Silicon MEMS + Photonic Systems

No online simulation and more for nanotechnology all right so good afternoon everybody today we’re really happy to have with us professor Sunil body from Cornell University Sunil is a very well known researcher in the field of medicine photonics and he got his bachelor’s and PhD degrees from Berkeley that was in 1998 in 2004 so he’s been with the Faculty of Cornell University since October of 2004 and he has received a lot of honours in the works I’m just gonna mention the NSF Career Award in 2007 in the DARPA young faculty award in 2008 a lot of his students have also received a number of awards actively photonics ultrasonics and IDM of this bill the conference’s that stimulus work has been recognized 50 students so we’re very fortunate to have him here and so thank you also for attending and let’s give a warm welcome thanks Dimitri is great to visit I’m told that last week was really bad to come to Purdue but this week is supposed to be nice and it is fantastic today I’m going to talk about the work going on in my research group at Cornell University on trying to combine micro electro mechanical systems MEMS with silicon photonics and what we are trying to work with them you know just to give an idea when I first started at Cornell I started working on out of micromechanical resonator structures trying to make as a lighter filters mechanical filters vibrating transistor devices some interesting MEMS resonators plus RF MEMS switches for sort of phase or a applications and this is because I came from that background it was the low-hanging fruit that I could attack immediately but as we start you know you start getting into the academic role and you start expanding and finding other applications of your research so I just put a collage of some of the current projects that are ongoing in my lab first I’ll talk about the projects that I’m not gonna focus my talk on one of them is trying to couple MEMS resonators with spin torque oscillators spin torque oscillators are micro oscillators that have been shown to have octave tuning range but maybe they are not so very good in terms of jitter or phase noise characteristics and so maybe we can couple micromechanical resonators with them in order to make some low phase noise yet octave tunable oscillators a second project is a little bit more far-out is trying to a couple cold atom systems with micro mechanical devices such as this 3d goblet here and this is mainly for the purposes of fundamental study of interacting atoms with membranes or MEMS resonators but also with the eventual goal of trying to make a chip scale sort of inertial sensor out of this system the third one which we already got some great results out of is trying to couple MEMS resonators here than aluminum nitride transducer with nitrogen vacancy defect centers in diamond and this is for manipulating the nitrogen vacancy defect centers to try to manipulate qubits in the diamond systems there are some applications on that as well including spin squeezing or state control using mechanical strain so all these are as you can imagine far out topics that are fundamentally driven and funded by NSF on the other hand the tropic I’m going to talk about today involves a very mature field of coupling micromechanical resonators with silicon photonics to try to demonstrate low phase noise up to acoustic oscillators and so I’m very happy to talk with you guys afterwards about these topics but today I just chose one of them and focus on this particular one here so before I start I want to give credit where credit is due everything that I’m going to talk about today is the work of these for graduate students they they really put in the time and energy in order to make these things happen and you can see some of the monumental work that went into not just a design and testing but more importantly the fabrication of some of these after mechanical systems what was so let’s start with defining the problem the problem is a very simple the problem is we have a MEMS resonator here is a picture of a comb drive which was first probably demonstrated by Bill Tang and Roger how at Cal in 1989 it involves a shuttle a proof mass and a shell that is say in this case vibrating up and down along the vertical direction you have a pad down here that where you can apply

AC signal and if that AC signal is at the right resonant frequency is going to sort of ping this device and this device has very high mechanical quality factor so it rings up as it rings up this sort of capacitor here is going to change its capacitance and you’re gonna get emotional displacement current coming out of that capacitor that will be measured out of this pad 1 here so the electromechanical transfer function that I’m trying to measure is this eye out over the input voltage signal and that’s going to look like some kind of a bicoid response where it’s a very high resonance which corresponds to the very high quality factor of this mechanical system and when you operate this comb drive I’d say 20 kilohertz or so you get extremely large signals because you get very large motion and that gives a very large displacement current out of this sense capacitor so you get very large signals out of this now as you scale this device let’s say from 20 kilohertz you say 200 megahertz may be going towards 2 gigahertz a lot of things go in the wrong direction first of all the structure starts becoming very small it’s no longer 200 microns but more like 2 microns in dimension what that means is it doesn’t really move as much it’s mechanically extremely rigid or very stiff so the actual signal that my displacement at a sensor can measure