Femtosecond laser writing of integrated photonic circuits

Francesco Ceccarelli, Roberto Osellame


1 Integrated photonics

Photonics is a key enabling technology that is ubiquitous in today’s society. It is heavily employed in information communication due to the large bandwidth, immunity to electromagnetic interference and long distance range it can achieve. It is also an important tool for sensing, allowing remote, distributed and/or harsh environment measurements. It is obviously a key ingredient for devices related to vision and imaging, and it is also explored as a powerful platform for computing, both classical and quantum. To address all these applications in real-world conditions, the use of discrete and bulk components is not viable to produce compact devices with a high degree of complexity and thus we have to rely on integrated photonics (fig. 1). Integrated photonic devices are realized by fabricating all the components on a monolithic substrate in which light is guided inside transparent circuits (named waveguides) that enable its manipulation. With an analogy to concepts that may be more familiar to the general public, the electric cable that brings the signal to your electronic device (e.g. the ethernet cable) is the analog of the optical fiber in photonics, whose main scope is the transport of information, while the electronic microprocessor that encodes, retrieves or transforms information is the analog of the integrated photonic circuits, made of several interconnected optical waveguides.

The fabrication of integrated photonic circuits is a well-established and vast field. Many standard technologies are available for this task. Many of them try to leverage on CMOS-compatible processes, i.e. they try to exploit the same facilities that are used for the creation of microelectronic transistors. These technologies are capable of creating integrated photonic circuits in silicon or derived materials (e.g. silicon nitride). From the economic point of view this is clearly a big advantage as a large quantity of devices could be produced at a reduced cost. However, this is also the reason why integrated photonics is not yet as widespread as it could be. In fact, these large foundries have no interest in developing processes for small batches of products and this complicates the uptake of integrated photonics in new fields or niche applications. In addition, standard microfabrication technologies for integrated photonic circuit production, all rely on photolithography, where a circuit layout, engraved in a mask, is transferred to the surface of the transparent substrate with a sequence of steps encompassing illumination of a photoresist through the mask and selective etching of the exposed regions. This kind of process is very efficient in producing a large volume of replicas of the same device. However, it is very expensive and time consuming for the development of new devices. This adds to the previous motivation on why integrated photonics is not used in many applications where it could provide an added value. In conclusion, there is a significant space for a microfabrication technology capable of producing integrated photonic circuits in small/medium series with rapid and cost-effective prototyping capabilities. This technology exists and it is direct writing with ultrashort laser pulses or femtosecond laser waveguide writing (FLWW).

The possibility to produce optical waveguides by focusing ultrashort pulses (in the order of a few hundreds of femtoseconds, i.e. a duration that compares to a second as the second compares to the age of the universe) was first discovered in 1996. In these last twenty years, this technique has developed significantly and it is now a reliable, industrial-grade microfabrication tool employed by an ever-increasing number of companies. In the next section we will briefly review the main characteristics of this technology, emphasizing the multiple advantages it provides with respect to the standard alternatives.

2 Femtosecond laser writing in transparent materials

A transparent material is by definition a substrate that does not absorb the impinging light. In more physical terms, this is equivalent to saying that the energy of each impinging photon is not sufficient to bridge the material bandgap. Hence, without any electronic transition to trigger, the photon does not interact with the material. However, when ultrashort pulses are focused inside the transparent material many photons arrive at the same point at approximately the same time, which makes the event of a nonlinear absorption very likely. By nonlinear absorption we mean all those phenomena in which multiple photons collaborate to interact with the material and are then absorbed. An example of one of those nonlinear processes is multiphoton absorption, where an electronic transition is triggered by the simultaneous absorption of multiple photons that can match the transition energy by summing up all their energies. It is clear that nonlinear absorption is highly dependent on light intensity (that is proportional to the number of photons per unit time and unit surface), for this reason absorption only happens at the laser focus, while it becomes negligible as soon as the beam diverges. This mechanism creates a very localized structural change of the transparent material right at the beam focus. By suitably tuning the irradiation parameters, one can produce a very gentle modification causing a local refractive index increase without perturbing the other properties of the material. As depicted in fig. 2a, the translation of the laser beam focus with respect to the substrate results in the creation of arbitrary circuits that can confine and guide light.

