Programmable Photonics - Wim Bogaerts - Stanford

all right so i'm gonna talk about programmable photonics and before i dive into the topic itself let me very briefly tell me where i'm coming from and david already gave a bit of an introduction there i'm from what we call the photonics research group at ghent university but we're also a part of the imec nanotechnology center which is one of the few technology centers in the world that can still do the development for really advanced cmos because the cost of those developments is getting increasingly more expensive but we're not doing cmos electronics we're doing photonics and since the past 20 30 years we've been mostly working on technology to bring optical functions to the surface of a chip and in particular a silicon chip so that's what we call silicon photonics which i'm going gonna touch on a bit uh later in my presentation uh we started out as a telecom data com group because that was where photonics was really a hot thing uh let's say like 20 30 years ago but since the dot-com bubble uh we started diversifying our activities going more into sensors there's a bit of a question whether we were first diversifying into sensors and that was the cause of the dot-com bubble or it was the other way around that's not entirely clear but okay around that time early 2000s we also started looking into a lot of different applications for photonics and since recently the past five years i've been working myself more and more into an area which we call programmable photonics now if we think about programmable and photonics these are two distinct words if you think about programmable things you think about all kinds of stuff which you can manipulate in software well if you think about photonics photons it's manipulating light or more particularly manipulating light on a microscopic scale so if you think about programmable photonics obviously we want to manipulate light in software on a microscopic scale now why the hell would we want to do this well for one good reason light contains information and light can transport information just to give you an idea light is an electromagnetic wave but with an extremely high oscillation frequency on the order of 200 terahertz compare that to radio waves and microwaves that are used quite a lot for uh communication the the frequencies of photonics are huge which means that there's an enormous bandwidth at your disposal just to modulate and transport information and that's the whole engine behind optical fiber communication is that you have this massive bandwidth of a photonic carrier so one of the applications of photonics that we're seeing more and more is that instead of using your processing of signals in the microwave regime at very high gigahertz or tens of gigahertz frequencies is that you transport your radio frequency signal into the optical domain process it in an optical system where you have all this bandwidth at your disposal and then at the very end transport it back into the microwave domain and you can do that with the system like you see here on the side you basically have a laser coming in with light you have your rf your microwave inputs on the right which modulate a signal onto your laser signal laser line then you have an optical s system in between which basically processes the optical system optical signal and then when you end up at the other side you basically translate your microwave signal back to sorry optical signal back to your microwave signal using photo detectors so with a couple of building blocks you can go back and forth between high-speed radio and microwave signals including digital communication signals to the optical domain and back now that's just one way of manipulating information with light if you know about beams of light you know that you can encode information in many ways you can encode information in the total power in the intensity profile of the beam the face profile the wavelength or color of light like think about red green blue as three different wavelengths that's three types of information and things like polarization where your light is oriented horizontally or vertically so how do you process this information that's in a beam of light well even if a beam of light just propagates it already does something with the information that is in there just take about take think about a light profile a spatial profile of light as you propagate that beam of light that spatial profile will change it will diffract according to maxwell's equations so essentially while the beam is processing it's actually solving maxwell's equations in real time so if you put in the right input profile you can essentially by looking at the output profile know what the solution is of maxwell's equations in free space now okay that's maybe that looks very powerful but it can only do one thing that's that's these that's solving this particular integral well yes and no you can put something in between the path of light to manipulate that information just a simple example if you put a lens in the path of your light and you basically look at how the information in your two focal planes of the lens relate to one another it turns out that the lens actually performs a fourier transform in real time in the time that it takes for the light to propagate through the lens you have done a real-time two-dimensional fourier transform of your optical pattern from your input plane into your output plane now that's kind of nice because now you have with the lens a very simple way without power consumption to perform for your transforms and you can do lots of other things you can use face plates you can use wavelength filters you can use diffractive elements you can use mirrors polarizers to basically implement different functions of your uh for of your light processor now if you want to do this and make your processor more complicated the classical way to do that is on an optical bench you basically put all your components in the right position you carefully align them and then you have a transformation of your beam of light however if all of these components like lenses and mirror are just there your