here is going to be significantly reduced so that sort of shows itself as sort of this peak Heights reducing as I scaled my resonator to higher frequencies an even bigger problem occurs because has this structure becomes very small you’ll find that this three and this bad one start coming very close to each other essentially the drive and sense transducers are very very close to each other and the shunt capacitance between the drive and sends passed through the resonator under the resonator over the resonator it is going to start dominating and at high frequencies that capacitance is very large or the impedance of that capacitance is very small so the feed through current the current that I’ll get simply due to those fringing fields is going to increase as I go to high frequencies at around 100 megahertz or so you’ll start seeing the typical micro electromechanical systems of the feed tube signal can actually mask out any of the electro mechanical displacement signal that we are trying to measure out of our system and this system doesn’t really work as an electrical system anymore and this is one of the biggest challenges that we faced as we’re scaling our devices to the gigahertz regime so why do we persist with this because some people will say well stop doing electrostatics work with body forces bulk forces piezo electric forces and they will give you much larger forces which will give you much larger displacement currents and you will get much larger signals and you’ll avoid this problem but the advantage that electrostatics gives you is that it’s actually an actuator at a distance it gives you this independence where the actuator is different than the resonating element and so it allows you to sort of design the two of them independent of each other you can choose a resonating element to have extremely high mechanical cue for example without having to worry about how am I going to attach a transducer to it and this is what allows us to get very high mechanical cue and that’s important because in the end what you are after is the frequency cue product the FQ product which essentially determines how good of an oscillator I can build using this system so we are we wanted to stick with the electrostatics but we wanted to somehow get rid of this feed through signal problem that was sort of facing us there are some electrical engineering techniques that we can use simply using differential circuits or trying to have shunt paths sort of grounded out shield electrodes and so on and so forth and we tried all of that and eventually we said we’re sort of running out of all possible ideas and it was around that time that I was walking by the hallway and I noticed that Mahalo Lips ins group in Cornell who works on silicon photonics there are pictures SEMS of her devices on her bulletin board that looked rather similar to some other ring resonators some disk resonator devices that we were building in our lab they seemed to be made out of soid wafers that 2 micron silicon there are some peroxide layer there were some time to etch release that was happening and it was her like what if we take a mechanical resonator and couple it to a photonic resonator so the idea we pitched forward or too dark but it was part of the young faculty award program was saying I’ll make a MEMS resonator and in the same layer couple of the photonic resonator I’m going to drive the MEMS the way I know it using RF signals and air-gap electrostatic actuators but I’m going to measure its displacement not electrostatically but optically and the way I’m going to do that is I’m gonna have a waveguide

going by this optical resonator here and this light is going to be coupled to this optical resonator and this device vibrates the light is going to get modulated intensity modulated by the vibrating frequency I can have a detector at the very end that sort of demodulates out the mechanical signal and I can measure the overall electromechanical response of the system to bring everybody on the same page let’s see how individual elements in this system operate first one is a MEMS resonator it’s actually very simple it’s essentially a disc that is vibrating it’s dilation or mode it’s like your eyeball ray it’s like dilation of your eyeball which is expanding and contracting the mathematics of it is actually very simple it’s basically given with the Bessel function mat you can find out all the radial mode of vibration of this thing right and they’re very close to just being simply one gig ours to regard so on and so forth and so that was straight fire on the men’s side one thing you notice as the dislike rates radially it’s also changing its circumference so that becomes important in the next part of the site of the device which is the photonic resonator in the photonic resonator as a disc expands and contracts or dilates its circumference is changing what that means is that the optical path length that the photons are traveling along the circumference of the disk is changing it’s getting modulated so if I had some light being swept through and I you know couple this light to this stationary disc I get a transfer function where the light couples in and does not couple in it couples in when integer number of wavelengths can fit in the circumference of the device otherwise it doesn’t couple in so I can do is I can simply maybe expand the device so a pseudo statically I just expand the device so I can sort of get my laser in resonance with the optical mode or I can contract the device in which case my laser will go out of resonance with the optical mode and if this device is breathing radially vibrating then it essentially can do this back and