Transcript of the presentation Platform for Photonic and Phononic Information Processing, given at the NTT Upgrade 2020 Research Summit, September 29, 2020

Thank you for coming to this talk. My name is Amir Safavi-Naeini. I’m an Assistant Professor in Applied Physics at Stanford University. And today I’m going to talk about a platform that we’ve been developing here that allows for quantum and classical information processing using photons and phonons or mechanical motion. So first I’d like to start off, with a picture of the people who did the work. These are graduate students and postdocs in my group. In addition, I want to say that a lot of the work especially on polling of the lithium niobate was done in collaboration with Martin Fejer’s group and in particular Dr. Langrock and Jata Mishra and Marc Jankowski. Now our goal is to realize a platform for quantum coherent information processing that enables functionality which currently does not exist in other platforms that are available.

So in particular we want to have a very low loss non-linearity that is strong and can be dispersion engineered, to be made broadband. We’d like to make circuits that are programmable and reconfigurable, and that necessitates having efficient modulation and switching. And we’d also really like to have a platform that can leverage some of the advances with superconducting circuits to enable sort of large-scale programmable dynamics between many different oscillators on a chip. So, in the next few years what we’re really hoping to demonstrate are few photon, optical nonlinear effects by pushing the strength of these non-linearities and reducing the amount of loss. And we also want to demonstrate these coupled, sort of qubit and many oscillators systems.

Now the material system that we think will enable a lot of these advances is based on lithium niobate, so lithium niobate is a ferrorelectric crystal. It’s used very widely in optical components and in acousto-optics and then surface acoustic wave devices. It’s a ferroelectric crystal that has sort of a built-in polarization. And that enables a lot of effects, which are very useful including the piezoelectric effect, electro-optic effects. And it has a very large chi-2 optical non-linearity. So it allows for three wave mixing. It also has some effects that are not so great, for example, pyroelectricity, but because it’s very established material system there’s a lot of tricks on how to deal with some of the less attractive parts of it of this material.

Now most surface acoustic wave, or optical devices that you would find are based on kind of bulk lithium niobate crystals that either use surface acoustic waves that propagate on a surface or, you know, bulk waves propagating through a whole crystal, or have a very weak weakly guided low index contrast waveguide that’s patterned in the lithium niobate. This was the case until just a little over a decade ago. And this work from ETH Zurich came showing that thin-film lithium niobate can be bonded and patterned. And photonic circuits very similar to photonic circuits made from three fives or silicon can be implemented in this material system.

And this really led to a lot of different efforts from different labs. I would say the major breakthrough came, just a few years ago from Marko Loncar’s group, where they demonstrated that high quality factors are possible to realize in this platform. And so they showed resonators with quality factors in the tens of billions corresponding to line widths of tens of megahertz or losses of, just a few, DB per meter. And so that really changed the picture and you know a little bit after that in collaboration with Marty Fejer’s group at Stanford they were able to demonstrate polling and so very large dispersion-engineered nonlinear effects in these types of waveguides. And, and so that showed that, sort of very new types of circuits can be possible on this platform.

Now our approach is very similar. So we have a thin film of lithium niobate, and this time it’s on sapphire instead of oxide or some polymer. and sometimes we put oxide on top. Some Silicon oxide on top, and we can also put electrodes, these electrodes can be made out of a superconductor like niobium or aluminum or they can be gold depending on what we’re trying to do. The sort of important thing here is that the large index contrast means that light is guided in a very highly confined waveguide. And it supports bends with small bending radii.

And that means we can have resonators that are very small. So the mode volume for the photonic resonators can be very small, and as is well known, the interaction rate scale is one over squared of mode volume. And so we’re talking about an enhancement of around six orders of magnitude in the interaction length, interaction lengths, over systems using sort of bulk components. And this is in a circuit that’s sort of sub millimeter in size, made on this platform. Now interaction length is important but also quality factor is very important. So when you make these things smaller you don’t want to make them much lossier. That’s, you know, you can look at, for example a second harmonic generation efficiency in these types of resonances and that scales as Q, to the power of three essentially.

So, you need to achieve, you win a lot by going to low loss circuits. Now loss and non-linearity are sort of material and waveguide properties that we can engineer, but design of these circuits, careful design of these circuits is also very important. For example, you know, because these are highly confined waves and dielectric waveguides they can, you can support several different orders of modes especially if you’re working for broadband light waves that span, you know, an octave. And now when you try to couple light in and out of these structures, you have to be very careful that you’re only picking up the polarizations that you care about, and you’re not inducing extra loss channels, effectively reducing the queue, even though there’s no material loss if you’re these parasitic coupling, can lead to lower Q.

