 It's already been said, I'm Benny Brecht from Paderborn University in Germany. I'm not a professor, I have to say, I'm a doctor. I'm working in Christine's Alba Horn's group, so she's the PI. And we're speaking today about modern technologies for quantum photonics. We have two very different lectures. The first one is very heavy, really, on technology, on device fabrication. How do we build devices that we can use for quantum photonics? The second lecture is more about underlying concepts, like how are we thinking, how are we using our degrees of freedom? And well, you've already seen the second slide. For those of you who don't know where Paderborn is, most Germans actually don't. So it's not a problem not to know. Paderborn is a small town in North Rhine-Westphalia, roughly in the center of Germany. It's a very historic town with an old cathedral. We have this historic town hall. As you can see from the pictures, we always have good weather, which is one of the perks of being in Paderborn. I'm going to show you what's about our group. The integrated quantum optics group in Paderborn largely separates into three subgroups. The first subgroup is the technology subgroup. This is our clean room. This is where we fabricate our devices and the group is led by Christopher Eigner. We're mainly working with lithium niobate and potassium titanol phosphate, so KTP. Since recently, we're also working with lithium niobate and insulator, which is a very promising new material, which I will say a few words on towards the end of the lecture. Here we're really doing everything in-house. We're buying lithium niobate wafers. We're dicing them into chips. We're indiffusing waveguides. We're doing periodic polling. We apply dielectric and facet coatings. We sputter on electrodes. We do bonding for electro-optic modulators, so everything is done in-house. The second subgroup is the devices subgroup, which is led by Harald Herrmann. And here we try to take all the capabilities from the clean room and turn them into multifunctional devices. We're looking at optical circuits. We're looking at modeling of our devices, characterizing them. Some key devices are electro-optic modulators, which I will come through later in the lecture. We're also doing very rudimentary amount of packaging and fiberpick tailing to really get our devices to a point where they are hands-off in operation, and we can just have them sitting on the table and doing the thing. The last subgroup is Quantum Networks Group, which I'm leading. Here we're looking into ultra-fast quantum optics and quantum walks. The idea here is really to look at fundamental quantum optics, quantum applications in the fields of quantum metrology, quantum communications, quantum simulations. We're very much working with techniques of time multiplexing. We're looking into topological effects, and that will be part of the next lecture. We're looking into temporal pulse modes and quantum pulse skating. Now, the outline of this lecture is following, I'll first speak a bit about underlying concepts. Why do we want to do integrated quantum photonics in the first place? Why don't we just do it in bulk optics? I'll then introduce the device toolbox you need to have at your disposal if you want to build integrated quantum optical circuits. There's a number of devices that you need to have to be able to implement all functionalities that you may want. I'll then speak about our fabrication technology. How do we fabricate our devices? And I'll finish with an example of our home on a chip circuit, which you can already see in the background of the slide. So why do we want to do integrated quantum photonics? These are two pictures. The left one is from Akira Furusawa's group in Tokyo. It's a picture of the continuous variable quantum teleportation experiment they set up. The right one is from a recent paper from Jan Waipan's group, which was on 20 photon boson sampling in a 60-mote interferometer. These experiments show incredibly nice results. And they're really beautiful. But it's my very personal belief, and it's not a general opinion, but it's my very personal belief that this way of building an experiment is not scalable. Which you can see here, a rate around the center block are 60 single-mode fiber couplings. And you can set them up. You can build 60 fiber couplings. If you're very dedicated, you can probably set up 100 fiber couplings, no problem. Now, if you're thinking, for instance, of a quantum computer, which you want to build with optics, you might need 1,000 or 5,000 or maybe 10,000 fiber couplings. And this is something you won't be doing in that architecture anymore. I don't think there is a way that you can actually do that. So one solution to that could be integrated quantum optics. And here is a paper from the late 60s from Stuart Miller. He was essentially the inventor of integrated optics back then. Because during the 60s, classical optics was facing a very similar problem. How do we build setups that we can actually scale in size and that we can make robust against environmental influence? Let me just read the abstract. This paper outlines a proposal for miniature form of laser beam circuitry. Index of refraction changes of the order of 10 to the minus 2 or 10 to the minus 3 in a substrate such as glass allow guided laser beams of widths near 10 microns. Photolithographic techniques may permit simultaneous construction of complex circuit patterns. This paper also indicates possible miniature forms for laser modulator and hybrids. If realized, this new art would facilitate isolating the laser circuit assembly from thermal, mechanical, and acoustic ambient changes through small overall size. Economy should ultimately result. And to date, this paper, in a sense, is very timely again. Because it almost reads like a roadmap for going from conventional classical integrated optics to integrated quantum optics. By just changing a few words, you can make that a paper that you could publish today without any problem. So you just check in quantum every year in there. Whenever it's about beam circuitries, you call them quantum circuitries. And at