forth so I essentially get intensity modulation of any laser that I put in at the output of this waveguide so once we have these two systems in place now let’s work in fabricate this thing seems pretty easy enough and Serratia was the first student I said well let’s do this how hard can it be we already know the MEMS the photonics group has already figured out the photonics let’s make this and so we came up with a process flow so just you know this is the Nano center so I think everybody would appreciate some of the good work as other challenges that we face here we start with an sy waiver mind you it’s a custom-made SOI wafer that’s a three micron bodied oxide I think if you’re doing photonics you’re probably aware of it you oxidize the silicon layer to form your hard mask and this is your electrode pattern those are the MEMS electrodes the MEMS resonator the photonic resonator and the waveguide so I’m just transferring that pattern into the photoresist that pattern gets transferred from the photoresist onto the silicon dioxide hard mask and then I transfer that hard mask pattern into the silicon device layer using a chlorine IC pH go a little bit further to expose areas of the Buried oxide and here comes one key step which is maybe you guys are wondering about it already but we’ll protect half of the device we protect the photonic disk in the photonic waveguide with take photoresist and didn’t do an ion implant the ion implant will allow the electrostatics to be heavily doped and electrostatics require heavy doping because we need free charges on the structure to move back and forth at generating the actuation of the device but if you have any kind of free electrons or heavily doped silicon on the photonic side the electrons or the doped silicon can potentially a sting called extinction of the photons earlier effectively affecting your photonic performance so this becomes an important step and once you do this you then strip off that blank photoresist mask put on your real sort of release mask here which is the square window and then do a timed BOE release that will release the oxide from leaving the MEMS structure suspended seems reasonable enough it took us about two years to realize this and here’s the RF electrodes here is the MEMS resonator right here there’s a mechanical coupler here’s a photonic resonator and that’s the waveguide that is going by no so why two years and so that we’re no longer just making a MEMS resonator or just making a photonic device the requirements are a little bit different for both and we have to get the yield right and the maturity of the process right such that you get 100

nanometer air gaps here you get a beam that is very well defined you get 130 nanometer air gap here you want to make sure the waveguide is designed in such a way that the photons go right through without leaking into the anchor so the anchor has to be optically preventing any photons from leaking out but mechanically should be robust enough to suspend the waveguide so you have to design an MMI structure there and last but not least and the e-beam guys you have a different feel pattern a lot of empty areas here they’re not many empty areas there so you have to do some correction techniques in order to correct compensate your mask and that went some challenges you can see the edge holes need to be properly designed because the time to edge to make sure you have an oxide pedestal anchor left behind you still are left with some white residue as you can see here which took us awhile to figure out as well so once you figure you have this thing done then you know maybe how does this look to give you an idea of size scales the actual device is probably around you know 20 by 10 microns or something like that it’s very small but we are doing microwave photonics so we need to have proper microwave routing you have your GSG bond pads on the mem side you have your optical grating couplers on the optics side and how do we test this we basically take our RF probe station and you replaced one of the RF arms with a fiber optic arm so we just have a silicon chip and just like we test our RF components we just step through and measure the performance of our device so you just RF signal coming in optical signal coming the other side so everything I show is measured at room temperature and pressure unless I specify otherwise for us the big learning curve was not the MEMS about the photonics so we want to make sure the photonics works properly and so here is the transmission response on the optical resonator you can see the broadband sort of filtering function of the grating couplers they act like sort of low Q filters and in the middle of it you can see the optical resonance Peaks these are the resonances of the optical disk if you zoom into one of them you find that we get a pretty reasonable optical quality factor around 50 thousand this was a concern because if you look remember on the SEM there is this stub here and if you have photons going around in a circle on the very edge it might be a scattering point for the photons if actually hinder the quality factor but we found actually there are particular optical modes that are just deep enough from the edge where that particular coupler does not really affect the performance of the device so it’s about choosing the right optical mode and so if you model it you can see yeah for particular wavelengths you will find that the effective refractive index of the disc now it matches the effective refractive index or the waveguide mode and this is where you’re gonna get critical coupling between the waveguide and the disc resonator for the EEM guys’ really I’m just