, the interactions between fs lasers and materials (e.g., metals, semiconductors, and crystals) introduce a large number of electrons due to the absorption of fs laser pulses^[138,139], which can introduce periodic structures mainly on the surfaces. LIPSSs can be achieved by either linear or multiphoton absorption, depending on the intrinsic material properties.

LIPSSs are essentially a kind of periodic nanostructure, also known as nanoripples or nanogratings, which were discovered in 1965 by Birnbaum for the first time^[136]. The formation of LIPSSs is influenced by both the types of materials and processing parameters (e.g., laser wavelength, laser polarization, number of pulses, angle of incidence, and scanning velocity). Generally speaking, two kinds of LIPPSs can be fabricated, low-spatial-frequency and high-spatial-frequency LIPPSs. Low-spatial-frequency LIPPSs feature a period ($\mathrm{\Lambda }$) close to the effective wavelength of the incident laser in the material (${\lambda }_{\mathrm{eff}}={\lambda }_0{n}_{\mathrm{eff}}$, where ${\lambda }_0$ is the laser wavelength in free space and $n_{\mathrm{eff}}$ is the effective refractive index). For high-spatial-frequency LIPSSs, the grating period is much smaller than the incident laser wavelength^[140]. It can be explained that the low-spatial-frequency LIPSSs are formed by the interference of an incident laser and plasmon polaritons excited on the surface, i.e., produced by a periodic energy distribution on the surface. However, the formation mechanisms of high-spatial-frequency LIPPSs are still under debate^[140].

Diverse LIPSSs have been demonstrated on various materials^[137–139]. For example, Xu et al. reported LIPSSs produced by fs lasers on the 2205 stainless steel (a kind of alloy) surface^[137]. It can be found that the homogenous low-spatial-frequency LIPSSs (LSFLs) with a spatial period of 890–940 nm are in the whole fs-laser-irradiated region, perpendicular to the polarization of the incident fs laser. The spatial period of 890–940 nm is close to the wavelength of an incident fs laser (e.g., the central wavelength at 1030 nm), which the classical interference model mentioned above could explain. Besides, high-spatial-frequency LIPSSs (HSFLs) with a spatial period of 380–470 nm have also been observed, parallel to the polarization of an incident fs laser. The spatial period of 380–470 nm is much smaller than the wavelength of the incident fs laser, which can be clearly observed only at high magnification. It is worth mentioning that LIPSSs, including LSFLs and HSFLs, have been observed not only on metals but also on many other materials, such as semiconductors and crystals^{[139,141,142]}. These nanostructures on material surfaces are largely identical, with only minor differences. The applications of LIPSSs include structural colors, modulation of wetting properties, engineering biomaterials, and so on^[143]. Combined with other photonic structures (e.g., domain structures and optical waveguides) written by fs lasers, LIPSSs may have great potential for improving PIC performances and fabricating novel 2D/3D PICs.

2D and 3D photonic crystal (PhC) structures with photonic bandgaps to inhibit light propagation^[144] are also fundamental building blocks for PICs; in particular, the 3D PhC structures are promising candidates for building 3D PIC devices with 3D waveguides and cavities^[145]. Laser nanofabrication of 2D^[146] and 3D PhC structures in various materials has been demonstrated^{[145,147,148]}. Given the extensive body of work in this area, we provide selected examples to illustrate key advancements rather than offering an exhaustive review.

Due to the 3D capability and flexibility of laser nanofabrication, it is possible to fabricate 3D PhCs via different mechanisms according to the material properties, for example, multiphoton (mainly two-photon, depending on the laser wavelength and materials’ absorption) polymerization in polymer materials^[149,150] and microexplosion in crystals (including glasses)^[148]. One example to show the unique capability of laser nanofabrication is displayed in Figs. 8(a)–8(d), in which Deubel et al. demonstrated the integration of 3D/2D/3D PhC heterostructures that can be fabricated by laser nanofabrication in a single step^[149]. More importantly, the intrinsic properties of the polymer 3D PhCs can be further tailored by doping different materials into the polymers^[151,152]. Furthermore, the refractive index of polymerized 3D PhC structures can be increased by introducing silicon material via a double inversion approach^[153] to achieve a complete photonic bandgap that is unachievable using low refractive index polymer materials [Figs. 8(e)–8(h)]. This method overcomes the difficulty of direct laser nanofabrication in silicon material, which has a strong absorption in the visible to near-infrared range (the typical wavelength region of the fs laser used for laser nanofabrication). On the other hand, laser nanofabrication can directly fabricate 3D PhCs in transparent high-refractive-index materials. One dominant example is the lithium niobate material (${\text{LiNbO}}_3$)^[148], which is the most well-known material for electro-optical tuning due to the strong Pockels effect. However, the high refractive index inevitably introduces spherical aberration^[34], which becomes more profound with increased NA and depth inside the materials. Therefore, adaptive aberration compensation based on a dynamic spatial light modulator (SLM) is necessary to achieve high-quality fabrication, which will be reviewed in detail in Sec. 3.

Nanofabrication inside nonlinear materials makes it possible to create nonlinear photonic crystals (NPCs). NPCs are a kind of micro/nanostructure with periodically modulated quadratic nonlinearity, which have been widely applied to produce and manipulate coherent lights at new frequencies^[154–157]. Figs. 9(a)–9(c) demonstrate the schematics of 1D, 2D, and 3D NPCs. NPCs could be produced using laser nanofabrication based on modification mechanisms or ferroelectric domain engineering. Shao et al. proposed an additional periodic phase method from order (i.e., bulk materials)/disorder (i.e., fs-laser-induced modifications) alignment, which could be applied to meet phase-matching conditions in arbitrary nonlinear crystals^[158]. This approach facilitates the fabrication of 1D NPCs, where periodic order-disorder structures are created via fs laser modification. They have realized deep-ultraviolet 177.3 nm coherent output in fs-laser-processed periodically disordered quartz crystals with an unprecedented conversion efficiency of 1.07 ‰. Following that, Shao et al. proposed an angular engineering strategy of additional periodic phase, by which a wide tuning of 111 nm in deep-ultraviolet had been achieved in fs-laser-processed periodically disordered quartz crystals^[159].

Domain inversion and domain erasure are two primary strategies of fs-laser-dominated ferroelectric domain engineering^[163]. Most ferroelectric crystals, such as ${\text{LiNbO}}_3$ crystals, are transparent in the NIR band. In ferroelectric nonlinear crystals, a tightly focused NIR fs laser could produce ultrahigh temperatures and steep temperature gradients in the focal areas^[58,164]. The resulted ultrahigh temperature could notably reduce the coercive field of a crystal, and the temperature-gradient-induced thermoelectric field may invert the direction of spontaneous polarization (i.e., ferroelectric domains). Based on fs-laser-induced ferroelectric domain inversions, various frequency-conversion devices and 3D NPCs have been constructed^[58,157]. During this process, domain inversion induced by the fs laser is essentially a phase change of the nonlinear optical coefficient (from 1 to $-1$). For example, using fs-laser-induced periodically inverted domain structures (i.e., 1D NPCs) in ${\text{LiNbO}}_3$ crystals, Chen et al. realized a frequency doubling of 815 nm with a conversion efficiency of 17.45 % in an optical waveguide^[58]. Čerenkov second-harmonic (SH) microscopy has been utilized to demonstrate the 3D profile [see Fig. 9(d)] of inverted ferroelectric domains. Using the fs laser domain inversion technique, Xu et al. fabricated 3D NPCs in ferroelectric barium calcium titanate (BCT) crystals^[165]. More recently, Xu et al. realized nanoscale inverted-domain structure fabrication in ${\text{LiNbO}}_3$ crystals by non-reciprocal 3D fs laser writing, which has great potential in achieving highly efficient frequency conversion and high-performance nanodomain-based electronic devices^[163].

Due to strong light–matter interactions, the crystallinity of the ${\text{LiNbO}}_3$ crystals could be completely erased or partly erased by a tightly focused fs laser, which is different from the domain inversion mentioned above. Wei et al. reported an experimental demonstration of 3D NPCs in ${\text{LiNbO}}_3$ crystals by selective fs laser erasing^[166]. It is possible to create long domains due to the spherical aberration originating from the refractive index mismatch of ${\text{LiNbO}}_3$ and fabrication immersion media, which increase with the depth, as shown in Fig. 9(e)^[161]. Ferroelectric domains with a period of $15\mathrm{\mu m}\times 15\mathrm{\mu m}\times 64\mathrm{\mu m}$ were realized in a barium calcium titanate (BCT, ${\text{BaCaTiO}}_3$) crystal^[161] [Fig. 9(f)], which has a much lower coercive field compared with the ${\text{LiNbO}}_3$ crystal. So, the domains in BCT crystals can be well controlled.