So the design is very important. This plot demonstrates, you know, the types of extrinsic-to-intrinsic coupling that are needed to achieve very high efficiency SHG, which is [unintelligible] unrelated to optical parametric oscillation. And, you know, you, so you sort of have to work in a regime where the extrinsic couplings are much larger than the intrinsic couplings. And this is generally true for any type of quantum operation that you want to do. So just low material loss itself isn’t enough, the design is also very important.

In terms of where we are, on these three important aspects like getting large G large Q and large cap up, so we’ve been able to achieve high Q in, in these structures. This is a Q a of a couple million. We’ve also been able to you can see from a broad transmission spectrum through a grading coupler you can see a very evenly spaced modes showing that we’re only coupling to one mode family. And we can see that the depth of the modes is also very large, you know, 90% or more. And that means that our extrinsic coupling to intrinsic coupling is also very large. So we’ve been able to kind of engineer these devices and to achieve this.

In terms of the interaction, I won’t go over it too much but, you know, in collaboration with Marty Fejer’s group we were able to pull both lithium niobate on insulator and lithium niobate on sapphire, and we’ve been able to see a very efficient, sort of high slope proficiency second harmonic generation, you know achieving approaching 5,000% per watt centimeters squared for 1560 to 780 conversion.

So this is all work in progress. And so for now, I’d like to talk a little bit about the integration of acoustic and mechanical components. So, first of all why would we want to integrate mechanical components? Well, there’s lots of cases where, for example, you want to have an extremely high extinction switching functionality. That’s very difficult to do with electro-optics because they need to control the phase, extremely efficiently with extreme precision. You would need very large, long resonators and or large voltages, it becomes very difficult to achieve you know, 60 DB types of, switching. Mechanical systems, on the other hand, they can have very small mode volumes and can give you 60 DB switching without too many complications. Of course, the drawback is that they’re slower, but for a lot of applications, that doesn’t matter too much.

So, in terms of being able to make, integrate MEMS, switching and tuning with this platform, here’s a device that achieves that, so that each of these beams is actuated through the Piezoelectric effect and lithium niobate via this pair of electrodes that we put a voltage across. And when you put a voltage across these have been designed to leverage one of the off diagonal terms in the piezoelectric tensor, which causes bending. And so, this bending generates a very large displacement in the center of this beam. In this beam, you might notice is composed of a grading, and this grading effectively generates, it’s a photonic crystal cavity. So it generates a localized optical mode in the center which is very sensitive to these displacements. And what we’re able to see in this system is that you know, just a few millivolts, so 50 millivolts here shifts the resonance frequency by much more than a line width, and just a few millivolts is enough to shift by a line width to achieve switching.

We can also tune this resonance across the full telecom band and these types of devices whether in waveguide resonator form can be extremely useful for sort of phase control in a large scale system, where you might want to have many phase switches on a chip to control phases with low loss. Because these wave guides are shorter, you have lower loss propagating across them. Now, these interactions are fairly low frequency. When we go to higher frequency, we can use the electro-optic effect. And even the electro-optic effect even though it’s very widely used, and well-known on a photonic circuit like these lithium niobate photonics circuits has, interesting consequences and device opportunities that don’t exist on the bulk devices.

So for example, let’s look at single sideband modulation. This is what an electro-optic sort of standard electro-optics, single sideband modulator looks like. You take your light, you split into two parts, and then you modulate each of these arms. You modulate them out of phase with an RF tone that’s out of phase. And so now you generate side bands on both and now because they’re modulating out of phase when they are recombined and on the output splitter and this mock sender interferometer you end up dropping one of the side bands and then the pump and you end up with a shifted sideband. So that’s possible you can do single side band modulation with an electro-optic device but the caveat is that this is now fundamentally lossy.

So, you know, you have generated, this other side band via modulation, and the sideband is simply being lost due to interference. So it’s getting combined, it’s getting scattered away because there’s no mode that it can get connected to. So, actually, you know, this is going kind of an efficiency less than 3 dB usually much less than 3 dB. And that’s fine if you just have one of these single sideband modulators because you can always amplify, you can send more power but if you’re talking about a system and you have many of these and you can’t put amplifiers everywhere then, or you’re working with quantum information where loss is particularly bad, this is not an option. Now when you use resonators, you have another option.