transformer does only one thing it's basically a fixed setter you need something if you want to do really interesting manipulations you need something that's programmable that you can reconfigure in software now the trick there is to cut up your beam of light into pixels essentially we know all about pixels a display has pixels an imager or a camera has pixels so if you now chop up your beam of light into pixels and then you come up with some system that processes that beam of light at the other side you record the pixels and you have a kind of discretized analog processor so you need some digital to analog converters and unlock digital converters now what is in the middle there some general system that transformed this plane of pixels on the left to the plane of pixels on the right how do we generate that plane that plane of pixels what with what we call a spatial light modulator which is essentially the same thing as a display except that it's usually it modifies the phase of the light rather than the amplitude of the light ideally it modifies both you can modify both for each pixel the phase of the light and the amplitude and the spatial light modulators can be implemented in many ways depending on whether you want a digital signal or an analog signal but quite the quite common implementation is based on liquid crystals now the nice thing of the spatial light modulator is that it basically behaves as a display you can put an image into it from software to determine what the pattern is that comes out so now you can build a system like this for instance this is called an optical convolution processor it has at the entrance a spatial light modulator where you have a picture you then go through a lens which does a fourier transform for you and then in the middle you have a kernel this is essentially a reference image the input images for your transform passes through the kernel and then it's for your transformed again together with the kernel and what you end up at the output is essentially a convolution of your input and your kernel image so in real time again the time that it takes for light to propagate from left to right you're executing a two dimensional high resolution convolution and this is not some some crazy magic idea this this really exists this is a product from a company in great britain called optolysis who have basically built such a lens system onto an expansion card for a pc and it's quite you need a quite bulky pc to fit this in but this essentially just runs real-time convolutions in such a system with multiple spatial light modulators and of course at the other side you have an imager to see what the pattern is like and you can further generally generalize this idea you can for instance go to multiple layers of these spatial light modulators as many as you like to basically get whatever operation you would like to to do and an example of this uh was done by a collaboration with nokia bellaps and the university of queensland and i believe that joel carpenter who made this nice picture uh is actually giving a talk in one of these seminars as well so stay tuned now the idea is that with this it's essentially a light bouncing back and forth between a source the target and the same spatial light modulator on the mirror so it goes seven through seven transformations from input to output still this is not fully generic if you want something that's really generic you essentially want your set of pixels on the left your set of pixels on the right and then a component in between that can perform any linear transformation between the input pixels and the output pixels essentially doing a matrix vector multiplication in real time but for that you need to be able to configure every individual matrix element in this intermediate block so how you do that well turns out the idea was already presented like almost 30 years ago you basically can discretize your whole system into a set of tunable phase shifters and tunable beam splitters so basically doing a cascade in the triangular configuration doing a cascade of uh beam splitters would split the light and combine it again and then in between you control the phase of the light so that you control essentially whether at each beam spreader the light interferes constructively or destructively or some something in between now this idea was already presented in the 90s but it has been laying dormant for almost 20 years until david miller from stanford here uh essentially dug up the idea reinvented it and came up with some very clever idea to combine it with clever self-configuring control algorithms essentially turning this system which can do a unitary transformation between the input and the output to turn it into a self-configuring beam coupler now this essentially looks like the holy grail you can do any transformation from left to right whatever you like to do by just programming all these phase shifters and these tunable couplers but it's still a bulky system if you want to build this on an optical table you're going to need a lot of these beam splitters and tunable couplers and uh and phase shifters and in the end it can become quite pretty pretty difficult to to keep everything properly under control and that's where we go into integrated photonics rather than building a system like that on an optical table why not put everything on the surface of a chip i mean this technique worked for electronics it worked really well by integrating electronics on a chip you could make much more complex circuitry everything started to work better was more reliable and then the end became a lot cheaper and with a lot lower power consumption by just bringing all these electronic functions together on a chip and with integrated photonics or photonic integrated circuits we basically want to do the same thing we want to bring all these common photonic functions like light sources wave guides filters signal modulations and detection all together on a single chip and you connect everything together with what we call a waveguide basically making little circuits that control light rather than controlling electrical signals now the key element here to