More recently, 3D multilayer domain structures containing fork, linear grating, and circular grating domain patterns designed for nonlinear wavefront shaping have been fabricated in the $z$-cut CaBaNbO (CBN) crystals [Fig. 9(g)]^[162]. The lateral period of these structures is 2 µm, which can be further improved to reach submicron by optimizing the laser nanofabrication conditions. To sum up, both modification mechanisms and ferroelectric domain engineering can be used to fabricate 3D NPCs, and they are essentially a kind of phase modulation. With these two approaches, more innovative and multifunctional 3D NPCs can be produced, which have immense potential applications in frequency conversion and beam shaping.

In summary, diverse 3D structures have been achieved through laser nanofabrication. The fabrication mechanisms vary depending on the material, allowing for the creation of different device structures, as summarized in Table 1. This flexibility in material choice is highly advantageous for PIC manufacturing, where various functional modules based on different material platforms must be integrated into a single chip.

In addition to fabricating 3D structures, integrating 2D materials, such as graphene^[169], graphene oxide (GO)^[170], perovskite^[171,172], transition metal dichalcogenides (TMDCs)^[173,174], and black phosphorus (BP)^[175], into 2D chips provides another attractive strategy. These materials have optical bandgaps in different wavelength ranges from ultraviolet (UV) to microwave regions, making them promising candidates for optical light sources for optical communication and quantum computing systems, as well as on-chip photodetectors from ultraviolet to terahertz regions^[176–180]. Due to the intrinsic low dispersion and large anisotropic light absorption of 2D materials, graphene^[169], GO^[181], TMDCs^[182], BP^[183], and MXenes^[184] are integrated to fabricate broadband polarization-selective devices, which are essential components in PICs. 2D materials with exceptional nonlinear optical properties have also been used to develop high-performance all-optical signal processing devices. These devices leverage nonlinear optical mechanisms, including four-wave mixing^[74,185], self-phase modulation^[186,187], and second-harmonic generation^[188] to achieve superior functionality. Flat optical lenses can also be fabricated by patterning ultrathin graphene^[189], GO^[190], TMDCs^[191,192], and perovskite^[193,194] films. Compared to traditional bulk devices, flat lenses made from 2D materials offer enhanced flexibility in both design and integration. This adaptability opens new possibilities for applications in imaging, sensing, and photonics.

There are numerous benefits of 2D material integration, including their extraordinary electrical and optical properties, strong light–matter interactions on an atomical scale, and many other novel exotic phenomena that can be used to design future spin electronic and quantum devices^[195,196]. Beyond their unique properties, 2D materials also excel in device fabrication. They are compatible with current CMOS technology and are easy to integrate onto chips via simple transfer- or solution-based self-assembly processes instead of the sophisticated deposition techniques that are usually required by bulk crystals. Those devices based on 2D materials have demonstrated huge superiority in both the footprint and device performance. Therefore, they are highly desirable in the fabrication of low-cost and ultra-compact photonic devices utilized in future PICs. On the other hand, the success of these devices relies heavily on advanced fabrication techniques. Among them, laser nanofabrication stands out with its maskless, in-situ, and one-step production capability, having been widely employed.

The strong light–matter interactions in 2D materials make it possible to use laser irradiation to locally modify their physical and chemical properties, providing a foundation for patterning 2D materials for multiple functional devices.

Laser modification of 2D materials induces various processes, including a thermal effect, single/multiphoton absorption, phase transitions, as well as photochemical reactions^[197–199]. These effects lead to different material changes, such as 2D material thinning, optical property tuning, for example, refractive index and PL tuning, as well as bandgap engineering, and conductivity modulation^[200–206]. For example, laser thinning of 2D materials, such as graphene, GO, and TMDCs, can achieve atomic-layer precision by optimizing laser fluence and scanning times^[199–202]. Figures 10(a)–10(c) illustrate a self-limiting thinning technique for on-demand thinning of TMDCs, which involves optically activated electrochemical etching that automatically halts once the TMD flakes reach the desired thickness^[202]. Laser-induced reduction and oxidation have also been employed in modifying the electrical and optical properties of materials. GO reduction [Figs. 10(d)] is a typical example while laser oxidation is used to modify materials such as graphene^[203], BP^[207], and WSe₂^[208]. In addition, laser-induced phase transition is another mechanism of property optimization for 2D ${\mathrm{MoTe}}_2$^[204], ${\mathrm{MoS}}_2$^[209], and ${\mathrm{PdSe}}_2$^[210]. Figure 10(g) shows the 2H to 1T phase transition in ${\mathrm{MoTe}}_2$, resulting in enhanced carrier mobilities^[204]. Laser-assisted doping featuring site-specific selectivity has also been widely utilized. By properly selecting doping precursors, n-type and p-type characteristics can be achieved in ${\mathrm{MoS}}_2$ and ${\mathrm{WSe}}_2$, respectively^[205,206]. For h-BN and TMDC monolayers, laser-based defect engineering, as illustrated in Figs. 10(h)–10(j), is also an efficient technique to generate and modulate the spin defects in these materials, which form the foundation of their applications in quantum computing devices^[211–214].

The system for laser patterning is similar to the one introduced in Sec. 2, in which the incident laser wavelength, pulse duration, as well as fluences, and scanning speed are finely adjusted according to different laser modification mechanisms as well as material characteristics.

As displayed in Fig. 11, a variety of 2D material patterns have been demonstrated, from gratings [Figs. 11(a)–11(c)] and simple dot arrays [Figs. 11(d) and 11(e)] to concentric rings [Fig. 11(f)], as well as more complex shapes, for example, microcircuits [Figs. 11(g) and 11(h)] and a Chinese knot [Fig. 11(i)]. With optimized laser fluences and translation speeds, a line feature size down to 400 nm can be achieved [Figs. 11(a)–11(c)]^[216,221]. Laser-patterned graphene nanodisks and the ${\mathrm{MoS}}_2$ nanohole array are shown in Figs. 11(d) and 11(e), respectively^[219], which are applied in large-scale integrated chips. In addition, arbitrary shape patterning can be realized by direct laser writing, for example, GO [Fig. 11(g)] and MXene [Fig. 11(h)] microcircuits. The in-situ patterning of $\gamma \text{-}{\text{CsPbI}}_3$ quantum dot films has also been achieved recently^[223]. One example of a Chinese knot is shown in Fig. 11(i). Compared with other patterning techniques, such as electron beam lithography^[224] and focused ion beam milling^[225], laser nanofabrication possesses high flexibility in both the available materials and patterning shapes without the requirements of specific mask design and additional etching process, which is highly advantageous for fabricating functional photonic devices in a single step, offering new possibilities for developing one-step and rapid patterning techniques.

With its capability for nanoscale patterning and property modification, laser nanofabrication has been employed to create a variety of functional photonic devices using 2D materials. These include on-chip photonic gratings, polarizers, waveguides, optical lenses, and innovative on-chip photodetectors.

Gratings, one of the most fundamental photonic devices, have been fabricated at both the microscale and nanoscale using 2D materials. For example, GO nano-gratings have been demonstrated using laser nanofabrication [Fig. 12(a)], in which the tightly focused fs laser beam simultaneously ablates the film in the line and reduces the surrounding area^[217]. The line width can be controlled to be around 400 nm [Fig. 12(b)]. Due to the ultra-broadband absorption property of the graphene family materials, the nanogratings can absorb up to 90% of light within the entire solar spectrum range to achieve a strong photothermal effect. On the other hand, the ultra-broadband absorption allows tuning the working wavelength region to NIR and mid infrared (MIR) by simply adjusting the period and thickness of the gratings. The creation of microgratings using fs laser ablation on GO thin films has been demonstrated [Fig. 12(d)], which can be applied as a polarizer^[226]. The polarizer shows a large extinction ratio ($\sim 20\mathrm{dB}$) and controllable working wavelength from 2 up to 25 µm. The thin film design allows direct integration of the polarizer on PICs by combining conformal coating and laser nanofabrication technology.