doing impedance matching between my input electrodes and my resonant cavity right so we have this in place and so now we want to test our electronic or optomechanical system hook it up with the network analyzer we have an RF signal going into the device this is the electrostatic drive and so we have a DC bias to linearize the actuation when this structure is actuated this will vibrate that will vibrate if a laser coming in is going to couple this photonic resonator come out then the photodiode will essentially downconvert give me back the RF signal and I can put the RF signal back in the network analyzer it’s a simple two-port measurement system and once we got it everything right we found that this thing works like a charm this is the effective Network analyzer transfer function essentially the gain of the measurement from the input RF to what’s coming out of the photodiode you can see the mode of vibrations this is a dilation or mode of the two disks there’s a in phase mode and out of phase more of the two disks so you can see sort of two peaks here you can see the second harmonic affair the third harmonic of it fourth harmonic you can actually go keep on looking at it until the tenth harmonic or so these are the modes that we were targeting there are some modes that you are surprised to see this is but you know this is the wineglass mode which has certain applications later on that I’ll talk about there are these low frequency modes and initially we were wondering what are those modes coming from it turns out it’s actually the floppy waveguide has its own vibration modes that can also be seen in the device and so what we are really built is an incredibly sensitive chip scale interferometer of our mechanical resonator the folks in the audience who MEMS people already know for Polytech interferometer this is really just a

Polytech on a chip we are really just measuring all the mechanical modes of vibration of our device right in proximity to it as a MEMS or oscillator designer when I see a transfer function that only requires 30 DB of loss and I can order a 30 DB amplifier from many circuits and try to build an oscillator out of this and so we get rid of the network analyzer by a low noise amp for all the cabling you need to get rid of it so you put a phase shifter to adjust for all the cabling and then close the loop and make your oscillator out of this simply and once we did that we were able to show ya we can see very clean-cut oscillations that depend on the laser power the DC bias of the actuators you know two tweaks one thing we did fix was in the previous transfer function I showed there were two resonances in phase and out of phase resonances and we have clever design techniques in order to mitigate or attenuate one of the modes while enhancing the other and you can see clearly here that’s one of the modes there’s the others of the oscillator safely locks to that particular model so we have the oscillator it’s a sort of with this random frequency 237 and so the challenge to the student was let’s make a gigahertz oscillator and so going up the learning curve now things are being stamped out of the cleanroom which must faster than two year kind of timeframe we swap out from a tethered anchor on the left which obviously DQ’d our performance to a balanced anchoring system that allows us to have very balanced high q mechanical resonators instead of disks we are going with the Rings now because in this case the Rings optical modes and optical properties are defined by the outer radius so we can choose whatever outer radius you want on the photonic side on the MEMS side the mechanical resonant frequency is dictated by the width of the ring so you can choose your inner and outer radii to be whatever you want but you can sort of basically decouple the mechanical design from the photonic design and once we do this we can start designing resonators that are much higher in frequency and so we did that and then we were able to show okay now we can close the loop and start making gigahertz kind of oscillator devices and this is where I called up my DARPA program manager and said okay I have something that that definitely works and we have something that has a very good phase noise so I said well good in a phase like that Cornell is not good enough for me this has to be verified by somebody else who is a higher authority and part of the problem with this photonic systems is they’re impossible to move out of a lab but our system will actually portable enough we just packaged it up mailed it to NIST and said okay NIST can verify the performance of our device and so we were able to then demonstrate actually their measurement was better than ours or rather their measurement system was better than ours and so you can see all the close to carrier signals are much cleaner in the mist system than our or noisy spectrum on a spectrum analyzer system but we were able to prove that we have a portable enough system but has decent phase noise and we can start using them for other purposes as well so what happens next well we try to improve on the phase noise that we just demonstrated there are two immediate things that we can do one of them is we need to improve the transducer in the sense that we need to reduce this gap or come up with the way to reduce the impedance looking from the electrode into the device the second is as we scale to high frequencies the intensity modulation that we are getting out of this photonic resonator well doesn’t move as much so we need to increase the optical quality factor of the photonic resonator so that we still get complete extinction of the light as the device vibrates so you