So, here’s a device that tries to demonstrate this. This is two resonators that are brought into the near-field of each other. So, they’re coupled with each other over here where they’re, which causes a splitting. And now when we tune the DC voltage, which tunes one of these resonators by sort of changing the effective half lengths in one of these resonators, it tunes the frequency, we can see, we should see an anti-crossing between the two modes. And at the center of this splitting, this is versus voltage, a splitting at the center at this voltage, let’s say here it’s around 15 volts. We can see two residences two dips, when we probed the light field going through. And now if we send in the pump resonant with one of these, and we modulate at this difference frequency we generate this red sideband but we actually don’t generate the blue sideband because there’s no optical density of state. So because this other side is just not generated, this system is now much more efficient.

In fact, so in Marco Loncar’s group ,they’ve demonstrated that you can get a hundred percent conversion. And we’ve also demonstrated this in a similar experiment showing that you can get very large sideband suppression. So, you know more than 30 dB suppression of the sidebands with respect to the sideband that you care about. It’s also interesting that these interactions now preserve quantum coherence. And this is one path to creating links between superconducting microwave systems and optical components. Because now the microwave signal that’s scattered here preserves its coherence.

So, we’ve also been able to do acousto-optic interactions at these high frequencies. This is a, this is an acousto-optic modulator that operates at a few gigahertz. Basically, you generate electric field here which generates a propagating wave inside this transducer made out of lithium niobate. These are aluminum electrodes on top. The phonons are focused down into a small phononic waveguides that guides mechanical waves. And then these are brought into this crystal area where the sound and the light are both confined to wavelength scale mode volume and they interact very strongly with each other. And this strong interaction leads to very efficient, effective electro-optic modulation. So here we’ve been able to see, with just a few microwatts of power, many, many sidebands being generated.

So, this is a effectively an electro-optic modulator where the v-pi is a few thousands of a volt instead of, you know, several volts, which is sort of the off-the-shelf, electro-optic modulator that you would find. And importantly, we’ve been able to combine these photonic and phononic circuits into the same platform. So this is a lithium niobate, the same lithium niobate on sapphire platform. This is an acoustic transducer that generates mechanical waves that propagate in this lithium niobate waveguide. You can see them here and we can make phononic circuits now. So this is a ring resonator. It’s a ring resonator for phonons. So we send sound waves through. And when its resonance, when its frequency hits the ring resonance, we see peaks. And this is, this is cheeks in the drop port coming out.

And what’s really nice about this platform is that we actually don’t need to, unlike many MEMS platforms where you have to have released steps that are usually not compatible with, you know other devices, here there’s no release steps. So, the phonons are guided in that thin lithium niobate layer. The high Q of these mechanical modes shows that these mechanical resonances can be very coherent oscillators. And so, we’ve also worked towards integrating these with very non-linear microwave circuits to create strongly interacting phonons and phonon circuits. So this is an example of an experiment we did over a year ago, where we have sort of a superconducting Qubit circuit with mechanical resonances made out of lithium niobate shunting the Qubit capacitor to ground.

So now vibrations of this mechanical oscillator generate a voltage across these electrodes that couples to the Qubits voltage. And so now you have an interaction between this qubit and the mechanical oscillator, and we can see that in the spectrum of the qubit as we tune it across the frequency band. And we see splittings every time the qubit frequency approaches the mechanical resonance frequency. And in fact this coupling is so large, that we were able to observe for the first time, the phonon spectrum. So we can detune this qubit away from the mechanical resonance. And now you have a dispersive shift on the qubit, which is proportional to the number of phonons. And because number of photons is quantized. We can actually see, the different phonon levels in the qubit spectrum.

Moving forward, we’ve been trying to also understand what the sources of loss are in the system. And we’ve been able to do this by demonstrating, by fabricating very large arrays of these mechanical oscillators and looking at things like, their quality factor versus frequency. This is an example of a measurement that shows a jump in the quality factor when we enter the frequency band where we expect our phononic band gap for this period, periodic material is. This jump you know, in principle, if loss were only due to clamping, only due to acoustic waves leaking out in these out of these ends, then this change in quality factor, quality factor should go to essentially infinite or loss should be exponentially suppressed with the length of these, but it’s not. And that means we’re actually limited by other loss channels.

And we’ve been able to determine that these are two-level systems in the lithium niobate by looking at the temperature dependence of these losses and seeing that they fit very well sort of standard models that exist for the effects of two-level systems on microwave and mechanical resonances. We’ve also started experimenting with different materials. In fact, we’ve been able to see that, for example, going to lithium niobate that’s doped with magnesium oxide changes or reduces significantly the effect of the two-level systems. And this is a really exciting direction of research that we’re pursuing. So we’re understanding these materials. So with that, I’d like to thank the sponsors. NTT Research, of course, a lot of this work was funded by DARPA, ONR, ARO, DOE very generous funding from David and Lucile Packard Foundation and others that are shown here. So thank you.