FLWW of photonic circuits has a clear drawback, which is the serial nature of the process. This means that it may be less suitable for very large volume production of devices with respect to photolithographic approaches. Nevertheless, the simplicity of the process and the very high writing speed that can be achieved (several cm/s), make this solution very attractive also for medium to large scale production. Another important limitation of the process is that the refractive index increase that can be obtained is limited (comparable to that of an optical fiber). This is much lower than what can be obtained in, e.g., silicon photonics, where the index change is achieved by stacking very different materials like silicon and silica. The consequence of this fact is that miniaturization of the devices is less extreme with FLWW. On the other hand, this technology has many unique advantages. The most evident one is the capability to produce circuit layouts in three dimensions and not just two-dimensional as with standard photolithographic processes. Completely new classes of device topologies, such as 3D waveguide arrays (fig. 2b and c), are thus becoming accessible. In addition, FLWW can produce waveguide structures in very different materials, from glasses to crystals, and thus it is rather straightforward to produce multimaterial devices. Finally, the similar characteristics of a laser written waveguide with respect to those of an optical fiber make fiber coupling of laser written devices extremely low loss.

It is worth mentioning that FLWW is just one of the multiple tasks that can be accomplished by the interaction of ultrashort laser pulses with transparent materials. An additional example is water-assisted laser ablation that allows the ultraprecise removal of transparent material resulting in a 3D microstructuring of the substrate (fig. 2a). FLWW and laser ablation can be combined to provide devices with better performances. In the following section, we will discuss an important device that has been recently demonstrated by exploiting a combination of the previous laser processes: the universal programmable photonic processor. This device, fully programmable by the end user, will become an important building block in more complex photonic systems, with a role similar to that of field-programmable gate arrays (FPGAs) in electronic systems.

3 Programmable photonic processors

Let us now consider a photonic circuit that acts on a given set of $N$ input optical signals $E_{\mathrm{in}}$ by transforming them in a set of $N$ output optical signals $E_{\mathrm{out}}$. If we consider only linear phenomena with no attenuation (i.e. no photon losses) we can model the behavior of this device as follows:

(1) $ E_{\mathrm{out}} = U E_{\mathrm{in}} $

where $E_{\mathrm{in}} $, $E_{\mathrm{out}} \in C^{N} $ and $U \in U ( N )$ is a unitary matrix whose elements are also complex values characterized by a modulus and an argument. As a relevant example, let us start by considering the simple case of $N = 2$, in which $U$ is a $2\times 2$ matrix that can be universally expressed as:

(2) see equation in PDF,

where $\phi$ and $\theta$ are arbitrary angles so that $\phi \in [ 0, 2\pi ]$ and $\theta \in [ 0, \pi ]$. Beyond the mathematical formulation, what is important is the physical meaning of these two quantities. Indeed, following the bulk optics implementation proposed in fig. 3a, this matrix can be seen as the cascade of a first transformation, which acts on the two optical signals as a phase shift $\phi $ ‫ and a second transformation, which acts on the two optical signals as a beam splitter featuring a power reflectivity $\mathrm{sin}^{2} ( \theta / 2 )$. Translating this description into the language of integrated photonics, a universal $2\times 2$ transformation can be implemented by resorting to a Mach-Zehnder interferometer (MZI) as the one depicted in fig. 3b. In this configuration, after acquiring a phase delay $\phi$, the light entering the first input is initially split in two equal parts by a balanced directional coupler (i.e. a 50/50 beam splitter based on two evanescent-wave coupled waveguides) and then recombined by a second balanced directional coupler. Depending on the relative phase difference $\theta$ acquired along the two internal paths, the second directional coupler will reflect the light with the same power reflectivity law as the beam splitter of fig. 3a, thus in agreement with the transformation $U$ (eq. (2)). Obviously, dual considerations can be made also for the second input.