In addition, TMDC grating waveguides have also been fabricated via laser nanofabrication. By directly patterning ${\mathrm{MoS}}_2$ using a fs laser, Mooshammer et al. demonstrated a grating coupler with a groove size down to 250 nm. The decent in- and outcoupling functionalities of the waveguide couplers were also tested and verified in the experiments of waveguide second-harmonic generation and nano-optical imaging, respectively^[227].

As mentioned in Sec. 2.3, optical lenses are one of the indispensable components in nearly all optical applications, and this is also the case for PIC integration. Conventional optical lenses rely on light refraction, which are bulky and unsuitable for on-chip integration. In the past decades, tremendous efforts have been devoted to designing ultrathin flat lenses to realize the integration and minimization of nano-optics and photonic systems. The emergence of 2D materials has opened new avenues for such advancements. In 2015, based on the unique and giant refractive index and absorption modulation of a GO film induced by laser exposure, Zheng et al. demonstrated an ultrathin ($\sim 200\mathrm{nm}$) flat graphene metalens that shows three-dimensional subwavelength focusing with absolute focusing efficiency of $>32\%$ for a broad wavelength range from 400 to 1500 nm^[228]. Figure 13(a) illustrates the laser fabrication process of graphene metalenses. The capabilities of both amplitude and phase modulation make the graphene metalens concept fundamentally different from the conventional flat lenses, which mostly rely on either phase or amplitude modulation solely and provide unprecedented flexibility in designing new ultrathin flat lenses. Combined with the new design method^[229,230], it is possible to design and fabricate 2D material metalenses with high NAs and arbitrary focal lengths with various focal intensity distributions^[231,232] for applications in 2D/3D photonic chips^[233]. As one of the important properties of light, the orbital angular momentum (OAM) can also be generated and controlled by graphene metalenses [Figs. 13(d) and 13(e)]^[234,235]. In addition, it has been demonstrated that by attaching graphene metalenses on flexible and elastic substrates, the focal length can be comfortably tuned by in-plane stretching the lenses, as shown in Figs. 13(f)–13(h)^[236]. Those design and fabrication methods can be generally applied to various 2D materials^[191–193], not limited to graphene.

Laser nanofabrication has also been employed to fabricate compact 2D photodetectors. Lateral homo- and heterojunction-based photodetectors have been demonstrated using laser nanofabrication. Recently, a laser scanning technique was used to fabricate a lateral ${\mathrm{WSe}}_2$ p–n junction. It was found that the laser-oxidized product ${\mathrm{WO}}_x$ was responsible for the p-doping in ${\mathrm{WSe}}_2$^[208]. A doped ${\mathrm{WSe}}_2/{\mathrm{MoTe}}_2$ photodetector was also realized using laser scanning to achieve localized doping in ${\mathrm{WSe}}_2$. This photodetector has an on/off ratio of $\approx 104$, a dark current of $\approx 1\mathrm{pA}$, and a response time of 72 µs under the illumination of a 633 nm laser at zero bias^[237]. Localized laser oxidation of ${\mathrm{In}}_2{\mathrm{Se}}_3$ was also used to fabricate lateral ${\mathrm{In}}_2{\mathrm{Se}}_3\text{–}{\mathrm{In}}_2{\mathrm{O}}_3$ heterostructure photodetectors [Fig. 14(a)], achieving an ultrahigh photodetectivity, which is two orders of magnitude higher than the similar multilayer-based devices^[238]. Beyond heterojunction photodetectors, rGO grating photodetectors have also been produced via fs laser nanofabrication, as shown in Fig. 14(b)^[239]. By exploiting the cylindrical focusing of a fs laser on GO films, researchers successfully demonstrated uniform subwavelength grating structures with a high speed along with a simultaneous in-situ photoreduction process^[239]. Triangular grating-based photodetectors have also been fabricated via direct laser writing grating patterns on ${\text{FAPbI}}_3$ perovskite films, which have higher responsivity and detectivity than unpatterned perovskite film photodetectors^[240]. In addition, a better polarization detection performance was also demonstrated with this device^[240]. Recently, a self-powered photodetector has been proposed by introducing laser-induced periodic surface structures on SnSe films. LIPSSs led to an enhanced optical absorption compared with that of the unprocessed film, enabling a high-infrared response without external bias^[241].

In summary, laser nanofabrication has been a promising technique in the fabrication of photonic devices with 2D materials for advanced light manipulation, such as focusing, coupling, OAM generation, and polarization control. Although challenges exist, laser nanofabrication is poised to enable significantly broader applications in the foreseeable future.

Advanced beam shaping is essential for achieving high-resolution laser writing of waveguides with precise cross-sections and feature sizes, particularly in 3D structure fabrication. There are three main beam-shaping techniques in laser nanofabrication: focal spot shaping, aberration compensation, and multifocal generation.

The resolution and the cross-section of the laser-fabricated structures depend on the shape and size of the focal spot of the fabricating laser beam. In general, the laser focal spot has an elliptical shape with a long axis along the optical axis caused by the inherent focusing nature of an objective lens. This asymmetry makes it unsuitable for fabricating highly symmetric structures, particularly optical waveguides, which require circular cross-sections for low loss. To address this challenge, advanced beam-shaping techniques are required to shape the laser focal spot and improve the quality of laser-fabricated structures.

Numerous beam-shaping techniques have been proposed for improving the performance of fs-laser-written waveguides, such as the slit beam-shaping technique^[242,243], astigmatic beam-shaping technique^[244,245], simultaneous spatiotemporal focusing beam-shaping technique^[246,247], deformable-mirror beam-shaping technique^[248], and spatial light modulator (SLM) beam-shaping technique^[249–252].

The slit method simply used a rectangular mechanical slit to limit the NA along one direction (the width of the rectangle), which effectively expands the width of the focal spot along that direction^[242,243]. Meanwhile, the NA along the other orthogonal direction (the length of the rectangle) is maintained to keep the resolution along the optical axis. In this way, the aspect ratio of the focal spot can be effectively tuned by the width of the slit. Figure 15(a) illustrates the laser writing in bismuth borate glass with slit beam shaping^[253]. By optimizing the slit width, nearly symmetric structural modification can be achieved, as shown in Fig. 15(b). Actually, the slit method can be considered as the simplest way to shape the laser focal spot. For the fabrication of curved waveguides, the direction of the slit should be rotated dynamically to ensure the width of the slit is always perpendicular to the direction of the waveguide. In this case, a mechanically rotating silt is needed^[254]. Recently, by introducing an adjustable mechanical slit, reduced birefringence and an optimized cross-section pattern of the refractive index modulation can be achieved in the laser writing of fiber Bragg gratings^[255].

Instead of placing a slit, the astigmatic beam-shaping technique utilizes an astigmatic cylindrical telescope to reshape the beam before focusing^{[244,245,256]}. Figure 15(c) illustrates a typical setup of a laser nanofabrication system with this method^[256]. By finely controlling the distance between two cylindrical lenses, the focal plane of the input beam in the $y$ direction can be adjusted, finally realizing the astigmatic difference tuning and beam spot reshaping. This method can flexibly shape the focal spot size, which allows for the fabrication of waveguides with circular cross-sections and variable sizes.

Simultaneous spatiotemporal focusing beam-shaping is a widely used technique to achieve 3D isotropic nanofabrication^{[247,257–259]}, a typical setup that is illustrated in Fig. 15(d)^[257]. A pair of gratings is employed to spatially separate the spectral components of the input laser pulse, generating an array of beamlets at various frequencies. The beamlets with different frequency components then recompile at the focal plane where the temporal focusing occurs, restoring the pulse with the shortest pulse duration. In this manner, the isotropic fabrication resolution can be improved significantly. Figure 15(e) compares the calculated laser intensity distributions at the focus without (top) and with (bottom) the simultaneous spatiotemporal focusing beam shaping in the $XZ$ plane^[246].