one point, you exchange laser for photon source. Then this abstract reads, this paper outlines a proposal for miniature form of quantum circuitry. Index of reflection changes of the order of 10 to the minus 2, 10 to the minus 3 in a substrate such as glass allow guided laser beams of width near 10 microns. Photolithographic techniques may permit simultaneous construction of complex circuit patterns. This paper also indicates possible miniature forms for photon source, modulator, and hybrids. If realized, this new art would facilitate isolating the quantum circuit assembly from thermal, mechanical, and acoustic ambient changes through a small overall size. Economy should ultimately result. That very much reads like the quantum technology roadmap of the European Union, in a sense. Now, the idea, at least our idea, our approach to integrated quantum optics is not to reinvent the wheel, but to keep established ideas, but adapt them to quantum optics. Integrated optics is doing a great job. All the devices are working, so we don't have to think about reinventing all of them. We just have to see which of those we have to exchange in order to build quantum integrated optics instead of classical integrated optics. If you think about a quantum optical experiment that you would want to integrate, you will find that most of them have a very common structure. You have to create photons at some point. You will have a circuit that manipulates your photons, and at the end of the day, you have to detect your photons. One objective of integrated quantum optics would be to actually integrate as many of those components as possible, maybe on one platform, if you can. If not, you have to use a hybridized platform made of different materials. But ideally, you want to stick to one platform as long as you can. So you want to make that overlap region in the center as big as possible for your choice of materials. And this is also not a terribly new idea. This is a picture from a paper from 2012 by Alanis Borogussi and Philip Walter, where they looked at a photonic quantum simulator. You will find that this also consists of a source part where you generate photons, a circuit part where you manipulate them, and a detection part. The circuit part is always assumed to be linear, as in linear optical quantum computation. The source part necessarily needs to have some sort of non-linearity to generate your photons. The benefit about that is that if you can build this in an integrated way, you can control many different spatial modes, because you don't have to align them. They are inscribed in your material. You have inherent interferometric stability, so you don't need face-locking techniques. You can build comparably compact devices. You have, I would say, a simple operation. Well, once you know how to actually operate your waveguides, it becomes simple. It's a bit of an uphill struggle to get to that point. But definitely once that thing is sitting on your table, you don't have to realign it anymore. It's just sitting there and doing a thing. And you can have high efficiency in these architectures. There's a small question mark because we always have to battle losses whenever we're propagating through our material. There will be some sort of loss, which we don't have in free space propagation. But we can engineer that. We can try to minimize that. And then we can build highly efficient devices. And the linear circuit part has been looked at. And I'm just flashing a few examples. It's not an exhaustive list. There's a lot of work from groups in Bristol who are looking on a silicon and insulator platform for integrated quantum photonics. There has been work in Rome from Paul of Mataloni's group, which they were looking at quantum walks in an integrated structure that was a laser written structure from Robato O'Salam's group. There has been work in Oxford and by other groups looking at boson sampling and integrated structures, again in laser written waveguides. However, the non-linear source part, something that's not as easy to do. I'll come to the big why in a few slides, but it has to do with the cartooning linearity which you need for some elements that you might want to use. Because of that, we chose to work with lithium niobate. Lithium niobate allows for the fabrication of low-loss waveguides. You can build integrated devices on lithium niobate. It has a K2 non-linearity, which you can use for photon pair generation and electro-optic modulation. So with the electro-optic modulation, you can build fast switches. So conceivably, you can do feed-forward and feedback operations in your circuit. You can build all sorts of devices on lithium niobate. You can build photon pair generators. You can do polarization converters, phase shifters, frequency converters. You can also include superconducting detectors directly on your substrate. Then you have to operate your substrate at cryogenic temperatures, but you don't have to couple to fibers anymore to go to your detectors. So we think that we can implement complex quantum circuits in lithium niobate, where for us, the most complex thing we've done so far is the Hormone chip sample that I will show you later. Now, what device toolbox do you need if you want to build integrated quantum photonic networks? First of all, you need some sort of passive routing. You may need power splitters. You may need polarization splitters, wavelength multiplexers and demultiplexers, just things to guide your light through your circuit as you would do in bulk optics. Then you need the generation and storage part. You need to be able to generate photons, probably change the frequencies and bandwidths and conversion processes. And ultimately, if you want to build large circuits, you want to have quantum memories as well for scalability. You need some active manipulation. If you want to reconfigurable circuit, you want to be able to do that fast. So what you need for that are, for instance, phase shifters to control the phases of interferometers, polarization converters that may act as half and quarter wave plates, and spatial switches that act as tunable beam splitters, which you