Among the qualities of integrated photonic circuits, the possibility of dynamically reconfiguring the optical transformation $U$ implemented by the device is something that is well appreciated, if not even required, in many applications. Therefore, the angles $\phi$ and $\theta$ are typically intended as variables that can be decided even at run-time. In other words, the device depicted in fig. 3b can be regarded as a programmable photonic processor able to implement any arbitrary unitary transformation on a set of two optical signals. Generally speaking, programmability is introduced in integrated photonic circuits by relying on different physical mechanisms and techniques, however in FLWW circuits the gold standard is the use of thermo-optic phase shifters. An electrical circuit is fabricated on the surface of the photonic device and, more specifically, resistive microheaters are patterned on the waveguides whose optical properties need to be reconfigured. By dissipating electrical power through the microheaters, it is possible to locally heat up the waveguide and, in turn, to have a reversible modification of the refractive index of the material. The final effect is a phase shift on the optical path targeted by the microheater. In order to implement effectively this technique in FLWW devices, a few design considerations must be made. First of all, waveguides must be fabricated as shallow as possible (depth < 30 μm from the surface) in order to guarantee a decent thermal coupling between the microheaters and the optical circuit. Secondly, thermally insulating structures must be employed in order to guide the heat diffusion towards the target waveguide, thus decreasing the power consumption of the device and limiting the crosstalk with other waveguides fabricated in the same substrate. Examples of insulating structures are microtrenches (fig. 3c) that are realized by 3D water-assisted laser ablation.

Such concepts can be extended in a straightforward way to produce processors able to manipulate $N$ optical signals. In order to do that, the $2\times 2$ MZI can be employed as the basic cell for a more complex interferometric network: a MZI mesh. Different planar MZI mesh layouts have been demonstrated to be universal and, among the most important, the square one (fig. 3d) is worth mentioning. The state of the art today is represented by the programmable photonic processors based on silicon nitride waveguides provided by the Dutch company QuiX and fabricated through the planar process provided by LioniX, world-renowned top-level fabrication facility. Such circuits are able to operate on $N = 20$ optical inputs/outputs (i.e. optical modes) with light between 900 and 1550 nm, photon losses of about 3 dB and an average fidelity of the implemented optical operation > 97.4%. On the other hand, photonic processors manufactured through a FLWW process have been demonstrated only up to 6 optical modes, however with a performance that is currently comparable to the state of the art and, in addition, with the fundamental advantage of being rapid, cost-effective and easily tailorable even to visible light with no detrimental effects on the performance. Moreover, FLWW photonic processors that feature insulating structures can claim also a total amount of power dissipation that can be orders of magnitude lower with respect to the silicon nitride technology, an advantage that cannot be ignored from a long-term scaling perspective.

Speaking of scaling, this is actually one of the hot topics of this fascinating field. The research on FLWW circuits is currently directed towards increasing the number of optical modes by keeping the same length of the circuit and, in turn, the same attenuation/losses. Indeed, this is particularly important in fields like quantum optics, in which the information is carried by individual photons and optical amplification is intrinsically not possible. If we consider a square MZI mesh, it is trivial to demonstrate that the optical length L of a square MZI mesh can be expressed as

(3) $ L=L_{\mathrm{MZI}} N$,

where $L_{\mathrm{MZI}}$ is the length of each MZI cell. It is evident from eq. (3) that optical length (i.e. photon losses) and complexity scale proportionally. A first strategy currently employed to face this problem is working on the compactness of the MZI cell, namely on $L_{\mathrm{MZI}}$, both in terms of optical and electrical circuit. On the other hand, an orthogonal strategy relies instead on leveraging the unique 3D capabilities of the FLWW platform in order to completely change the waveguide layout of the photonic processor and, thus, achieve a more favorable scaling of the circuit length L as a function of the number $N$ of optical modes. In particular, two main limits of the square mesh have been identified and addressed: first, the fact that the interference happens only in well-defined positions of the circuits (i.e. the directional couplers) and, secondly, the fact that the circuit (and thus the optical interaction) extends over only one plane. The novel layout proposed by Hoch et al. goes exactly in this direction by taking advantage of a 3D continuously coupled FLWW waveguide array (fig. 3e). First, thanks to the continuous optical coupling, the interference between the optical modes is distributed over the whole length of the circuit, with no dead space. Secondly, the 3D arrangement guarantees the most efficient interaction among the modes and, in particular, the triangular lattice (fig. 3e) chosen for the cross section of the array results in the best possible packing efficiency. All these features give a key contribution on the maximization of the optical interaction per unit length and, thanks to them, the scaling rule reported by eq. (3) now reads:

(4) $ L = C \sqrt{N} $,

where $C$ is a coefficient that depends essentially on the strength of the optical coupling within the array. Programmability can be introduced relying again on thermo-optic phase shifters fabricated on the surface of the device and, currently, a 32-mode programmable processor featuring 18 microheaters (fig. 3e) has been already demonstrated and exploited in the applications thanks to its large reconfigurability and low losses (3.5 dB). However, further work is needed to fully exploit this powerful approach. Just to make an example, a control algorithm that allows the end user to program the processor and implement an arbitrary unitary transformation is still missing. In particular, this point has been successfully addressed for discretely coupled MZI meshes, but for the continuously coupled arrays a black-box approach based on machine learning currently represents the most promising way to demonstrate full control of these promising devices.

4 Applications

FLWW has very diversified applications. It can provide integrated optical sensing in lab-on-a-chip for biophotonic analysis or photonic interconnects for telecom applications, where planar lightwave circuits need to interface with multicore/multimode fibers. Here, we will focus on two recent and very promising fields where FLWW can play a major role thanks to its unique properties: quantum technologies and astrophotonics.

Quantum technologies are nowadays promising to revolutionize the way we acquire, manipulate, and transmit the information. Although the development of quantum technologies has been driven by different applications, the development of a large-scale quantum computation platform is considered by all the Holy Grail of the field. The large effort concentrated around this goal is motivated by the promise of an algorithmic advantage in problems that are today considered intractable by classical (super)computers like the simulation of certain chemical reactions for drug design and discovery. Despite the humongous advances reported in the last years, the achievement of such a revolutionary milestone still looks far from our reach and it is very likely that this goal will not be attained even for the next ten years. As a result, the research in the field moved its attention on an intermediate, yet fundamental, step like the achievement of the regime of quantum advantage, i.e. the experimental proof of a quantum device able to perform a specific computational task that is currently not possible to address with classical computers in a reasonable time. In this scenario, Aaronson and Arkhipov proposed in 2013 a computational problem that consists of sampling the probability distribution of $n$ identical bosons scattered through a linear interferometer featuring $N$ input/output modes. Such a problem is called boson sampling. Under the absence of any special structure for the interferometer (i.e. the unitary transformation must be chosen “randomly”), boson sampling is strongly believed to be an intractable problem for classical computers.

Although this problem regards in general any bosonic particle, since the beginning the photonic implementation of boson sampling (fig. 4a) was considered the most promising approach for achieving the level of complexity required to challenge and eventually outperform a classical computer. Indeed, although not universal, a photonic boson sampling setup can be implemented by using far fewer physical resources than a universal quantum computing platform. This made it the ideal candidate for demonstrating the quantum advantage in the near term. In this framework, FLWW photonic circuits have been pivotal in attaining the first experimental demonstrations of boson sampling. Today, the quantum advantage regime has been achieved in different works and some of them have taken advantage of photonic boson sampling setups.

Although integrated photonic processors represent a natural way to realize an $N$-mode interferometer, an integrated implementation of boson sampling capable of reaching the quantum advantage is still missing. The reason is that, in order to effectively implement a large-scale experiment, it is not only necessary to increase the number of modes of the device, but it is also important to keep the photon losses as low as possible, otherwise the complexity of the process could be jeopardized. However, as we have already mentioned in sect. 3, scaling the number of modes and keeping low losses at the same time is not trivial in an integrated processor. An important step towards the solution of this issue was the demonstration of the 3D photonic processor fabricated by FLWW and already presented in sect. 3 (fig. 3e). For the first time, an integrated processor achieved a number of modes as high as 32, along with total photon losses as low as 3.5 dB. Moreover, a set of 18 thermal phase shifters enabled a high degree of reconfigurability, which is important to evaluate the effective randomness of the platform. Although universality was not fully demonstrated, with this device we successfully showed the implementation of a large set of random transformations and, then, we implemented and validated 3- and 4-photon boson sampling experiments, demonstrating the feasibility of using such a processor for future large-scale computational systems.