During laser nanofabrication, a dynamically tunable focal spot is usually preferable to optimize fabrication efficiency and waveguide cross-sections, especially in the writing of structures with complicated geometry. Adaptive optical elements, such as liquid crystal (LC) SLMs, deformable mirrors, and digital mirror devices, have been widely implemented for advanced beam shaping as well as dynamic modulation of focal spots^[25,260]. Here, we briefly summarize these techniques. More detailed introductions of their operational mechanisms and applications have been reviewed previously^[260]. One example of the SLM beam shaping is the adaptive slit generation^[261]. Figure 16(a) shows the corresponding experimental setup. A pinhole with an adjusted position is placed before the objective so that only the first-order diffraction light from the blazed grating on SLM passes completely. In this case, one can define the effective slit aperture in the pupil according to the shape of the grating region on the SLM. Using this method, researchers have fabricated single-mode waveguides with a propagation loss of $<0.4\mathrm{dB}/\mathrm{cm}$^[261]. One of the most attractive features of the SLM approach is its high flexibility in shaping beam geometries. The phase shift introduced by each pixel on the SLM can be designed to create desired patterns, which can be used to realize the arbitrary shaping of the focal spot. Figure 16(b) displays various laser spot geometries achieved, from ring and triangle to hexagon, by applying different hybrid holograms on SLM^[262]. In addition, more advanced beam structures, such as Bessel beams^[263,264] and Airy beams^[265], can also be realized for the fabrication of different writing tasks. Similarly, deformable and digital mirror devices allow adjustable surface curvature or orientation switch of the micro-mirrors, enabling arbitrary beam shaping at the focal plane^{[248,266,267]}.

In addition, the rapid development of metasurface optical devices provides more possibilities for laser pulse shaping. By designing specific surface nanostructures and patterns, metasurfaces enable the manipulation of the amplitude, phase, polarization, or spatial wavefront of the incident light^[268–270]. For instance, a metasurface-based Fourier pulse shaper has recently been developed for synthesizing arbitrary spatiotemporal ultrafast pulses^[271]. By optimizing the metasurface structures, arbitrarily engineering the instantaneous polarization of the incident pulses can also be achieved^[270]. These techniques are promising to be deployed in future laser nanofabrication systems to further expend the flexibility in fabricated structures.

In addition to beam shaping, aberration compensation is crucial in the case of manufacturing high-refractive-index crystal materials, for example, lithium niobate. Spherical aberration causes strong deformation of the focal spot, seriously degrading the fabrication quality. This becomes increasingly problematic when using high-NA lenses. This deformation degrades the quality of the fabricated structures, making aberration compensation necessary for achieving high-quality waveguide structures. In principle, the aberration can be corrected by applying the opposite phase, usually achieved using an adaptive optical element placed in a plane conjugate to the objective pupil, as illustrated in Fig. 17(a).

Figure 17(b) compares the structures written in diamond with and without aberration compensation^[273]. The writing depth is fixed at 80 µm beneath the surface. Due to the high-NA focusing and a large refractive index mismatch on the sample surface, a severe spherical aberration along the laser incident direction ($z$-axis) can be observed. This aberration causes distortion in the fabricated structures, which can be substantially mitigated using aberration compensation.

An example of the laser nanofabrication system with dynamic aberration compensation is presented in Fig. 17(c)^[274]. In this system, a feedback loop is designed in which a pulse tailoring unit (using an SLM) corrects the incoming pulse spatial phase. The detection of the sample modification provides a real-time evaluation of the dynamic laser focal condition. With sequence-by-sequence processing, an optimized phase pattern can be applied to compensate for the aberration. A comparison of the effects of theoretical and optimized correction phases is also presented in Fig. 17(c). Using a feedback loop system, a more effective aberration compensation can be achieved at the same depth in the sample. In addition to monitoring the sample modification, a direct wavefront sensing technique has also been employed in the feedback loop^[276,277]. Based on a single phase-only SLM, a multichannel interferometric wavefront sensing technique has recently been developed and applied in a two-photon laser writing system^[276]. This technique provides more accurate detection of the beam aberration, therefore offering better dynamic aberration corrections. With the assistance of machine learning technology, the accuracy and efficiency of aberration correction will be further improved^[278,279].

Compensating for the spherical aberration is especially critical when fabricating complex nanostructures with a high degree of uniformity. Figures 17(d) and 17(e) illustrate the laser-fabricated 3D gyroid photonic crystals in chalcogenide materials with a high NA objective with and without aberration compensation, respectively^[275]. Comparing the two fabricated structures, a clear improvement in uniformity and reduced elongation can be achieved when applying aberration compensation. Recently, laser-written optical neural networks have attracted increasing interest^[280,281]. Aberration compensation is critical in the fabrication of these devices, which require a uniform network structure and smaller device footprints.

To meet the demands of large-scale production for real-life applications, the speed of laser nanofabrication should be substantially improved. One of the most effective ways to speed up the process without compromising the resolution is to generate multifocal spots, which can enhance the speed hundreds of times.

Microlens arrays were proposed and used to produce multiple spots for parallel laser fabrication [Fig. 18(a)]. Using this method, 2D/3D structures with a lateral resolution of up to 100 nm can be fabricated^[282–285]. Figure 18(b) depicts an example of the fabricated micro-letter array of “N” using this technique^[283]. However, the microlens solution only supports fixed beam patterns and requires input beams with perfectly homogenous beam intensity, which is quite challenging in practical fabrication processes, therefore limiting its widespread application. In contrast, multifocal array generation with SLM features reconfigurable arbitrary diffractive patterns, providing a more practical and efficient technique^{[250,286–288]}. Based on this technique, Lin et al. generated a multifocal circularly polarized vortex beam and fabricated an array of 3D split-ring (SR) patterns in polymer [Fig. 18(d)]^[286]. The experimental setup is illustrated in Fig. 18(c). Zhang et al. successfully utilized an SLM to shape a fs laser Gaussian beam into a laser beam possessing discrete annular intensity distribution. This approach enabled the rapid fabrication of circular depressed-cladding waveguides inside ${\text{LiNbO}}_3$ crystals in a single step^[250]. This one-step processing technique dramatically improves the fabrication efficiency in transparent materials, showcasing the transformative potential of SLM-based multifocal array generation for high-throughput laser nanofabrication.

In addition, Li et al. reported an SLM-based technique for the rapid fabrication of 3D waveguides with circular cross-sections^[249]. In this work, SLM can change a fs laser single focus into a multifocal array. By adjusting the distance and arrangement between sub-focal points in the multifocal arrays, the cross-sectional morphologies of the fs-laser-written optical waveguides can be conveniently and flexibly controlled. The schematic of the experimental setup is indicated in Fig. 18(e). Figures 18(f) and 18(g) depict the cross-sections and mode profiles of the waveguides fabricated by different multi-foci arrays. The circular cross-sectional single-mode waveguides could be directly written by the multi-foci array. This beam-shaping technique is significant for flexibly tailoring cross-sectional morphologies of waveguides and improving the guiding properties of waveguides.

It should be noted that the high-performance waveguides could still be fabricated without using any beam-shaping techniques. For example, when the waveguide depth is not a critical requirement, low-loss single-mode waveguides with circular cross-sections could be produced by optimizing the fs laser processing parameters (e.g., repetition rate, pulse energy, scanning velocity, and focusing depth). For example, in 2022, Liu et al. successfully fabricated low-loss single-mode waveguides in borosilicate glasses (Eagle XG) with this parameter-optimization method^[290]. Essentially, this kind of parameter-optimization method reduces the effects of spherical aberration by optimizing fs laser processing parameters. This method is also practical to produce high-quality waveguides, especially in cases of low requirements for waveguide depth.

PICs are composed of various 2D/3D functional components, which may be fabricated with different materials. Integrating and connecting these components efficiently without losing the high performance and functionality is critical to achieving a PIC. Heterogeneous integration refers to the integration of separately manufactured components into a higher-level assembly that, in aggregate, provides enhanced functionality and improved operating characteristics. Previous work has demonstrated that laser nanofabrication, especially direct-write two-photon lithography, is a viable tool for connecting photonic chips of different materials—either by 3D free-form single-mode waveguides, i.e., the so-called photonic wire bonds, or by facet-attached beam-shaping elements such as micro-lenses that allow for highly efficient coupling with relaxed alignment tolerances.

Photonic wire bonding (PWB) utilizes in-situ additive nanofabrication to create free-form polymer waveguides between pre-placed photonic chips. The 3D shape of the photonic wire bonds can be adapted to the exact positions of the chips such that high-precision alignment of chips becomes obsolete, rendering the technique amenable to automated mass production^[83,291].