make this work is what we call a waveguide what is a waveguide well imagine what you have if you don't have a waveguide if you have a set of rays or beams of light they would diffract outward a waveguide is essentially a bus for light where your light is confined internally through total internal reflection and so it bounces back on the side walls and and propagates along the waveguide itself and that works if you have a core that has a high refractive index surrounded by a cladding that has a lower refractive index and the best example that you have for this kind of device is an optical fiber an optical fiber essentially consists of a small core of 10 micrometer diameter of glass surrounded by a slightly different glass with the lower refractive index and that gives you like a mode with roughly the area of this circle 10 microns across and so that essentially in the core of such an optical fiber you have this kind of distribution of light sitting there in the center now 10 micrometer across is quite large and the reason that it's large is because the the index contrast between the core and the cladding in an optical fiber is quite small it's a very tiny difference in refractive index if you go to different materials like semiconductors you can shrink this mode down to micrometer size or in the case of silicon even to sub micrometer dimensions like half a micrometer wide and that brings me to silicon photonics silicon photonics is essentially a technology which allows you to make wave guides optical waveguides for infrared light that are smaller than a micrometer in size and the nice thing is that you can also bend them very tightly in a few micrometer bend radius and you can space them close together so that means that if you you have these tiny waveguides and they're made in silicon you can put them on the surface of a chip you can now put a lot of these waveguides on a certain area of chip and that gives you an enormous scale advantage these sub micrometer waveguides which come from the high refractive index contrast of silicon basically makes it possible to make really large circuits and on top of that silicon photonics has another very important scale advantage it's compatible with large-scale manufacturing the word silicon is not coincidental silicon photonics is essentially a technology that allows you to use the same technology for electronics manufacturing but now to make photonic chips so you basically can re use all this technology which has taken trillions of dollars investment just to realize to build these massive high production fabs you can now use that same technology to make photonic chips and as a result if you look at the price advantage of electronics well you have the same advantage in photonics you can make really compact low cost high volume chips so just an example this is this is a recently fabricated silicon photonics chip that we made it's about two by two centimeters in size but it contains a lot of experiments so at like 200 different circuits and in total about 100 000 photonic components on this chip for photonic chips that's quite a lot it has like 500 interfaces for optical fibers and about 2 000 interfaces for electrical signals so that's that just gives an idea of how densely you can integrate circuits on a silicon photonics chip and that integration makes it possible to now also make these complex photonic circuits now for those of you who don't know how you make such a silicon photonics chip you basically need on the same chip to make a transport for light like waveguides filters signal modulation detection etc so how is how is this made well we start typically off from a substrate in silicon which has a thin layer of silicon dioxide glass on top of it and on top of that an even thinner layer of silicon and it's in that top layer that we're going to make our little waveguides so you start with a blank surface typically on a 200 millimeter or 300 millimeter wafer and then you coat that surface with a material called photoresist which is essentially a photosensitive layer which you can then illuminate with a uv pattern that come that you have prefabricated before on a mask now that uv pattern defines where your waveguides are going to be so essentially where the resist is illuminated it remains the other resist can be selectively removed with the solvent and then the remaining layer of resist can now be used as a mask to transfer the pattern into the silicon with a reactive ion etching finally you strip your resist away and what you end up with is these little waveguides and this is only the start of your process you can pattern a second layer like for instance partially etched you can also implant dopants so now you can make pn junctions inside your waveguides which you can use for very fast modulators you can implement the photo detectors by locally growing a layer of germanium inside a window like this you grow the germanium you planarize it with some polishing step and then basically you can you have now these optical and electrical components you can put some electrical connectivity like little plugs of tungsten copper wires and then finalize it with some aluminum bond pads this is a typical process for all being simplified for a silicon photonics platform and what you end up with is a cross section that looks a bit like like this where you have so your your silicon wave guides you have uh modulators on the far end you have photo detectors and you have an electrical connection for driving everything with electrical signals and thanks to this type of technology we've seen over the past decades we've seen the number of components on silicon photonic circuits steadily grow so circuits becoming ever more complex and we're going from these simple like static circuits where basically everything is fixed to a circuit where you can tune for instance your configuration of your filter so you get all these tunable elements there to circuits that can do multiple functions at the same time now if you want to make these circuits more complex obviously your components that govern your circuit should become better and better and so one of the things that a lot