can basically tune from zero to 100% reflectivity by applying a voltage. Finally, you may need interfacing. You won't be able to do everything on a single substrate, either because your substrate doesn't provide the functionality or because it's not big enough, or because you have external detectors, or because you want to send photons to a remote location. So at some point, you will have to couple your photons to fibers. To do that, you need efficient fiber wave cut coupling and ideally dielectric coatings to reduce reflections at interfaces between different devices. Every of those components has already been built for classical photonics. So we're not reinventing the wheel. They also have been optimized for classical photonics. What does that mean? That means that if you're talking to classical telecoms people, they will tell you that losses up to 70 dB are perfectly acceptable because you can just re-amplify your light. Check in an amplifier somewhere in your transmission line and your golden. That doesn't work with quantum optics, unfortunately, because if you lose a photon, it's gone. You can't amplify it. So that means that you do have to re-optimize these devices for quantum applications in the sense that you really have to work on reducing the losses. 3 dB insertion loss for a spatial switch is not an option for a quantum integrated network. You can't afford to lose 50% of your photons at every beam splitter. But other than that, the devices exist and people know how they work. Now, let's go through them and see how they work and what you can do with them. And we'll start with the passive routing which you do via so-called directional couplers. Directional couplers are devices with two waveguides. You bring them closely together such that the light can couple from one waveguide to the other. They are together in a central section with a certain coupling length and then you separate them again. And depending on how you design that device, you can build just simple power splitters. You can build wavelength division multiplexers. You can build polarization splitters. You can build any sort of interference device. So it can act as any sort of beam splitter that you would use otherwise in bulk optics. How does it work? Well, as long as the two waveguides are separated and there is no coupling between the waveguide, so your coupling constant is zero, you can describe the light in the waveguide mode description. So you're looking at either modes in waveguide one or modes in waveguide two. If your waveguides are identical, they will just have the same propagation constant. So they're just traveling along your waveguide, not talking to each other. You bring your waveguides closely together such that you have a coupling. You can't use the waveguide mode description anymore. You have to use an eigen mode description where you have a symmetric eigen mode that basically extends over both waveguides. You have an asymmetric eigen mode. Propagation constant of those will be modified by your coupling constant leading to an interference between these two modes. You get a beating and superposition of these two modes. If you just add them up, you will end up with light in waveguide two and no light in waveguide one. If you subtract them, you will end up with light in waveguide one and no light in waveguide two. So as these modes propagate along your coupling region, you will have beating between them and you will couple back and forth from one waveguide to the other. You then, at the right point, separate the waveguides again. You have managed a couple light from one to the other. Here's an example of a wavelength division multiplexer where the aim is to separate light at 1,320 nanometers from light at around 890 nanometers. Now, here you make use of the fact that different wavelengths have a different mode area. 890 nanometer light will be confined more strongly than 1,300 nanometer light. So while 1,300 nanometer light can couple between the waveguides, there is no coupling for the 890 nanometer light. This means that if you look at the output power in the cross-state, meaning you're sending light in here, you're looking at the output down there, you see that there is no coupling for the 890 nanometer, regardless of the coupling length, but the 1,320 light follows the sinusoidal pattern and somewhere around nine millimeter coupling length, you have complete coupling of the 1,320 to the lower channel, whereas the 890 is still in the upper channel. And this is where you want to operate your device. Looking at a polarization splitter, they are slightly different in geometry. They are so-called zero-gap couplers where the two waveguides will actually touch each other in the center region. And these are couplers that you operate in a so-called low coupling order. This means that the light will not bounce back and forth between the waveguides many times upon propagation through coupling length, but only once or twice. This means on the one hand that you do have to hit your coupling length very precisely. Ideally, you want to be exactly at the trough here as soon as you move out of that little bit, you basically lose in splitting ratio. On the other hand, you get a very robust device as in it's comparably broadband in wavelength regime. And one thing you can see here is that for a coupling length of, in that case, half a millimeter, that's a very short coupler, you get roughly 20 dB of splitting, which is sufficient for most applications that you want to do. And we typically have excess losses in these devices of below half a dB. Now, let's turn to the active modulators where we use electro-optic modulation. And for that, I quickly have to introduce, first of all, tell you that you do need a K2 non-linearity for that. You can't build electro-optic modulators, for instance, in glass, so you do need the K2. Second, I have to say a few words about non-linear polarization of a medium. The dielectric polarization of a medium is induced via an electric field that's coupled to the non-linear tensor of your medium, or dielectric tensor of your medium. And as long as the response to your electric