want to improve the optical cue of this resonator as well being an engineer you want to separate the two problems attack one problem in one way solve the other problem the other way then eventually merge the two solutions together so step one is reduce the impedance from the mechanical side in order to reduce the impedance from the mechanical side you can always go with an array of devices an array of devices means you get larger trends do area largest transducer area means I get more oomph more energy transfer from the electrical domain into the mechanical domain we took this idea from Professor Clarke neurons paper here at transducers 2009 where they use a mechanical lever system to couple a large array of MEMS resonators to another resonator at the very tip remember in the optics side I want very large change in circumference on the men’s side I want very low impedance very good power handling so essentially we we could build an array of MEMS resonators to get our employ

impedance and then lever it out to very large displacements on the end resonator that will allow me to get very large optical or rather extinction and they forget a very large signal-to-noise on the optics side so we built exactly that and so here’s a system of seven MEMS resonators on one side balanced out to a single photonic resonator on the other side and you can still see the waveguide going by a couple of things you can see the actuators are only around the center for not around every every device there and that has to do partially with the routing challenges that a single layer of SOI will provide you there are certain reasons why this might be there one of them being momentum balance mechanically are to be momentum balance to get high mechanical Q out of the device so various considerations that went into play in order to make this device the other thing we found is that as we scale the devices to gigahertz kind of frequencies our optical time constants and our mechanical time constants started coming out to me in the same ballpark of course they’re running at a gigahertz so the mechanical time constant is pretty straightforward isn’t about a nanosecond or so now the photons are circulating in this disk the photons are at around higher and 70 terahertz but they have a queue of like 50,000 to 100,000 or something like that so they’re in there for a long enough time and they might be in there for a long enough time that the mechanical device starts coupling dynamically to the optical performance so the device and so what we thought was well if that’s the case then in addition to intensity modulation of light we might actually be measuring some phase modulation some frequency modulation of light and so that was interesting so we hooked it up we draw the same set up but now we put up a fabry-perot in there to see those side bands are we seeing those phase side bands and lo and behold yeah we we were able to see mechanically generated side bands to a pump that is one gig ARDS away from the actual pump this is a silicon device that is mechanically causing phase modulation of light more importantly we can actually just generate photons on one of the side bands by adjusting the detuning of the device now granted the efficiency of conversion is still a work in progress but just showing the first proof of concept was actually pretty exciting and so we push this as a new type of program you know so you can try to increase this number a little bit the other thing to do was to take advantage of this kind of spectroscopy that is available now to us readily and so you know here I plot out the mechanical spectrum from DC to like 10 gigahertz and you can see the whole thing light up and so then you get console is there and single crystal silicon so everything is very predictable so you can run console figure out all the mechanical modes of vibrations find out which families they are from whether the radio family or the wineglass family and map them out and measure the quality factors of individual modes their harmonics the mode families so on and so forth now why would we want to do that is because then you can do your fundamental study of F Q versus frequency now if you have a particular mode family let’s say in this case a radial mode it seems like it’s topping out sort of around 1 in 13 on the other hand the wineglass modes seem to be continuously increasing as I go to higher frequencies now the circuit designer what this tells me is if I choose a radial mode of vibration it really matter whether I design a 2gig arse oscillator rs7 gigahertz oscillator it’s probably going to give me the same phase noise the same ballpark at least on the other hand if I choose a wineglass mode of vibration and if I actually try to try to get this one here I will get much better phase noise out of this device then say one of the triangles down here so actually pays to go to higher frequency with these devices so that’s what this kind of spectroscopy gave us an insight into and of course now what we are saying is we’re going to switch from radial vibration modes to wineglass modes and try to lock an oscillator at 10 gigahertz at those frequencies and try to take real advantage of this FQ scaling so what are we doing today we want to make better and better devices and higher and higher frequencies we started with rings but maybe discs are better in terms of packing density dye space constraint we got some new e-beam that can design you can use it to make some very tiny gaps in our structures we use LD Illumina in order to reduce the gap sizes which will further reduce our electromechanical impedance and also surprisingly it acts as cladding and