of a lot of effort has been invested in the past decades is making tunable elements that are compact have a short optical length a low optical loss and a low electrical power consumption and that's not so trivial especially that last item is kind of difficult because the standard way to tune a photonic component on a chip is putting a heater next to it either a metal resistor on top or a metal resistor on the side so for instance in this cross-section you could have a tungsten heater on the top or you could have these doped silicon heaters on the side now the problem with these heaters is kind of obvious i mean you have this kind this resistor which is just burning power it's just heating locally an element of your chip if you have thousands of those you really need good cooling and you're essentially basically killing the planet so can we come up with something better than heaters well there's a lot of research going on looking into carriers which are very good for really fast tuning and modulation but also to lower power systems like using liquid crystals or using microelectro-mechanical systems like moving little moving elements so especially the last two are two activities that we've been working on quite a lot because the type of circuits that we make we want to tune so many elements that we really don't want to rely on heaters in the long term so for instance one of those is liquid crystals if you have a liquid crystal which is a very birefringent material they have these little sausage-like molecules if you apply an electric field over the liquid crystal you're basically reorienting your molecules and the light sitting in your waveguide feels that so you can actually by doing this you can tune the behavior of a photonic element just by applying a voltage so you're not sending a current to a liquid crystal you're just applying a voltage that doesn't consume a lot of power now how do you integrate these liquid crystals with all these other components in your silicon photonics circuit because your waveguide is sitting here at the bottom under a thick layer of oxide well you open up that layer and then you have this little well which you can then fill with liquid crystal that looks easier than it than it actually is it's always easy to draw these things in powerpoint but the idea is that if you have these wave guides sitting there and you have this liquid crystals getting infiltrating that gap in between the waveguide and the little piece of silicon next to it with a very small voltage you can actually flip the liquid crystal direction 90 degrees which gives you a very strong effect and it takes a bit of tweaking to get a liquid crystal actually inside these little cavities uh and most more recently we're doing this with inkjet printing so that helps and indeed you see if you apply a voltage over the liquid crystal that your response of your photonic circuit changes so you can with this technique program your photonic circuit every individual element in a low power way another technique that we're working on is micro electromechanical systems essentially moving waveguides and this is not a new thing wave guide mems have been around for for years but in the past three years we've been working together with other people in a european project called morphic on a technique to integrate these moving elements inside our silicon photonics platform that already has all these other functions and that's quite a challenge because now you need again to open this area where the waveguides are locally without damaging the waveguides and then with after putting some protection layers down you have to undercut locally the area where your waveguide is so that it can move and all of that without affecting the performance of everything else in your silicon photonics platform now after quite a bit of development this is what we got and this makes for such nice sem pictures you see these freestanding wave guides and if you move a bit closer what you see here is this kind of wave guide which has a ski next to it that's key is essentially a an optical phase shifter so if we move that ski we're actually moving silicon material that little ski from close to the waveguide too far away from the waveguide and that has a very strong effect on our on our response of our waveguard so here there's another variation of that so you see this kind of waveguide coming in from the top and then exiting on the other side and the the part at the bottom can actually move vertically in the picture moving closer or further away from the waveguide and this applies what we call an optical phase shift and it works pretty well this is a this is an example of a response curve so with a fairly small component we're talking about 30 to 40 microns of length we can induce a complete 180 degree or pi phase shift and these mems can do other things as well for instance this is an example of a directional coupler so your light comes in from the bottom and it exits on the top and at the bottom on the right side and how much is the splitting ratio is determined by how close these two wave guides are together so if you move one of the waveguides with this kind of checkerboard actuator here then you basically influence the coupling of these two waveguides and now if you have these mems you can now combine these tunable couplers which is essentially the same thing as the tunable beam splitter that i introduced earlier and this tunable phase shifter which is the same as the phase shifter i used earlier to into circuits which could then be programmable now it takes a bit of work to get this operational but now with like with this mems components we now have a library that doesn't contain just all the bells and whistles of the existing silicon photonics platform but also has low power tunable couplers switches and phase shifters so if we have these building blocks we can now use them to make circuits so and as the circuits become ever more complex we need more of these building blocks more tuning elements and more electronic control as well now if we if we look at these circuits that i draw here and also all these circuits that we see on this