field is purely linear, as in that case here, you're speaking of a linear medium. This would be just your refractive index. As soon as this response becomes non-linear, which could happen, for instance, by increasing the driving strength of the field, you might see non-linear terms, or just by having a second order non-linear response from your medium directly, you're speaking of a non-linear polarization. You then typically do is you do a tailor expansion of your polarization in terms of orders of non-linearity. So you have the linear term, which is your refractive index. You have your second order non-linearity for which you need a non-centrosymmetric crystal, and which is used, for instance, for electro-optic modulation. The third order non-linearity, which would be the cause for four-wave mixing, or self-phase modulation, cross-phase modulation of pulses. And electro-optic modulation is the interaction between optical and electric fields. So one of these fields would be an optical field, the other one an electrical field. Works in the following way. On the upper right, you see a picture of a sample. You have a wave guide with an electrode on top, and another electrode sitting next to it. And you apply a voltage between the two electrodes. This will lead to an electric field through your wave guide. If your wave guide is very small, that field will primarily point downwards into your sample. If you choose your crystal orientation appropriately, you can make use of what's shown here, the R33 component. That is the component of the electro-optic tensor, which in Lithium Niobate is strongest. There are different components, R33 is strongest, so you have the strongest interaction between the electric field and the light field. What this electric field now does for you is it induces a change of the refractive index in your wave guide. Because your wave guide has a finite area, you have to average over the wave guide cross-section to do via this integral term. Then you get an effective index change that is dependent on the applied voltage. Because wave guides are very small, you only need very low voltages, so you can achieve very high driving speeds. You can achieve modulations in the order of up to 125 gigahertz in wave guide modulators, which is something you couldn't do in bulk cells, they're just too large. The most common sort of modulator is a Maxinda modulator. So it's a Maxinda interferometer, which you use for intensity modulation. And in that case, what you want to have is a very sophisticated arrangement of electrodes. There's three electrodes in total, two on top of two wave guides, one to the side, they have to be impedance matched, they have to be velocity matched, so it's really a feat of engineering building these things. But if you have one, you can operate them for instance at 40 gigahertz, which would be the standard telecom speed. Good thing about Maxinda modulators is you can just buy them. They are standard telecoms components. If you're working in telecom wavelength, you can go to any telecoms company and they will sell your Maxinda modulators. They typically look like that, they are fiber pigtails, they have input and output fibers. You typically have one plug for your radio signal, which you use to basically quickly modulate the light. You have a few more pins to apply some bias voltage to choose the operation point. For instance. However, again, with these modulators, you're facing a problem of insertion loss. Typically, they would have something like 1 dB if you're careful with your fiber splices, meaning that you have 10% loss at a modulator. So you have 90% transmission, that doesn't sound too bad. Assume you want to build 100 by 100 network of modulators. 0.9 to the power of 100 times 100 is an incredibly small number. So you wouldn't want to use these modulators for large quantum applications. Look in a polarization converters. We have to add one more thing which we hadn't before and that's that gray structure inside the waveguide here. This is a periodic polling and you need that to achieve face matching between the two polarization modes. The reason for that is that lithium niobate is a birefringent material, meaning the two polarizations experience different refractive indices. So they will have different wave vectors inside your waveguide. You want to convert from one polarization to the other. You have somehow to find a mechanism that allows you to match the two wave vectors. And you do that by introducing an additional momentum component via this periodic polling, by which you just flip the sign of the non-linearity of the crystal periodically. And the polling period, which is called capital lambda here, will define this additional component in your face matching. You can achieve what's called quasi face matching. You can build polarization converters. I won't be going into much more detail than that on that. Now, here are curves from a device we built in Paderborn. This is the response as a function of wavelength. And you can see we have an optimal conversion for something like 1587 nanometers in that specific device. Or with a bandwidth of about two nanometers, which one integrated device isn't too bad. Two nanometers actually means that you can use it for pulsed light. You don't have to be CW. You don't have to hit that wavelength on spot. You can work with a reasonable bandwidth. At the same time, we use the conversion efficiency between TE and TM polarization as a function of the applied voltage. We can see we can basically switch from one polarization to the other. That would be a halfway plate rotation with a voltage of about 10 volts. So that's something that's still doable. And you can still switch it reasonably fast. Now, moving on to the next part, generation and storage. I will be very brief on that one because the next lecture focuses very much in parametric down conversion. So I don't want to double all of what I'm saying. First of all, again, we're looking at non-linear optics in that case. Only now we're looking at the interaction of optical and optical fields. So where before we applied an electric field, now we have