The second example of application field where FLWW is having and will have an important role is astrophotonics. The field of astrophotonics promises to improve the observation of celestial objects and phenomena by interfacing integrated photonic circuits with telescopes. Coupling the collected light into single-mode waveguides provides spatial filtering that, combined with the stable and controlled interaction between different waveguides, can produce interference effects with greater visibility and easier scalability than bulk optical interferometers. This results in an enhancement of the angular resolution in imaging astronomical objects. To this aim, FLWW in glass has several specific advantages. First, it is rather straightforward to optimize the writing process in order to have single-mode waveguides at any wavelength in the transparency window of the material, with the possibility of performing observations in the whole visible and near-infrared range with very good performances in terms of insertion losses and reproducibility. Second, laser written waveguides show very low birefringence, thus all devices are polarization transparent, i.e. their optical performance does not depend on the polarization state of the input light. This allows one to maximize the available signal by analyzing all the faint celestial light, since no polarization filtering at the device input is required. Finally, the 3D capabilities of FLWW are very important for stable and compact pupil remapping (i.e. reorganizing the signals collected in different points of the telescope pupil to arrange them in different ways) and efficient beam combining.

A technique benefitting from these features is aperture masking interferometry. In this framework, the light imaged on the telescope pupil plane is sampled, by means of a mask or a segmented mirror, in some spare points, whose interference pattern is then analyzed for reconstructing the original image. This technique provides enhanced resolution and robustness to atmospheric aberrations if compared to classical imaging, and for this reason it is widely used when studying far astronomical objects. Integrated optical waveguides can be used to sample and remap the signal, routing it to a multimode integrated interferometer. Such an interferometer is typically complex and difficult to scale for a large number of sampling points. For this reason, a new type of device, named discrete beam combiner (DBC), has been proposed and realized as a scalable interferometer. This intrinsically 3D element is based on a continuously-coupled waveguide array arranged in a triangular lattice. The fabrication of the DBC has been accomplished with FLWW in borosilicate glass for operation at 1550 nm. The device has been tested on-sky at the William Herschel Telescope in 2019. The reported circuit (fig. 4b) was composed of a 4-input DBC with 23 waveguides, followed by a fan-out region reformatting the output modes in a linear configuration for further spectral dispersion and analysis. The interferometer was preceded by a length-matched pupil remapper, necessary to route the 4 selected sub-apertures of the telescope pupil into the proper interferometer inputs. Notably, the device has shown a polarization-insensitive behavior, thus enabling the analysis of all the collected light, without polarization filtering.

Another relevant application of the DBC component is represented by the interferometric combination of the light beams collected by different telescopes, with the purpose of increasing the angular resolution of the observation creating a synthetic aperture that equals the distance of the telescopes and not the actual aperture of each of them. To perform this task, complex interferometers are typically used, based on the pairwise interaction of all the beams. However, this is hardly scalable to a large number of telescopes. A DBC produced by FLWW has been characterized in the laboratory, showing its capability to simultaneously combine 6 different signals in the J band (around 1300 nm), potentially coming from 6 different telescopes.

5 Conclusion and perspectives

In the last forty years, thanks to the development of important photonic technologies like the laser or the optical fiber, we have witnessed a true revolution in fields as diverse as medicine, industrial material processing and telecommunications. Now, it is time to move the implementation of all the photonic components we know to a fully integrated platform in order to unlock the true potential of many groundbreaking applications. However, it is not clear at the moment if, similarly to what happened for the electronic circuits and the CMOS technology, we will have a winning integrated platform also for the photonic circuits. As a matter of fact, different technologies are currently establishing in the field for the implementation of specialized tasks and, as a result, hybrid integrated platforms are now gaining a lot of attention from those applications requiring a thorough optimization of the entire photonic system, i.e. from the light source to the optical circuit and, eventually, to the detection. Our vision is that FLWW will play a paramount role in such a scenario thanks to the possibility of interfacing different photonic components in a very efficient fashion (i.e. with very low photon losses) and thanks to its flexibility in adapting photonic circuits to different situations. FLWW optical interconnects are already an industrial reality and now other applications like quantum and astrophotonics are blossoming and establishing thanks also to the high value of this fabrication platform. Throughout this article, we have reported examples of FLWW devices with unique features that have been already exploited with success in the aforementioned applications. However, we believe that the number of applications in which FLWW devices can be key is much larger and, as a natural consequence, we envision that in the next years this technology will finally prove itself as one of the most prominent integrated platforms of the entire photonic landscape.