nice little scaling graph there's one thing we should remember all of these circuits have been designed for one particular purpose they're essentially asics application specific circuits now what's the implication of this well take this example if you have an optical link in a data center and you want to design or make a new optical link you have to design fabricate package and test a chip so that means if you choose a particular type of optical link for instance according to a protocol called parallel single mode fiber you have to design a chip with four modulators for detectors arranged in this particular circuit topology and essentially your light comes in it goes out to your fibers and then your light from the other direction comes back and is connected to the photo detectors but if for some reason you want to change this link into what we call for instance coherent link want to upgrade this link we need to take the same modulator same photo detectors again and now design a different circuit that does coherent modulation and has a coherent receiver and if we want to switch protocols again to something called wavelength division multiplexing we now have to build a circuit that has four wavelength channels and multiplexes those with a filter bank or has and has a filter bank that demultiplexes the wavelength channels into four photo detectors every time you want to build a new chip like that or you want to change protocol you need a new chip and that basically takes you about a year to develop and quite a lot of money and so in terms of a roadmap for product development this is not a very good thing so can we do with electronic with photonics the same thing as you do when you want to prototype some new electronics if you want to prototype some new electronics you basically just take an fpga or a digital signal processor or a microcontroller and within a couple of weeks you have a working prototype that you can test with a customer and only if you really need it you're going to design an asic afterwards so can we have something that's similar to field programmable gate arrays i mean if you look at the history of field programmable gate arrays these are digital configurable logic with programmable interconnects and they're extremely successful i mean this this is really a success story of electronics now people have tried to do this for analog functions as well like the analog functions we want to use for photonics but there the story is far from successful there's a couple of technologies that that try to accomplish this but it doesn't scale well so this is an interesting opportunity for photonic fpgas can we combine the best of both worlds and do analog optical functions that can be digitally programmed for high bandwidth signals and that of course should be similar as fpga is extremely successful so can we do something like photonic fpgas or programmable photonics i want to throw in a definition here with photonic fpgas i mean photonic integrated circuits that can be reconfigured in software to perform different functions and i think all of these are important it's not just configuring software for one function it should be really for different functions in order to fulfill the same role as an fpga so that means that we're going from this kind of complex application specific circuit to something that is more generic and you see that this already has a more generic layout that can be programmed to perform different functions and this is what we call a general purpose photonic integrated circuit which is fully tunable which can route arbitrary optical paths and can also implement generic functions so how this would work is that you have your laser source you have your microwave inputs which are translated into optical signals using modulators and then you have your microwave outputs which are basically converted by photo detectors and of course you also want to process optical signals directly so you have optical fiber inputs and output parts as well now at the heart of this is what we call a two by two optical gate this is a mesh of wave guides which has two by two gates two inputs two outputs now how does this work well you have an input wave in this optical gate and somehow this gate controls how much the two inputs are being coupled and what the phase delay is between the two outputs and this can be described by a very simple matrix a unitary matrix so what we have here is a component that in real time performs a two by two unitary transformation and the way we can implement this on a photonic chip is by using either a tunable coupler and a phase shifter or by using a little circuit of two one or more phase shifters called a maxender interferometer now the simplest one to understand is the top and just to give you an understanding of how you could control such a tunable gate let's look at a tunable gate where we have two beams of light coming in and we want to couple all the light out into output number two now for that we basically put a detector at output one to monitor what's going on and we use that detector to tune the face of that phase shifter in such a way that the two beams falling on the detector have as much destructive interference as possible essentially it turns the two inputs into antiphase so that they cancel out and in the second step the signal from the detector can be used to change the coupling ratio between the two essentially making sure that all the light ends up in output number two and the nice thing of this two by two matrix is that you now can cascade that into networks so you can make a chain of those for instance to couple all the light now into output four by systematically tuning gate by gate and trying to cancel out first the light in output one then the light in output two and then finally with third gate you cancel out the light in output tree and as a result by systematic simple local optimizations you can configure this little circuit into an automatic beam coupler that whatever combination of beams that comes in on this side the circuit will automatically adjust itself to couple all the light to the four outputs now the trick here is that you