two optical fields coupling to the second order non-linearity of our material. The processes you will most often encounter in quantum photonics are second harmonic generation or frequency doubling, by which you double the frequency mainly of laser light. You would, for instance, do that if you were working in continuous variables. You needed your laser to actually also serve as a local oscillator for measuring quantum states. The next one is parametric down conversion, photon pair generation, where a strong pump induces the generation of signal and idler fields at lower frequencies. On the bottom two, sum and difference frequency generation, essentially a processes to use a strong pump field to translate the photon, the frequency of a weak field to either the sum of the frequencies or the difference of the frequencies. You could use that, for instance, if you're thinking about, you have quantum dot emitting photons at around 900 nanometers. You want to convert them to 1550 for telecommunications applications, sending them through fibers. You would use these kind of interfaces. You're quick words on parametric down conversion. In our case, guided wave parametric down conversion, we have a strong pump laser just propagating through our wave guide, which again has this periodic polling structure. Every once in a while, it might happen that a photon from the pump laser decays spontaneously into a pair of photons, then labeled signal and idler. Process has to obey energy conservation, meaning the frequency of signal and idler have to add up to the pump frequency. Process has to obey momentum conservation or face matching, meaning that the propagation constants of signal and idler plus this additional contribution from the periodic polling have to add up to the pump wavelength. This helps you to choose which signal and idler wavelength of frequency matched so it helps you to tune the wavelength of your generated photons. We're propagating through a wave guide, so we naturally have a strong field confinement, strong as opposed to focusing with a lens into a bulk crystal. And we can basically propagate with a focused spot of a long propagation distances. This leads to a high efficiency. Typically, guide wave parametric down conversion is three to four orders of magnitude more efficient than bulk parametric down conversion. It's not necessarily a cleaner system. You do not necessarily get better quantum states, but you get many more of them, which in some cases can be very helpful. With lithium niobate, you can go up to propagation length of about 90 millimeters. This is limited by the available size of lithium niobate wafers, which is four inches. So you can buy them in discs of about that size and it's really hard to get larger material in a good enough quality for quantum applications. First guided wave parametric down conversion was demonstrated in Paderborn in the group of Wolfgang Sohler back in 1986. This was in lithium niobate and they use birefringent phase matching, so no periodic polling, but using the birefringent of the crystal. And that's the spectrum, the mashat where you can see the peak, the parametric down conversion peak at around 680 nanometers. The first quasi phase matched parametric down conversion in a periodically pulled wave guide was then demonstrated 93 in Nice. And the University of Nice is still a very strong center for integrated photonics, lithium niobate, wave guides and everything you can do with them. Frequency conversion, I already briefly touched upon that. Assume you have a solid state system, a matter qubit. This could be an iron, a trapped atom. This can be a quantum dot. This can be a color center and diamond. Pick your favorite. All of them typically will have wavelengths somewhere around the UV to visible region where attenuation and optical fibers is prohibitively large. So you wouldn't want to send a UV photon through a fiber, it just doesn't make any sense. You would want to send your photon somewhere in the telecoms region. So either in the O-band around 1300 or the C-band around 1500 where the fiber losses are lowest. Mainly you would target the C-band because that's the telecommunications band where classical telecommunications operates. However, for quantum communications the O-band might also be interesting because you don't have to worry about all the bright classical signals going back and forth through your network. Basically it's 1300 nanometer band is empty so you don't have noise. Anyway, to go from that to that you need some sort of frequency conversion device. And this is just to give you a very brief overview of what people have been looking at in terms of quantum frequency conversion. It's essentially converting single photon level light from one wavelength to the other. Most of the works have focused on photons somewhere in the visible range. These can be from quantum dots, these can be from nitrogen vacancy centers in diamond, these types of systems. There was one demonstration in Southampton and Oxford where people tried converting light from 422 nanometers to 1550. Basically the idea was that that would be an interface to link strontium ions to the telecommunications band and strontium ions are interesting candidates for ion trap quantum computing. And in our group we are running two of these interfaces. We have one that is highly engineered. I will touch upon that in the next lecture which is converting green to the telecom. And we've built another one which is actually going down to 369 nanometers that's a calcium transition which we're also linked in that case to the telecom O band. I have to say that 369 conversion was done in KTP not in lithium niobate. Lithium niobate starts to absorb as hell somewhere below 400. So you can't use lithium niobate at 370 nanometers. Good, but that become to the last part of the toolbox and that's interfacing. And I'll start with the dielectric and facet coatings. Now just give you an example. The interesting curve here is the red curve. The green curve is the transmission characteristics of a coating which we measure on a sample substrate made of glass. Glass has a different