therefore improves our optical losses as well we start becoming professionals in RF testing we switch from platinum bond pads to real gold bond pads so we can really last but not least maybe switch from a santech cavity laser to a real low noise fiber laser that so you make a list of things that are slowly go through this and first things I would like to show is the partial gap coating process so you have the MEMS resonator and if you you know if you look at from the electrode looking into the device the transducer efficiency is simply given by this equation here what you can see here is that that gap square becomes really important if I make this gap 10x smaller my transducer efficiency goes up as 100 so it’s definitely advantageous reduce a gap and so we said how can we introduce this into our process flow so we start with our release device remember we want the Ald coating everywhere so we have to have a release device and then with the release device we take this and put it into our Ald machine so now there is Ald everywhere it coats the gaps close the device’s makes this gap smaller makes the electrode to gap smaller as well so you know getting better transduction on the electrostatic side two challenges remain one is well wait a minute I had metal bond pads and now I just quoted it with Ald and so how am I going to open up that contact again and so you can’t really spin photoresist on a release chip because it’ll break off and fly off and so we just spray coat the chip with photo so we have a spray coder and we’re just like literally spray code for micron thick for others on the device the bond pads are relatively big compared to the dam that is shown here there were 100 100 microns so now you can do lithography on this to try to open up this area to remove that coating but to our surprise what we found is that when we did the developer it turns out the developer actually has a little bit of T mAh and that is sufficient to actually etch away the alumina so we were worried that our chip will then have to somehow go through some wet chemical processing or some dry RIE etching and it has all these release structures in there that could be harmed but actually the developer itself is acting as an etchant to sort of expose our bond tab and so we were very cutely able to show this coating everywhere the aplex is coated the MEMS is coated the gap is shrunk and we have an exposed Mountain sort of get everything and once we do that you can immediately see the performance improvements you can see the black line here is showing the mechanical resonance is that the spectra that this particular system picked out and you can’t really see any resonances after say 2 gigahertz but once you have the Ald coding it improves a transducer efficiency and now you can see resonance is much higher so yeah definitely this coating helped and to be quite frank it was that like release of that one Parrish was actually more key than any of the performance benefits the other advantage was on the optical side and so we simply put you know light in and try to measure if this Ald is going to degrade our optical performance it turned out it didn’t the optical cue before and after the L decoding was pretty much exactly the same so our optical cue was not affected if anything for some reason it seemed to have improved our bandwidth of our grating couplers we will take that any day so if you’re very happy that the LD did not hurt our optics at all in fact now we have some evidence to show that if you use it right at the right thickness it will actually improve the optics as well the third thing we did is you don’t always make sure that your critical couple between the waveguide and the MEMS resonator or the optical resonator if you don’t have critical coupling then you want to get proper coupling of light and the e-beam has enough of a variation that there are certain gaps at couple and certain gaps that don’t and so you start using a MEMS actuator to sort of control that gap and by applying DC bias voltages we can either a couple light in or not couple light in so we sort of have that extra little tweak so that every device in the array couples light into the resonator so with all these three handles then we are able to now start making oscillators that work around two gigahertz or so but now we can start seeing these harmonics all the way going to about 12 gigahertz and we have pretty good phase noise all these devices up to sort of x-band kind of frequencies so what’s the next step well of course is this power out here is very small there are minus 30 DBM we want to try to get it to around zero DBM so trying to get more oomph out of the

oscillator or improve its figure of Merit so here’s how they’re doing it you can start seeing the power of the fabrication technology as well as the student as they mature and start making more and more intricate devices here’s an area of telling 25 devices a couple to one optical resonator at the very end there if this is enough then maybe you make an array of 100 this really shows how far we’ve gotten since sorry this is four years from 2008 to 2013 this is a millimeter of a millimeter in size so you’re talking about arrays that are millimeter size scale there are hundred such resonators in there zoom in each resonator is around three to five microns in diameter all of them pretty much the same diameters and then this is sort of credit to the student and the e-beam and the proximity correction algorithms that go along with it and then if you zoom in further more then you can see that everybody has a very tiny gap around 50 nanometers gap so we really can go from millimeter size scales with nanometer