need something like detectors there but if you want to cascade this into multiple layers you need something what we call a transparent photodetector so essentially we want the light that hits the photo detector we want most of that to go through and then we can put a second layer of gates after that and essentially the third layer as well so you basically can get a kind of triangular grit or triangular mesh of waveguides and this was proposed by rec and then also by david miller later on so it's essentially the equivalent of this scheme of free space beam beam splitters and phase shifters into an optical circuit and this is just one possible implementation of such a mesh you can also do a mesh like this which can essentially perform almost the same function meaning a unitary transformation from the left to the right the first demonstration of such a circuit was already in 2015 by the group of by the university of bristol this was on in a glass chip was a quite large chip the first demonstration in silicon was by our group at ghent university one year after that on silicon and that was a much much smaller chip like a few millimeters across and already a year later we saw this massive chip from mit with 26 inputs and 26 outputs so the scaling goes pretty quickly now why would you want to build a mesh circuit like that well it's interesting because if you represent your inputs the amplitudes of your inputs as a set of complex numbers amplitude and phase and you do the same for the output what this circuit essentially does it acts as a matrix that performs a matrix multiplication on your input vector so in real time the time that it takes for a light to propagate from left to right you have done a matrix multiplication a linear operation also called a matrix vector or multiply accumulate operation and this turns out to be one of the basic operations in a lot of pattern recognitions and neural network operations and it's also one of the fundamental operations in linear quantum optics so it's not surprising that we've seen a lot of the research going on in these meshes for quantum applications and also for accelerators for artificial intelligence now these circuits are already pretty generic but they're typically also that they're typically focused still on one particular type of functions these linear operations from left to right so they're programmable but they're not fully generic if you want to make them really fully generic you need to step away from this left to right architecture and go to something that's a bit more powerful what is missing in this left to right or forward only architecture well all that parts are actually the same length so programmable delays are a bit difficult there's also no feedback the light also only has feet forward there's no wavelength filtering in there there's no dispersion engineering if you want to do time signal processing or operations like integration or differentiation or even full-scale programmable signal processing you need something more and that's where the architecture of a recirculating mesh comes in essentially a waveguide mesh where all the ports are equivalent and can act as an input as an output and the way to implement that was already proposed in 2016 is by looping your wave guides into rings and then coupling these rings together and the rings can have four ports or three ports or six ports there's different variations but if you now inject light it basically starts recirculating and coupling from ring to ring and you can still couple your light from any point to any point in your circuit and the same you can do with this hexagonal mesh and again the basic building block is just that same two by two optical gate and if you have a measure of these gates you can now essentially by programming these gates from bar state to cross state you can just route your light anywhere in your mesh the way you like it even with multiple routes you can even send light through the same coupler not just side by side but you can even cross your light which is kind of interesting and nothing stops you from basically tuning your couplers to a partial complex date so you can use it as a splitter tree to redistribute your light and you can even route the light back again to combine it and this gives you essentially an interferometer because you can now introduce path length differences with along the two parts that your light is following and that gives you a wavelength dependent interference so you get a wavelength filter by just programming the routing in your in your mesh and you can even route the light in a loop which is eventually gives you a resonator that gives you these very sharp filter lines now the first practical implementation of this device was done by the university polytechnical of valencia by the group of jose carmani and already in this seven cell mesh they could implement more than 100 optical functions like basically different delays along different parts different splitter trees splitter ratios with different delays so and you can basically tune the splitting ratio that you want to do by just tuning the coupling ratio of every individual coupler you can do these interferometric uh forward-only circuits so essentially doing the same thing as in the forward-only mesh but now in this recirculating mesh and you can make filters like in this case a double interferometer again a double interferometer or in this case a ring resonator and because you can program every individual element you can tune the exact filter characteristic you can tune it which wavelength the slide is filtered and how how much of the light is filtered you can there are different ways to make these filters you have some limitations because of course you have to have this unit cell of length that you have to repeat so you have a set of discrete types of filters that you can build with this but it's already quite flexible skip this so how do you scale up these programmable meshes well we already saw that we needed really good couplers and tuners so we needed these cop and for instance the mems and the liquid crystal is one way to do it but you also need a lot of those and