reflective index than lithium niobate which you have to correct for. So the red curve is the response of the coating on lithium niobate. The setting we're looking at is parametric down conversion. You use strong light at 800 nanometers. You coupled it to a wave guide. You generate photons at around 1600 nanometers. And at the end of the day you want the photons at the output but not your strong pump light. At the same time you want the strong pump light to enter your wave guide efficiently. Lithium niobate has a high refractive index meaning you have an interface loss of about 16% for 800 nanometers. You don't want to lose 16% of your pump light just because you didn't apply a dielectric coating. So what you can do is you apply coatings just as you would for instance for mirrors on the end facets and in that case the input facet has a coating that has a transmission of around 96% at 800 nanometers. We don't care about the longer wavelength for that coating. However at the output we want to get rid of our pump light but still transmit the photons efficiently. So you need a multi-band coating which has a low transmission for the pump light in that case transmission for the pump light is below 1% but a high transmission for the generated photons which again is in the order of 95%. Being able to use these coatings actually helps you because you can get rid of a lot of additional filtering in your setups. Every component you can get rid of essentially means you have lower losses again meaning you can add more components to your system meaning you can build bigger networks and that's what it's all about at the end of the day. Then wave guide to fiber coupling. That's a bit of a tricky thing because if you look at the spatial mode of a single mode fiber that's a nice round blob. Now lithium niobate waveguides at least the titanium-indiffused waveguides we are using are waveguides that are sitting exactly at the surface of the sample. So at the top of your waveguide you have a refractive index jump from lithium niobate about 2.2 to air which is 1.0. That means that the modes coming out of a lithium niobate waveguide are slightly squeezed along this upper line here. That's basically the edge of the sample. However, if you do it correctly you still get overlaps of about 93% without working too hard. So you optimize your fabrication parameters. You do your fiber coupling. You have less than 1 dB of losses to fiber interface. You can increase these efficiencies by further modifying your sample. For instance, you can add a cover layer on top of your waveguide which has a refractive index that's lower than lithium niobate but bigger than air. That would lead to slight symmetrization of your mode and that would increase these coupling efficiencies. Again, we're looking at a problem here where we're saying, well, we already have 93%. Why do we actually care? Well, we do care because if you can't go above 99% we don't have to bother with trying to build big circuits where we have for instance 100 photons. Because 0.93 to the power of 100 is again a very small number. That's just a picture of the pigtailing setup. This is a sample. This is a fiber ferrule. This blue light are UV LEDs which we use to cure the glue with which we glue the ferrule to the sample. So you basically try to optimize the coupling by monitoring the power. Once you've found the ideal spot you apply your glue, you cure the glue and you hope that the glue actually cure such that the fiber isn't moving anymore. All right, waveguide fabrication, basically our fabrication technology. First of all, we fabricate waveguides. We do that by depositing a thin layer of titanium onto a lithium-nibre substrates and then coating that with a photoresist. We then do mask photolithography to pattern the photoresist such that only strips are remaining. The next step is chemical etching of the titanium and removal of the photoresist which means that the titanium only survives underneath the photoresist so you have titanium strips running along your sample. Now you do indiffusion of the titanium. The titanium indiffuses into the lithium-nibre. It increases the refractive index and you have a guiding structure. Indiffusion happens at around 1060 degrees for roughly eight hours for telecoms waveguides and around 30 hours for MIR waveguides. We can do both with roughly these parameters. Your waveguides will be single mode so guide only one spatial mode. We typically do the diffusion in a platinum box. This is to avoid out diffusion of titanium and also to avoid indiffusion of other particles. The purer the titanium, the lower the waveguide losses. After fabrication, we do a mode characterization where we couple light from an EDF-A that's just amplified spontaneous emission. You want incoherent light for that. Into a waveguide sample, we monitor it with a CCD camera and this is again a comparison between a fiber mode and a typical waveguide mode at telecoms wavelength. After that, we do a loss measurement. Our samples have an infested reflectivity of about 14% at telecoms. You can think of them as a Fabry-Perot resonator. You couple CW light to your resonator. Now you slowly change the resonator length. You do that by heating your sample. This will lead to intensity fringes at the output of your sample. If you know the reflectivity of your mirrors very well, the contrast of the fringes will tell you something about the losses inside your resonator. So you learn about the propagation losses in your waveguide in an undistructive way. You don't have to do cutback measurements. We take a long sample, measure the coupling, cut it into short pieces, measure the coupling again. That's a typical trace that you see. From that, you can then basically calculate the losses and we're generally below 0.1 dB per centimeter. We can do better if we work really hard. Periodic polling. Few words about physics behind that. This crystalline structure of lithium niobate, the red bloods are the oxygen layers and the green and gray blobs are the niobium and lithium ions. And you can see that there is always one empty layer and then a mixture of lithium and niobium. That