size scales while maintaining good electromechanical and optomechanical performance of these systems and so these are these are some of the devices that are currently being evaluated in my lab remember I said that we’ll do it to divide and conquer approach and so we showed you a bunch of array pictures and this is work in progress the second part was can we improve the optical cue of these devices we found out that silicon pretty much has some kind of a limit practical limit associated with it ali-a DB’s group at Georgia Tech has shown queues of like million but we could not reproduce those kind of results at Cornell on the other hand silicon nitride devices have been shown by various folks including folks here at Purdue of showing extremely high optical cue in the systems and at Cornell we have a lot of inbuilt expertise and very high mechanical quality factor silicon nitride devices so we said okay let’s maybe silicon nitride as a material of choice other interesting properties include working at the wafer scale order than at the chip scale and you know using regular G line with ography tools instead of a beam so there are a lot of benefits to this so we started with making silicon nitride as a platform so you start to the fabrication process where you start with depositing of silicon nitride on the 4 micron thick thermal oxide so you have your red silicon nitride this giant thick layer of oxide you have your mass you transfer your mask to the nitride and then you put the PECVD cladding this allows the light to be well confined within the nitride once you have that cladding you open a window where you want to release the photonic device now this is a key step I could have taken this device and put it in BOE to release the device but it will undercut laterally a lot so we sort of h back as much as a little bit down so we have a smaller time of release and that becomes a very critical step so that when we do a timed release at the very end the waveguide sections and the other sections of the chip are not released not attacked by that HF and only the mem structure is released so we have something like this seems simple here the picture of it is a silicon nitride ring you see the pedestal anchor with four suspensions there’s a waveguide coming in trust me I would not show this picture to you if it was shorted like that that’s just charging from the SEM and simply you want to see you can make up a light into this does it really have a decently high optical q and maybe what is the mechanical performance if we can see it so that was our belief so divide-and-conquer trying to make silicon nitride after mechanical systems step one move away the RF probe tip there is nothing conductive on the device so now you’re just doing optical measurements and step one is simply to measure the optical response and yeah we were able to get much sharper optical transitions as we expected out of the silicon nitride and we got you know even at first shot we got a very high turn Q of two hundred thousand but we got very decent optical cues all the way to millions depending on the device and the geometry of the system so we had the optics in place and it was around the time that the DARPA orchid program started which is essentially using optomechanical forces for heating and cooling of optima chemical systems essentially using photons to generate forces onto the mechanical devices we had a release device that coupled light very well so of course we will try to see whether we can do Heating and Cooling with it and so yes we coupled some light into it and we got not you know you got some RF power out

but then the radiation pressure forces took over started self oscillating and we got some RF signal out so really the heating the optomechanical radiation pressure based heating effects were measured maybe this figure does not tell you everything by itself this is what we’ll tell you is at low powers low laser input powers if I just look at the RF spectrum coming out I see the mechanical brownian noise of the device the Brownian vibrations of that particular mode of vibration are imprinting itself onto the photons and then you can see it when you’re down convert the light that again tells you the amazing power of the light it’s really just an incredibly good clock with very high fast sampling speeds and I’m able to see the Brownian noise now if you choose the correct detuning there is a particular delay in which the photons enter and the response mechanically and you can choose it such that you get negative damping effects out of this so the photons will eventually start pumping energy into this mechanical mode of vibration and you start seeing this threshold leading to self sustained mechanical oscillations you delight so what was just simply Brownian motion now the photons are imparting energy to the phonons and launching self oscillations of the device so you’re able to see this threshold and in the clean cut resonance this is you know observed by other groups as well so what’s so special about this what was interesting to me was the phase noise performance and if you look at the absolute numbers not nothing special here in terms of absolute performance what was very interesting was it was 1 or F square all the way right makes sense I don’t have an amplifier in a sustaining loop that is adding 1 or F noise into the system it’s just a self oscillatory system and so it’s just gives me the 1 or F square is essentially the Brownian noise the white noise of the mechanical system is being seen and more importantly it matched Theory extremely well this was like spar on the you know we literally ran out of spectrum analyzer we