so that means that you need to electrically control them and you need to know what's going on on the chip so you need to come up with a way to drive your phase shifters and tunable couplers with efficient electronics so you need to get good connections between your photonics and your electronic drivers and that's still quite of a challenge ideally you have your photonics and your electronics on the same chip but in practice you end up with more something like this where you go package them or in best case you do a flip chipping or 3d stacking but that requires quite a lot of dedicated very specialized electronics design now one of the solutions that we're working on again in that morphic project that i told you is a kind of packaging approach where we have an interposer where the photonics and the electronics are connected with up to 3 300 connections between your photonics and your electronics chip and on top of that driving you need some way to control it so that means that you need monitors in your circuit that are read out that couple back into some control loops which can be implemented either in analog electronics or digital electronics or even just software but the control of these meshes is not straightforward especially these recirculating meshes because they have can you can have resonances in there they're pretty difficult to control so there's a lot of need for new control algorithms to drive these meshes and that this is a kind of a wide open space to come up with really good algorithms to control these because you need not only to go back and forth you need up front to know where you have to put all your photo detector monitors in order to measure what you're going to control and then on top of that you need to add multiple layers of control like not just the local control layers but also ways to program your new functionality into your circuits and then adaptive mechanisms that make sure that your circuit keeps working as it should so that essentially brings you to new ways of designing circuits rather than drawing circuits as patterns on a chip or even as traditional circuits connected by waveguides you now have a way of designing circuits in a chip that is already there and where you just program the functionality as essentially a software routine and that gives you the the same kind of functionality as fpgas for electronics you can now in software program something very rapidly it also creates a lot of opportunities for coming up with new types of algorithms new types of design ip for these new kinds of programmable photonic circuits like how to define a circuit in this waveguide mesh or how to control it or even how to incorporate a waveguide mesh now in a new specialized photonic circuit just like you put a microprocessor core in your own custom chip and so quite a lot of work is needed in this area to come up with good algorithms and one of those for instance is uh graph based algorithms that we are working on where we're trying to capture these photonic circuits in a graph based representation which is kind of difficult because we ended up with something that looks a bit more complicated than you would think intuitively but it works and you can then use existing graph algorithms to route these uh paths of light through this photonic mesh circuits you can also do distribution problems like splitting your light up depending on boundary conditions you can do a kind of tap base distribution or you can split up your light as early as possible now this gives you quite a lot of layers that you need in order to make these things work you just don't just need the photonics you need electronics on top of that both analog for the drivers and digital for the control and then software layers to program everything so eventually you end up with something that looks a bit like this so you have your photonics chip your driver electronics sitting on top of that you need fiber assembly array fiber ray assembly which is still a costly proposition if you want to process microwave signals you need some radio frequency amplifiers there you need some radio frequency packaging you need everything in a nice stable housing also temperature control and then on top of that on top of all these mechanical implementations you need your software layers to make it work so this is a much more elaborate technology stack than we see today with standard regular photonic circuits but if you bring all of that together what you end up with is this kind of generic programmable optical processor which has its radio frequency inputs its radio frequency outputs its photonic inputs and outputs and then this waveguide mesh in the middle that controls everything and now to go back to our previous example you can now program your photonic transmitter receiver your transceiver into this kind of circuit by just connecting everything together like connecting the modulators correctly and then linking all the inputs and outputs correctly to the right fibers and then the other fibers directly to the photodetectors and that way it gives you your barrel single mode fiber transceiver but you can just as well configure it as a coherent modulator by splitting the light up routing it to four modulators and then bringing everything together to a single output fiber which gives you coherent transmitter and the nice thing is that you have on the same chip you have that laser you can use that as a local oscillator for your coherent receiver as well and that gives you in the same chip a coherent transmitter and receiver and same thing again because in such a mesh you can implement wavelength filters you can also do this wavelength multiplexing to basically do wavelength division multiplexing transceivers so in this case four wavelengths four laser lines combined together and at the same time at the output also four laser lines of incoming light which are then split up and routed towards the photo detectors and it doesn't stop there i mean this is a chip that can do more than just being a transceiver first you can use it as a switch eight inputs eight outputs and essentially just connecting them together however you like connecting