The main technique for PWB is based on two-photon lithography in ultraviolet-sensitive polymers, such as SU-8 and negative-tone photoresist^[83,292,293]. Figure 19(a) illustrates a multi-chip system fabricated using PWB^[83]. Both the chip-to-chip and chip-to-fiber interconnections are included in the system. The fabricated PWB chip-to-chip interconnect is presented in Fig. 19(c). Efficient coupling of III–V light sources to silicon photonic circuits has also been achieved via the PWB technique^[292,294]. In this work, direct-laser-writing two-photon lithography and negative-tone photoresist were used to fabricate a hybrid photonic multi-chip module, realizing chip-to-chip and fiber-to-chip connections, respectively. Figure 19(d) shows the SEM image of a hybrid multi-chip module. Insertion losses of PWB1 to PWB4 are also indicated, demonstrating a minimum insertion loss down to 0.4 dB.

Optical communication engines that rely on PWB for connecting arrays of silicon photonic modulators to InP lasers and single-mode fibers have been demonstrated recently^[36]. Figures 20(a) and 20(b) illustrate an eight-channel transmitter that combines the InP laser array with electro-optic modulators on a single silicon photonic chip. Using dedicated test chips, automated mass production of photonic wire bonds was also verified, featuring an insertion loss of $(0.7\pm 0.15)\mathrm{dB}$. As shown in Fig. 20(c), the fabricated wire bonds exhibit good resilience in environmental-stability tests. In addition, recently, Ma et al. realized a compact and efficient photonic convolution accelerator based on a hybrid integrated multi-wavelength DFB laser array by PWB^[295]. The photonic convolution accelerator operates at 60.12 GOPS for one $3\times 3$ kernel with a convolution window vertical sliding stride of 1 and generates 500 images of real-time image classification.

In addition to PWB, laser-fabricated free-form optical components, such as microlenses, micromirrors, and grating couplers, can also be used to directly connect 2D and 3D chips in PICs. Figure 21 illustrates how these components are used to connect different functional devices in a multi-chip system. A printed free-form microlens and grating coupler support the connection between InP lasers with the SiP chip. Beam expanders are used to connect the single-mode fiber array from the outside. These laser-fabricated optical components offer good mode-field adaptation, thereby enabling low-loss coupling and relaxed alignment tolerances^[103].

Efficiently coupling optical fibers to SiP processors is critical in telecom and data centers. Figure 22(a) depicts a coupling strategy between a single-mode fiber array and an array of edge-emitting SiP waveguides using high-precision 3D-printed facet-attached microlenses, which are fabricated via multi-photon lithography. The microlens allows for low-loss coupling of $\sim 1.4\mathrm{dB}$ to edge-emitting SiP chips as well as pluggable optical connections based on simple mechanical alignment structures^[296]. In addition to planar connection, a 3D free-form coupler that supports vertical coupling to devices has also been demonstrated^[297,298]. Figure 22(b) shows a laser-fabricated low-loss fiber-to-chip vertical coupler on the silicon photonic platform^[297]. Such a coupler features low insertion loss ($<1\mathrm{dB}$), providing a broad working wavelength range for both TE and TM polarizations over the entire C-band. Another advantage of this device is its large tolerance for misalignment of the coupling fiber, up to 4.5 µm for a 1 dB loss, which enables the development of relaxed alignment techniques^[297].

Using ultrafast laser nanofabrication, waveguide fan-in/fan-out devices can be fabricated to directly connect SiP chips to optical fibers. In 2007, Thomson et al. developed a fan-out device that enables each core of a multicore optical fiber to be connected to a single-mode fiber held using a fiber V-groove array^[299]. An 84-channel fused silica 3D waveguide fan-out device was also demonstrated for high-density edge coupling of multicore fibers to a SiP chip^[300]. To fabricate the device, fs laser irradiation, combined with chemical etching, was used to create alignment sockets, allowing for quick assembly and precise locking of multicore fiber positions^[300]. A photo of the fabricated silica fan-out device is displayed in Fig. 22(c).

In addition, using laser nanofabrication, 3D polymeric bridge waveguides were also fabricated on prefabricated planar optical chips to directly connect ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguides. Figure 22(d) illustrates the structure of the 3D bridge waveguide with the insets showing the cross-section mode profiles at different positions. The fabricated device is shown in Fig. 22(e). Its adiabatically twisted shape along its axis enables the geometrical rotation of linear polarization on the chip^[301].

Another 2D/3D integration in PIC fabrication involves the use of 2D materials. As discussed in Sec. 3, photonic devices incorporating 2D materials offer significant property and footprint advantages, promising to be implemented in high-performance PIC fabrication. 2D material integration is essential for the fabrication of these devices. While the integration techniques have been reviewed previously^[28,302,303], those reviews have primarily focused on 2D devices rather than the 3D scenarios. In this subsection, we provide a summary of these techniques and discuss their potential applications in future 2D/3D device integration.

Depending on the synthesis method of 2D monolayers, there are different integration strategies, mainly including dry and wet transfer, inject printing, direct deposition, as well as layer-by-layer assembly techniques, as illustrated in Fig. 23.

Dry and wet transfers are used to directly stack the as-prepared 2D monolayers onto a target structure. Auxiliary media, such as PDMS and PMMA, are needed in these transfer processes^[28]. For wet transfer, chemical solutions are also needed to detach the monolayer from its deposition substrate followed by subsequent cleaning. The fabricated hybrid devices integrated with graphene^[304,305] and black phosphorus^[306] are illustrated in Figs. 24(a)–24(c), respectively. The dry and wet transfers were mostly based on manual operation in the early stages. However, with a decade’s development, automatic transfer techniques that are advantageous in mass production have also been proposed. Inkjet printing is adopted to integrate liquid-exfoliated 2D materials. This technique allows rapid coating on various substrates and supports maskless patterning of 2D films. In addition to transfer and printing methods, direct deposition of 2D materials onto 3D structures has been demonstrated recently. For example, atmospheric pressure CVD was employed to grow graphene directly on silica^[307]. A modified two-step CVD method has also been developed for depositing monolayer ${\mathrm{MoS}}_2$ on silica structures^[308]. In addition, using a thermally assisted conversion and atomic layer deposition (ALD) method, area-selective growth and conformal coating of ${\mathrm{PtSe}}_2$ films onto substrates having different topologies, including trenched substrates and striped waveguides, have also been demonstrated^[309]. Figures 24(f) and 24(g) show the SEM images of the top and side of a silicon trench coated with ${\mathrm{PtSe}}_2$.

Self-assembly is another solution-based integrating technique for 2D materials. Mechanisms such as electrostatic attachment^[310], hydrogen binding^[311], and ionic charge transfer^[312] are implemented to realize layer-by-layer integration. Compared with the above-mentioned transfer and deposition approaches, self-assembly offers more flexibility in 2D/3D structure fabrication. The layer-by-layer coating process allows precise control of the layer numbers, thus enabling film thickness control on the nanoscale. This unique feature also brings benefits in sidewall coating, allowing conformal applications on complex device structures, such as a silicon waveguide^[29] and nanopillar array^[34], as shown in Figs. 24(h) and 24(i). Moreover, self-assembly is highly scalable, with the film area being limited only by the size of the solution container^[44]. Using larger containers, this technique has a high potential to be used in future industrial production.

Although current integration techniques have achieved significant success in fabricating devices with 2D configurations, their application in 3D PIC fabrication remains challenging and requires distinct strategies. Dry and wet transfer methods enable only surface integration, presenting difficulties in achieving conformal coatings on 3D structures. They are better suited for integrating 2D materials on substrates with embedded 3D structures or for coating the input and output side ports of 3D waveguides. For integration with devices featuring 3D configurations, layer-by-layer self-assembly and direct deposition techniques are more feasible, as they allow for conformal coatings of 2D materials on the structures. However, limitations in production efficiency and process stability still hinder the industrial deployment of these techniques. In practical 3D PIC chip fabrication, integration becomes even more complex, as multiple functional modules must be incorporated into a single chip. Therefore, optimizing existing solutions and developing more efficient 2D material integration techniques are crucial.

Thanks to their exceptional device performance, 2D/3D PICs have found applications across a diverse range of fields, from optical communication systems and quantum computing to optofluidics in biomedicine, optical metrology, and astrophotonics. In this section, we provide a broad overview of these applications to highlight the extensive potential of PICs and cover as many areas as possible.

PIC was initially born for applications in optical communication systems to overcome the limitations of conventional electronic approaches in bandwidth, processing speed, and power consumption. On-chip photonic components, including light sources and amplifiers, as well as optical modulators, couplers, and photodetectors, are the key building blocks of an optical communication PIC. Laser-fabricated functional devices have been widely utilized to realize these functions.