gives you spontaneous polarization. They too form an electric dipole. Now, periodic polling means that you apply an external electric field above the coercive field strength of lithium niobate. That's 21 kilovolts per millimeter. Our samples are half a millimeter thick, so we have to apply fields of around 11 kilovolts. What that will do is shown in the following animation. I hope, yeah. If you apply the electric field, the niobium ions will stay in the layer, but the lithium ions will be squeezed to the adjacent layer of oxygen. So you flip the direction of the spontaneous polarization of your material. This flips the direction of your non-linearity. This in turn gives you the periodic polling. If you turn off the electric field, that polling is permanent, so you don't have to worry about domains flipping back. Again, we do periodic polling on the waveguides by starting with a layer of photoresist. We do photolithographic patterning. Now we carve out small holes from our waveguide. We apply the high voltage with liquid electrodes. Upon application of high voltage pulses, these domains start to grow, and if you stop at the right point, you get an exact 50-50 duty cycle of your periodic polling. This is an example of a periodically-pulled waveguide. That one has a polling period of four and a half micrometers. When we started in Paderborn roughly 10 years ago, state of the art was something like seven micrometers, and people were telling us you can't pull waveguides with smaller periods. Today, we can do sub-2 micron polling periods. That doesn't work on every sample, but we can do it. Four and a half microns we have under control. We can do that reliably. That specific waveguide has propagation loss of 0.02 dB per centimeter, so you're not losing a lot of light. We're coupling to single-mode fibers of more than 85%, and we're characterizing our samples in every way possible. We're doing linear characterization of losses of spatial modes. We're doing nonlinear characterizations for face matching, so we're looking at second harmonic generation, some frequency generation, parametric down conversion. We do that for every waveguide on every sample, just to give feedback to our technology. So if you were to collaborate with us and you would get a sample from us, you would know exactly what you were getting for every waveguide on that sample. Drowned by a waveguide sample from commercial vendors and asked them for characterization, it's a bit of an uphill struggle to get any information. Now, dielectric coatings, we use iron-assisted deposition, whereby target material is deposited onto the samples. The iron-assisted deposition technique helps to form high-density coatings, which means they don't degrade over time due to instance for instance due to water absorption, so the coatings just stay there as they are. This is a picture of our coating machine and our coating technicians. We have two full-time clean room technicians. One is the expert on periodic polling, the other one is the expert on dielectric coatings. I've been told by our clean room guys, it's more than just knowing the technique, you have to have the feeling for it, which I obviously don't have, so I'm staying with my quantum networks. Anyway, you can build mirrors from dielectric coatings. This is just an example of the reflectivity of a quarter-wave stack, so each layer is a quarter-wave length of the target that you want to achieve. You can use different materials, we typically use titanium dioxide and silicon dioxide. So another example again of a three-band coating now, we are low-transmissive at 532 nanometers with high-transmission bands at 81550. That was again a parametric down-conversion source where we pumped with green light and generated photons at 81550. This is the coating machine, you have a collot into which you mount your samples, you have the two materials, you have your iron source, your electron guns, and then materials are deposited on the rotating collot. Right, with that, in the last few minutes, let me show you, for this lecture, let me show you the Hormone chip source. And first of all, some quantum devices that we've built. We've built a fully integrated plug-and-play single-photon source. It's fully fiber-coupled, you just plug in your laser on one end, you get out your photons at two fibers at the other end. And that source has been running for three years without any degradation of performance. We've built an integrated noon source where you pump with a laser pulse, you have a periodically pulled directional coupler at the output, you have a noon state which is already separated into waveguide channels. We've built a photon triplet source where instead of generating photon pairs, you're generating photon triplets. We've also built a polarization entanglement source which generates polarization entangled belt states which, again, are already separated into two waveguides. So you can do fiber-butt coupling, you don't have to do any additional external manipulation. The sample I want to talk about is our Hormone chip sample. I think most of you are familiar with Hormone interference, just a brief recap. Two photons interfere at a balanced 50-50 beam splitter. There are four possibilities in principle. Both photons can leave the beam splitter at the left port or the right port, or both photons can be transmitted or reflected. If you do your quantum calculus and if the photons are indistinguishable, the center two possibilities cancel out and you're left with photon bunching. You measure that by looking at coincidence counts between the outputs of your beam splitter as a function of indistinguishability, for instance, a time delay. If you scan your time delay and you measure coincidences, you will get this homomandal dip, which ideally dips to below 50% visibility that tells you you're non-classical. Right, what do you do if you want to build a homomandal experiment? In bulk optics, you would start with a parametric down conversion source. That's a type two source, photons of orthogonal polarizations. You separate them on a polarizing beam splitter. You introduce a time delay. You swap the