any input to any output you can use it as an optical beamforming circuit where you split your light into a channels each channel goes through their own modulator and then they go to their output coupler and by tuning the modulators you can now basically direct where the output beam is going or and this actually brings me back to the very first example that i showed you can use it as a microwave processor where you have your radio frequency modulation coming in and then essentially your local oscillator you combine it in your radio frequency output and yet now you can implement the filter function right in the middle to correct the imperfections in your radio frequency signal so it's quite diverse type of functions that you can do and you can imagine that you can use this kind of chip in many many locations i mean it could be used as a top of rack switch in a data center or a programmable transceiver in a data center those are the kind of obvious communication applications but you can also imagine that essentially it it's it's served as the hub for a building where you have your fiber link coming in and then the signals are distributed to the different subscribers and even to adsl subscribers because the chip can also handle microwave signals and a chip could like that could also be the hub for 5g antennas feeding the optical signal directly into the microwave antennas it's also very well possible to use this as a sensor hub for instance a readout for fiber sensors that monitor the structural integrity of a building or a hub in automotive sensors more and more trends to go towards optical sensors and optical communications in cars and while we're talking about optics and cars it could also serve as the engine of a lidar for instance for all self-driving vehicles or as the engine for a very lightweight microwave radar that could be patched into a drone it could be it could be used for all kinds of sensors for healthcare like bio biosensor readouts uh or optical coherence tomography tomography to find cavities and teeth or inspect a retina of your eye and finally one of the interesting functions that's coming up more recently is tricks like optical computations like hashing functions or blockchain algorithms or even quantum functions like quantum key distribution so these programmable circuits could be used for a lot of these applications which in themselves might not warrant the fabrication of their own dedicated chip now does this give us the one chip to rule everything to make everything work well yes and no it has the same drawbacks as a general purpose electronic chip in the sense that it's typically a large chip that is has higher losses and higher power consumption than a dedicated chip but on the other hand there are some cost advantages as well because you now have a chip which would be fabricated like as an electronic chip and a photonic chip together in the traditional chip model the entire cost of the development would be paid for by the customer which then turns that into his product so that means that the entire chip fabrication cost is all for one party now if you have a general purpose chip you can imagine that there's a lot more customers that would be interested to use such a chip so the cost for such a general purpose chip or programmable chip is distributed over many more customers so that means that all the costs that you have to implement this a lot of the costs are actually non-recurrent engineering which can be distributed over many more parties of course you still have quite some proportional costs but if you look if you put this all these costs in a bit of a model a very simple model you see that for three different scenarios of chips you indeed see that the programmable chip is more expensive to make because it's larger it takes longer to design and large and it's more complex to package but if you can fabricate it in a much larger volume you end up with a per chip cost that's a lot cheaper because this is a logarithmic scale that's a lot a lot cheaper than the asics and that means that if you want to do specialty low volume applications or you want to be quick to the market such a programmable chip can really be a game changer and so for small volumes or for short development times this essentially takes the supply chain between wafer scale fabrication and new product and cuts it in half and cuts the lead time from five months or a year to three days so essentially it gives you like off the shelf standard programmable chips okay so that brings me to the end uh i hope i didn't overwhelm you well me with too much information programmable the message is that programmable photonics can really be a game changer it's still in early early days for this field but the ability to have flexible and programmable optical functions that you can prototype right away in software is could really revolutionize the way that we use photonics there's a bit of work going on there you need you need not just the optical chip you need the electronics you need the algorithms and a lot of researchers still needed to fully complete that technology stack but we're getting there and the benefits could really put photonic chips in the hands of a much larger community now if you want to know more about this there's an entire book written by jose carmani and daniel perez of the technical university of valencia there's also a review paper that i wrote together with several people including david miller and jose carmani and daniel perez about the field of programmable photonic circuits and what is needed to get us to the point where we can actually use these chips so with that i would like to thank also my colleagues who did quite some of the work and i would be very happy to take some questions if we still have time for that well thanks very much i'm happy to open this up now for any questions that anyone has you can unmute to ask a question if you want so

Channel: Photonics Research Group - UGent-imec

Views: 15,696

Keywords: programmable photonics, silicon photonics, stanford, photonic integrated circuits

Id: 3-Ci3xgLbHw

Length: 54min 52sec (3292 seconds)

Published: Thu Apr 01 2021