The capability to generate lasers on-chip has been a long-sought-after goal for fully integrated photonic circuits^[313,314]. A planar waveguide laser fabricated by direct laser writing is one of the promising solutions. Based on doped glass and crystal material platforms, such as Yb-doped bismuthate glass^[97], Ho-doped fluorozirconate glass^[167], as well as Nd- or Yb-doped YAG^[315,316], and Tm-doped ${\mathrm{YVO}}_4$^[317,318], the fabricated waveguide lasers were capable of emitting laser pulses with a wavelength ranging from NIR to MIR band. Though most of these proof-of-concept demonstrations were based on isolated device structure, they can be easily connected to other photonic components on one chip via simple PWB or overall chip layout design and fabrication. Recently, incorporating 2D materials and their heterostructures provides new possibilities in waveguide lasers. Their excellent light emission and nonlinear optical properties bring numerous benefits, including improved emission efficiency and laser energy^[319,320] as well as more flexible manipulation of laser pulse duration^[321,322].

Optical modulators are critical in optical communication systems, being used to manipulate the phase, polarization, or intensity of a light beam. Based on the operational mechanisms, these devices can be divided into three categories: electro-optical, all-optical, and thermal-optical modulators. Due to the large electro-optical (EO) coefficient, ${\text{LiNbO}}_3$ is a perfect platform for the fabrication of on-chip electro-optical modulators. Laser-fabricated ${\text{LiNbO}}_3$ 2D structures, such as Mach–Zehnder interferometers (MZIs)^[325,326], waveguide Bragg gratings^[323,327], and microresonators^[328], have been used for electro-optical modulators, achieving a maximum extinction ratio larger than 10 dB. In addition to ${\text{LiNbO}}_3$, on-chip polymer EO modulators have also been realized by direct-laser-writing MZIs in SU-8 with nonlinear polymer cladding, as shown in Fig. 25(a)^[98]. Though the achieved EO efficiency is smaller than the devices based on ${\text{LiNbO}}_3$, this work demonstrates the flexibility of laser fabrication in different material platforms. The recent demonstration of laser-fabricated high-speed all-optical modulators based on polymer nanofiber Bragg gratings^[329], as well as organic-inorganic hybrid thermal-optical modulators^[330], provides more options for on-chip modulators in optical communication systems.

Wavelength conversion is critical for broadband signal generation and processing. Second-harmonic generation (SHG) is the main mechanism for these devices. In terms of the material platform, ${\text{LiNbO}}_3$, which features strong second-order optical nonlinearity, is an excellent candidate. Based on birefringence phase matching design, direct-laser-written single-line waveguides in ${\text{LiNbO}}_3$ crystal have been demonstrated as good wavelength converters for frequency doubling of 1064 nm light^[331,332]. Laser-fabricated on-chip ${\text{LiNbO}}_3$ microresonators have also been fabricated for converters operating at the telecommunication band^[333,334]. In addition to birefringence phase matching, quasi-phase matching is another strategy to enhance SHG, therefore improving the wavelength conversion efficiency. Due to its 2D/3D poling structure as well as reconfigurable capability^[160,163], fs laser poling has been a promising technique to achieve quasi-phase matching in ${\text{LiNbO}}_3$ and fabricate high-efficiency wavelength conversion devices^[58,335,336]. The fabricated devices exhibit a maximum efficiency up to 17.45% at near-infrared wavelengths^[58].

In addition to active devices, laser-fabricated waveguide couplers and grating structures have been functionalized as various on-chip passive devices, such as polarization multiplexers^[324,337], mode multiplexers^[338,339], and fan-in/fan-out components^[300,340]. Figure 25(e) gives an example of an on-chip polarization directional coupler designed for polarization multiplexing^[324]. This device demonstrates effective polarization projection on different pairs, showing a great potential for applications in large-scale polarization-based optical communication systems.

PIC for quantum computing is advantageous by manipulating photons, which allows for room temperature operation and long-distance propagation. Three main building blocks, including quantum state generation, manipulation, and detection, form a quantum photonic system.

Single-photon generation forms the basis of photonic quantum processing. The most common single-photon generation is currently from spontaneous parametric down-conversion sources^[341,342] to quantum dots^[343,344] or defect-based sources^[345,346]. Diamond with nitrogen vacancies is one of the representative platforms. Fs laser writing has been successfully used to induce nitrogen vacancies on the surface or in diamond for high-performance single-photon generators^[347]. Efforts have also been made to directly fabricate integrated single-photon-emitting devices, which are increasingly desirable in large-scale quantum information processing. Though several raw demonstrations of diamond waveguide structure were reported, the most promising approach for photon pair generation is to utilize nonlinear processes, such as spontaneous four-wave mixing and spontaneous parametric down conversion, in integrated platforms, for example, silica and LiNbO₃. Laser-fabricated silica waveguides have been proposed and demonstrated for generating heralded single photons via spontaneous four-wave mixing^[99]. Figure 26(a) shows the experimental setup used for the single-photon generation. Based on spontaneous parametric down conversion, hybrid chips by combining periodically poled ${\text{LiNbO}}_3$ waveguides and laser-fabricated borosilicate glass circuits were fabricated to generate non-degenerate photon pairs at the telecommunication wavelengths^[342,348]. In addition to the traditional platform, 2D materials, such as h-BN^[345,349] and ${\mathrm{WSe}}_2$^[346] with atomic defects, have also been demonstrated to be decent single-photon sources. Integrating these materials with waveguide structures or directly fabricating waveguide structures based on them provides another attractive strategy.

Quantum logic gates are critical in quantum computing systems. Single- and two-qubit photonic gates have been created with laser-fabricated 2D/3D directional couplers. For example, a single-qubit photonic quantum state was realized using a laser-written waveguide circuit featuring high two-photon interference visibilities^[341]. In 2011, Snigirev et al. demonstrated the first integrated photonic controlled-NOT (CNOT) gate for polarization-encoded qubits^[352]. The chip was fabricated by laser writing partially polarizing beam splitters on a glass chip. Using the same technique, Zhang et al. further realized a path-encoded CNOT gate [Fig. 26(b)]^[350]. The gate was formed by five directional couplers, each with precisely controlled splitting ratios and stable symmetric phases inside glass^[350]. Figure 26(c) illustrates the experimentally constructed CNOT logical truth table, in which an average fidelity higher than 0.98 can be observed. Recently, a two-qubit photonic processor has been realized with a low-loss reconfigurable photonic chip fabricated via fs laser writing^[353]. An exemplary application to estimate the ground state energy of an ${\mathrm{H}}_2$ molecule was also demonstrated using the variational quantum eigensolver algorithm.

Photonic quantum memory is another core element in quantum networks, which has been realized in laser-written waveguides^[354–357]. The single-photon storage was first demonstrated using laser-fabricated type I waveguides in ${\mathrm{Pr}}^{3+}:{\mathrm{Y}}_2{\mathrm{SiOB}}_5$^[354]. Based on the type-IV waveguide fabricated on the surface of ${\mathrm{Eu}}^{3+}:{\mathrm{Y}}_2{\mathrm{SiO}}_5$, spin-wave atomic frequency comb storage was also demonstrated, achieving a high interference visibility (up to 97%±1%) between the retrieval pulse and the reference pulse^[355]. By introducing a Stark modulated atomic frequency comb, the same group further demonstrated on-demand storage of time-bin qubits in a similar waveguide structure with a controlled storage time exceeding 2 µs, far beyond the pulse width ($\sim 100\mathrm{ns}$)^[356]. Moreover, quantum optical memories operated in the telecommunication band have also been achieved in ${\mathrm{Er}}^{3+}:{\text{LiNbO}}_3$ waveguides recently^[358,359], allowing storage of up to 330 temporal modes of heralded single photons with 4 GHz wide bandwidth at 1532 nm and a 167-fold increase of coincidence detection rate with respect to the single mode^[359].