polarization of one photon. You interfere them on a 50-50 beam splitter. You look for coincidences. We thought, let's do that on chip. There we go. Polarizing beam splitter is easy. That's a directional coupler. I've shown you that we can build them. Half-wave plates are easy. These are electro-optic modulators. We can build them. Beam splitters are easy. These are just directional couplers. We can build them again. The question was, how the hell do you build a tunable time delay on an integrated chip? You can't just stretch and compress a wave guide. It doesn't respond very well to do in that. It will typically break. So how do you do that? And it turns out that for once in a lifetime in integrated optics, birefringent is not your foe. Typically, birefringent is an absolute pain in the ass if you excuse my language. But in that case, it's actually your friend. So what's happening? Birefringent means that two photons travel at different velocities. We generate our two photons. They propagate through the wave guide. We flip the polarization of one. We send them to the beam splitter. They arrive at the beam splitter at different times. This would mean we're measuring a point somewhere out there at our home dip where nothing interesting happens. However, what you can do is you can add an additional polarization converter directly after your sample. What that one does for you, PDC generates your photons. They walk off from each other. You flip the polarizations. Formally fast photon becomes the slow photon and vice versa. Now the slow photon has time to catch up during propagation in your sample. If at the right time you switch the polarization of the now slow photon, they will propagate together and meet at the beam splitter. That's exactly what's happening here. You flip the polarization of the photons. They meet up. You flip again, you send them to the beam splitter. Now, we've built a segmented polarization converter, meaning we can choose where to flip the polarization of the second photon, giving it more or less time to catch up with the first photon. That allows us to sample discrete time delays along the home dip. If we turn off the first polarization converter, we can sample the wings of our home dip. This is what you get from the experiment. The blue curve and the gray shaded curve are theory, gray shaded, including experimental imperfections. The dots are the measurement points. We see a dip visibility of 93%, which isn't too bad. It's also not too brilliant. But the important point is that we've managed to implement a tunable time delay of about 13 picoseconds on an integrated chip. So we can now delay our photons on demand by about 13 picoseconds, which is something that you might need in a large network for synchronization purpose or anything else. So we've built all of that. Where do we go from here? What are the next steps? I mean, essentially, we've kind of managed to take one of the quantum optical experiments and put it all on a single chip, like what's the big next step? Well, big next step is following. We've built that thing. That waveguide chip is already 90 millimeters long. There is no way we can fit more components onto one single chip. The reason for that is that we're working with weekly guiding waveguides, so we have to build comparatively large structures. Contrast to that, there's that whole field of silicon and insulator photonics. This is a picture, as an example, from Doug Englund's group at MIT. That chip comprises 88 max-inter interferometers with two thermo-optic phase shifters each, and a sample that's about five millimeters long. Why can they do that? They can do it because silicon waveguides have this incredible high refractive index change to air. You have very small structures. You can do very small banding radii. You have this high integration depth, and that's been optimized by industry. Computer chips are built on silicon technology. Combine these two. Take the high integration depth and the functionality of lithium niobate. You end up with lithium niobate on insulator. You have a thin film of lithium niobate sitting on an insulator. The technique to build this was, again, developed in Pada Born by an involved gang solace group. The guide responsible for that was Huihu. He's now back in China. He's running a company. He's selling that stuff. By now, this material is good enough for quantum applications. The losses are low enough for quantum applications. People are looking at that. People are looking at cascaded processes. People are looking at high-efficiency couplers to reduce losses. People are looking at low-less circuits. People are looking at niobate electro-optic modulators where they show modulation of up to 125 gigahertz. And I think that this is something where lithium niobate has to go towards to realize large quantum photonic networks on chip. Samaria talked in that lecture about the underlying concepts of why we want to do integrated quantum photonics in Pada Born. Why we think that bulk optics ultimately will not scale up to really large networks? Not saying that you shouldn't use bulk optics because for many applications, it's the easiest thing to do. But I don't think it will scale up to something like 10,000 nodes. I don't see it. I've shown you the device toolbox that you need to have at your disposal if you want to do integrated photonics. I've shown you our fabrication technology with which we can fabricate every single device from the device toolbox and finally shown you that we can combine them onto multifunctional chips and we can build quantum photonic circuits with our technology. But that I want to thank you for the first lecture. It's a picture of our group. I'm also thanking our funding agencies. Just to let you know, we do have open positions as for PhD students as well for postdocs. And we typically have open positions even if we don't advertise them. So if you should ever be in a position of looking for a new position and you think this could be interesting, just get in touch with us and we can see what we can do. That's it for that lecture. I'm happy to answer questions if there are any. Thank you.