Single-photon detectors (SPDs) are essential for detecting and counting individual photons in quantum computing systems. Though stand-alone single-photon detection units have been widely used, integrating these units into PICs is more attractive either in the detection efficiency or in a compact footprint^[360,361]. To fabricate integrated SPDs, electron beam lithography is the leading technique to directly pattern superconducting nanowire detectors, for example, NbN and NbTiN, on Si^{[360,362,363]}, ${\mathrm{Si}}_3{\mathrm{N}}_4$^[364,365], and ${\mathrm{Ta}}_2{\mathrm{O}}_5$^[366] platforms. Recently, a series of inspiring achievements using laser nanofabrication have been reported in this area. The first example is a laser-written waveguide-coupled NbN nanowire SPD [Fig. 26(d)]^[351]. The chip is composed of a 3D waveguide array, including detector waveguides and auxiliary waveguides with NbN nanowires on top of the waveguide cross portion. All-laser fabrication of integrated SPDs has also been demonstrated via direct writing WSi superconducting nanowires on a silicon substrate^[367]. By optimizing the wire width and length, saturated detection efficiencies at 775 and 1550 nm were verified. What is more, laser-fabricated photonic circuits can also be used in the detection of non-classical states of light. For example, sampling the statistics of multiphoton states of light was demonstrated in a fused silica multiplexing waveguide chip, which was fully fabricated by fs laser inscription^[368].

Biomedicine-optofluidic devices, which merge the benefits of integrated photonic circuits with microfluidic chips, have attracted increasing attention for accurate biosensing and imaging applications in the past decade. Laser nanofabrication with high flexibility in both the device structure and material platforms plays an important role in the fabrication of these devices.

A laser-fabricated microfluidic channel is the first example. By simply adjusting the inscription and following etching parameters, 2D/3D microfluidic channels with arbitrary lengths and configurations have been achieved in glass and fused silica via direct laser writing^[369–371], featuring a channel resolution down to sub 50 nm^[372].

A more exciting advancement is the integration of different functional micro-components into one optofluidic chip by direct laser writing. A microfluidic biosensor consisting of laser-fabricated microchannels and longitudinal optical waveguides in glass has been fabricated for single-blood-cell detection^[373]. With a similar device structure, a curved waveguide was combined with microchannels to perform the classification of microspheres and algal cells [Fig. 27(a)]^[374]. Laser-fabricated MZIs have also been widely integrated into on-chip optofluidic devices^[375–378]. Figure 27(b) illustrates an example of this kind of device fabricated in fused silica, in which the sensing arm of the MZI crosses the microfluidic channel with a reference arm passing over it^[375]. The 3D structure allows for spatially resolved sensing, achieving a detection sensitivity down to ${10}^{-4}$ refractive index units in glucose solutions.

A microlens array fabricated by two-photon polymerization is another attractive functional structure for optofluidic chip applications. These microlens arrays enhance signal intensity and excel at simultaneously imaging micron-scale objects over large areas, significantly improving detection efficiency^[379]. High-sensitivity and high-throughput droplet and single-cell analyses have been demonstrated by integrating a microlens array and micromirror in an on-chip microfluidic system^[379]. A 3D-cascade-microlenses optofluidic chip composed of four detection channels has also been fabricated recently. The detection channels feature different microlens configurations [Fig. 27(d)], allowing adjusted sensitivity by varying the microlens number^[380]. A photograph of the fabricated chip is displayed in Fig. 27(e), with a general size comparable to a five-cent Swiss franc.

In addition, microresonators with high Q factors, including microspheres^[381], microrings^[10], and microdisks^[382], have also attracted intense interest in high-sensitivity biosensors. Laser nanofabrication has emerged as a powerful tool for fabricating these resonant structures, offering the precision and versatility needed for future advancements in optofluidic chip design. This technique holds promise for creating highly integrated optofluidic systems by seamlessly incorporating laser sources, detectors, and a variety of functional waveguides and microstructural components into a single chip. These advancements pave the way for the next-generation devices with unprecedented functionality and sensitivity.

Ranging and object detection are critical for the automobile industry in autopilot assistance, anti-collision, and target identification^[9,383,384]. Electronic chip-based radio detection and ranging (RADAR) encounters limitations in the generation and processing of broadband signals, therefore challenging to achieve high spatial resolution^[385,386]. PIC-based light detection and ranging (LiDAR) technology is promising to overcome these limitations, enabling highly precise and ultrafast detections^[387,388].

Optical phased array (OPA) LiDAR is one of the representative devices^[388,389]. Figure 28(a) shows a silicon-based OPA, in which a 2D array composed of $64\times 64$ optical antennas is integrated on a silicon chip^[389]. By applying different voltages on each antenna, various phase combinations can be obtained, generating a sophisticated radiation pattern in the far field for LiDAR applications. Polymer-SiN OPA has also been designed and fabricated recently^[390,391]. The fabricated hybrid OPA exhibits decent line beam scanning performance, demonstrating its high potential in the fabrication of practical LiDAR devices. Although these current advancements are mainly based on traditional CMOS techniques, laser nanofabrication is capable of OPA fabrication and will bring more benefits in fabrication efficiency as well as 3D structure flexibility^[392,393]. Laser nanofabrication also plays an important role in LiDAR chip packaging. A complete LiDAR system includes a variety of functional components, such as laser sources, detectors, as well as many other optical and data processing units. To integrate these different components, photonic wire bonding is an ideal technique, which is especially attractive for connecting 2D/3D structures^[394,395].

In addition to the LiDAR technique, the all-vision approach is another technical roadmap for autopilot assistance. This approach relies heavily on high-performance cameras for 3D imaging as well as advanced computer vision algorithms. As discussed in Sec. 2, two-photon direct laser writing is capable of fabricating versatile 2D/3D microlenses as well as more complicated multi-lens systems, which are required for high-resolution and large-field imaging^{[101,102,228,396]}. Recently, researchers have directly printed a multi-lens system onto a CMOS image sensor for high-quality imaging. Figure 28(d) shows a fabricated device with a laser-printed doublet lens array^[101]. The CMOS image sensor possesses a total active area of $2592\text{pixel}\times 1944\text{pixel}$ and a pixel size of $1.4\mathrm{\mu m}\times 1.4\mathrm{\mu m}$. Figure 28(e) displays the image of the USAF1951 resolution test chart taken through one hexagonal array of 19 printed doublet lenses, where the numbers and lines of the elements of group 2 can be clearly observed. This proof-of-principle demonstration paves the way for laser-fabricated optical devices in high-quality imaging applications.

Astrophotonics is the development of advanced imaging and spectroscopic instruments that can capture high-resolution images and spectra of celestial objects^[397]. The application of laser nanofabrication in astrophotonics was first proposed in 2009^[398]. To date, a variety of laser-fabricated PICs have been used in this field. For example, integrated photonic lanterns written in glass platforms [Fig. 29(a)] have been employed to enhance the resolution and sensitivity of adaptive optical systems in telescopes by coupling multimode light to single-mode waveguide arrays^[25,339,399]. Figures 29(b) and 29(c) show the transmission microscope images of the input facet and the diffraction-limited pseudo-slit output in a fabricated photonic lantern, respectively. The on-sky testing of these integrated devices in 2013 demonstrated the spatial reformation of a telescope point spread function into a diffraction-limited pseudo-slit^[399]. A nulling interferometer is another example that is critical to realize high-contrast imaging of exoplanets and circumstellar discs. By directly writing single-mode waveguides and evanescent directional couplers in one chip, a photonic nulling interferometer can be fabricated. Figure 29(d) is an on-chip nulling interferometer fabricated in borosilicate glass^[400]. The waveguide arrangement and coupling region in the chip are illustrated in Fig. 29(e). This chip has been tested in the Subaru Telescope, demonstrating a null-depth precision on the sky of $\sim {10}^{-4}$ and achieving milliarcsecond accuracy of the angular diameter of stars.

The past decade witnessed a rapid development of artificial intelligence (AI). The construction of artificial neural networks (ANNs) that can mimic the structural, functional, and biological features of human neural networks forms the basis of AI technology^[401,402]. PICs with high compactness and high stability are an ideal platform for ANNs.

Integrating phase-change materials with SiN waveguides has demonstrated on-chip photonic synapses^[403]. Figure 30(a) illustrates a device structure that resembles the neural synapses. The synaptic weight can be tuned by setting different optical pulse numbers sent down the waveguide. In addition to the isolated neutrals, a fully optical neural network has also been realized using a programmable nanophotonic processor composed of a cascaded array of 56 programmable MZIs in a silicon PIC [Fig. 30(b)]^[404]. The test on a vowel recognition problem verified that the chip possessed an accuracy comparable to a conventional 64-bit computer. In addition to 2D structures, 3D ANNs, which feature a larger number of neutrals and more complicated problem-solving capabilities, have attracted increasing interest recently^[280,405]. Among them, direct-laser-written 3D networks stand out. Using two-photon laser writing in a photoresist, as shown in Fig. 30(c), 3D Steiner tree networks with varied periodicities and rod sizes have been fabricated^[280]. Figure 30(d) presents a representative 3D Steiner tree network with a 2 μm unit size and a 200 nm rod diameter^[280]. Although these laser-fabricated 3D structures have not yet been applied in device-level AI applications, the potential for laser nanofabrication in this field is highly promising.

The past decades have witnessed tremendous progress in ultrafast laser nanofabrication in creating multi-dimensional functional photonic devices. This review offers an overview of the latest advancements in laser nanofabrication, focusing on its advantages for writing 3D structures and manufacturing functional 2D material devices for PIC applications. Advanced beam shaping techniques and 2D/3D integration based on laser techniques are also included and discussed. Additionally, we provide a comprehensive summary of the applications of laser-fabricated devices, demonstrating the tremendous potential of laser nanofabrication for future PIC fabrication. Despite these successes, challenges and new demands for future development remain.

Material platforms form the basis of device applications. For PICs, silicon^{[186,404,406]} is still the leading platform while other materials such as ${\mathrm{Si}}_3{\mathrm{N}}_4$^[2,407], SiC^[408,409], ${\text{LiNbO}}_3$^[323,327], as well as functional polymers^[98,410] and 2D materials^[29,195] have also emerged as promising candidates. Although laser nanofabrication has demonstrated its capability in diverse device fabrication based on these platforms, its application in the PIC industry is still limited by the trade-off between material processability and device performance. Materials that are highly compatible with laser nanofabrication may not always deliver the desired performance for specific device functions. For example, high-resolution 3D structures with arbitrary geometry can be achieved via two-photon laser writing in photo resists, which is highly advantageous for the 2D/3D photonic chip integration and wire bonding for chip packaging. However, these polymer-based materials usually encounter intrinsic limitations in thermal and mechanical stability as well as susceptibility to chemical degradation, therefore posing challenges in practical device applications. 2D materials bring significant performance benefits to PICs and have been regarded as one of the next-generation platforms for both electronic and photonic applications. But device fabrication is still challenging not only for laser-based techniques but also for the other COMS-compatible techniques. Thus, researchers continue to seek and develop advanced materials that feature both good fabrication capability and device performance.

Production efficiency is the most critical challenge for laser writing techniques when it comes to commercial device fabrication. Although impressive achievements with laser-written PIC chips have been reported, these remain proof-of-principle demonstrations conducted in the laboratory. Compared to industry lithography methods like EUV-photolithography and nanoimprint lithography, laser nanofabrication’s point-by-point writing process struggles to match in terms of speed. Multi-beam parallel writing based on SLM beam shaping is a feasible solution to improve the production yield^[249,411]. Realizing fully automatic processing can significantly improve laser fabrication efficiency. Adaptive optical components, such as SLM, digital mirrors, and gratings, are required to dynamically change and optimize the laser beam for different structure fabrications. Mechanical accessories, especially high-speed and high-precision translation stages for both the chip holder and objective lens, are also indispensable. In addition, in-situ monitoring and characterization of the fabricated structures are needed to provide real-time feedback for dynamic writing parameter adjustment and optimization. Moreover, integrating the laser writing system into the entire chip fabrication flow is essential but challenging, which needs to include new layout design and expanded facilities.

Another aspect is fabrication resolution improvement. Although previous reports have demonstrated a sub-10 nm 3D structure feature size using two-beam laser lithography in a two-photon absorption photoresist^[27], achieving nanometer laser writing resolution remains challenging, especially when balancing production efficiency and material generality. To improve the resolution, finding new materials with new fabrication mechanisms and applying new beam shaping techniques with adaptive optic components, a high-NA objective lens, and more accurate high-speed translation stages are needed.

Emerging technologies are continually enhancing laser nanofabrication, with artificial intelligence (AI) playing a transformative role. Over the past decade, AI has gained global traction impacting nearly every industry. Section 6 highlights the significant potential of laser-fabricated devices in developing neural networks for AI applications and the relationship is reciprocal. With its powerful capabilities in material prediction, property analysis, and functional device optimization^[412,413], AI is highly attractive in laser nanofabrication. The integration of AI into the laser nanofabrication system enables the streamlining of the previously repetitive and time-consuming processes, particularly in device design and laser writing parameter optimization. In addition, an AI-assisted feedback loop system offers the promise of real-time fabrication adjustments, which is highly advantageous when it comes to fabricating various functional units in a single PIC chip. Recent advancements in AI-assisted waveguide design^[414] and processing window optimization^[415,416] have confirmed a promising future of this field. As technology advances, the connection between laser nanofabrication and AI will only grow stronger.

Integrating 2D materials with functional 3D waveguide structures gives birth to new possibilities for high-performance PIC fabrication. To achieve optimized device performance, 3D conformal coating of 2D films on PICs with no gap in between is quite essential. Previous sections have discussed and compared various on-chip integration techniques of 2D materials in which conventional transfer mechanically exfoliated and CVD-grown 2D materials suffer from poor attachments on sidewall coverage and, therefore, are not capable of the 3D conformal coating process. In contrast, techniques based on layer-by-layer self-assembly and direct growth of 2D films have been two promising approaches by offering high flexibility in both the device structure geometry and coating layer control. This has been demonstrated by the recent fabrication of hybrid integrated photonic devices with GO and TMDC monolayers^[74,186,308]. However, one should note that these conformal coating techniques are still distant from being utilized in industrial mass production because of their limitations in large-scale manufacturing, integration, and production efficiency. Modifications based on these methods need to be further developed to meet the demand for practical mass device fabrication in the future.

PICs feature the integration of various functional components into a single chip. These components can be simple fiber-to-chip couplers, passive devices such as filters, or active laser sources and optical modulators, as well as more complicated units for quantum and photonic computing applications. Instead of only fabricating a single device, laser writing a whole PIC chip at one time is preferable to avoid contamination and connection failure during the following integration process, which is particularly important considering the continual increase of device complexity and chip density^[21,22,417]. This should involve the highly adjustable fabrication of different device structures and also in different active materials, requiring a system with much more flexibility than that used for single-device fabrication. In addition, laser fabrication of 2D-material-based devices is still in its infancy. As discussed above, for a single 2D material, challenges remain in high-resolution patterning, precise property modification, and 3D structure conformal coating. Recently, van der Waals (vdW) heterostructures that assemble different 2D crystals offer many new features and possibilities for PICs^[418,419]. Laser fabrication of corresponding devices is even more challenging with considering the interface interaction and twisting angles between different 2D layers.

The commercialization of laser nanofabrication technology is being driven by a growing number of companies that specialize in advanced laser systems and nanofabrication solutions. Companies such as Nanoscribe (Germany), Heidelberg (Germany), Optoscribe (UK), and Innofocus (Australia) offer cutting-edge laser-based systems for precise nanoscale patterning and 3D structure fabrication. The market for laser nanofabrication is expected to grow significantly in the coming years as demand for integrated photonic devices and systems with nanoscale structures and components increases exponentially with the boom of integrated photonic circuitry, quantum photonics, augmented/virtual realities, and head-up displays. Industries like semiconductor fabrication and PIC technology are experiencing a shift toward more versatile, cost-effective, and precise nanomanufacturing methods, where laser-based techniques offer clear advantages over traditional approaches. The widespread deployment of laser nanofabrication technologies accelerates innovation, reduces production costs, and enables the realization of more sophisticated and efficient devices, driving the growth of these industries. What is more, the ongoing research into improving the resolution and material compatibility of laser systems will further enhance the commercial viability of laser nanofabrication, making it a central technology for future industrial applications.

The capability of laser nanofabrication is only beginning to be fully explored, offering a vast and largely untapped potential for advancing PICs. This emerging technology allows for precise manipulation of materials at the nanoscale, enabling the creation of novel structures and devices that were previously unattainable with conventional fabrication techniques. By leveraging this new capability, researchers can develop advanced materials with tailored properties, innovative device designs with enhanced functionality, and unique fabrication mechanisms that push the boundaries of current technologies and integrate more devices on a single chip. These advancements promise to open new avenues for improving resolution, expanding functionality, and increasing the possibilities for PIC development. The exploration and optimization of laser nanofabrication hold the key to addressing longstanding challenges in photonics and unlocking unprecedented opportunities for the next-generation PICs in applications ranging from telecommunications to quantum computing, biomedical sensing, and beyond.