had to actually get a signal source analyzer the Agilent SSA which becomes your best friend for getting these low offset measurements and we found that actually all the way to about two Hertz or so we were able to get F Square and what limits it and this is something that is you eventually do get one or F and that coming from the laser it’s not coming from the mechanics it’s just coming from the laser you don’t have to just oscillate radial modes this was the example of a radial mode of oscillation but you start seeing some other interesting modes of oscillation as well here’s an example of one which is a sort of a wine glass or rubber band mode and there are a lot of MEMS applications that are more interesting from this more standpoint and the radial breathing modes that you’re oscillating so we want to use our sort of MEMS tricks to sort of control which ones oscillate alright so because we want to build you know people to control and so we change the suspension design off say in this case these two couple devices where we can choose via the chillers is not by luck we can choose whether we oscillate the radial mode or the wineglass mode and not oscillate the other all right so if you’re making a clock you probably want the radial modes if you want to make a sensor you probably want the wineglass modes and so you can pick and choose which modes you want and so that that’s what are the exciting and you know now you’re pushing this a little bit further ahead so I assure you oscillators the first thing that you can say like well okay if you have an oscillator and I come from the inertial sensors backgrounds like well if it’s a it’s a it’s so sensitive displacements I should be able to just make an accelerometer with it and so methyl Ibsen’s group as well as Oscar painters group had shown this optical disk stack resonator was essentially an optical super mode between two photonic resonators this with a silicon nitride stack or a silicon stack you can do both what we did was insert releasing both these devices sort of just kept the bottom and anchored so just one anchored and the second one is free to move and that becomes sort of the proof mass that the accelerometer proof measures just depending on the acceleration will either move up or droop down and that changes the gap between these two rings that changes this gap that changes the wavelength of the super mode that’s sitting between these two rings and so we fabricated this device we did have a waveguide coupled into it or using a tape or fiber but by just flipping the device up and down from one g2 plus what plus 1 j2 minus 1g we were able to get sort of a change in the modulation index so if you have the same laser you will

either get some light out or a little less or a little bit more where it turns out actually you get quite a bit of modulation around 22% 4 + minus 1g just to compare with electrostatic accelerometers that are out there these days you will get something around the order of 2% modulation 4 + minus 1g so not surprisingly you get about a 10x improvement in your sensitivity or your scale factor when you’re doing this optomechanical sensing compared to say electrostatic sensing of these devices you can say ok I can keep on pushing this further as an engineer you will say well that’s too much that’s almost too much sensitivity what I should be able to trade this off for something else something else meaning bandwidth something else when is dynamic range you know so there are creative things that we need to pursue that will try to improve its dynamic range or your bandwidth while maintaining reasonable sensitivity and so that’s what we’re exploring next so you know so we have this tool set now sort of forgot about that oscillator right we were going to plug this back into the oscillator that’s because there is no progress on that front but we have this thing it’s its own project chip by itself and so let’s see what else can it do and so you have a device that works is a fantastic accelerometer that has a pretty decent phase noise more importantly it doesn’t have long-term drift potentially and maybe it also is so sensitive displacement you can measure Brown neither the structure itself so really I have something that should be a toolset that can use to build all kinds of sensors and the sensor that most interested in is a gyroscope so we we pitched this program he said okay traditional silicon gyroscopes our additional MEMS gyroscope are made out of silicon silicon has a load from elastic damping so the Q’s are a little bit lower it has an ionized satrapy depending on the crystalline orientation typically made into DS and have this problem when 2d structures have outer plane parasitic modes of vibration the preferred mode of transduction is using electrostatics you know we have conductors we can use electrostatics what’s appealing to investigate was can we make gyroscopes out of non conducting materials like silicon nitride or silicon dioxide now why because they’re amorphous so fundamentally they’re isotropic it’s really easier to do frequency matching or mode matching they have extremely high thermo elastic dissipation so their quality factors are potentially higher like you saw in silicon nitride you can mold them into 3d shapes and those 3d shapes fundamentally have very little anchor loss so you have very high Q on the anchor last side of things and those 3d shapes also give rigidity against parasitic out of main modes as 3d and let oh wait I cannot use electrostatics but now I have light that are uncoupled to this so it was more opportunistic that I had up to mechanics working in my lab and I wanted to make a gyroscope out of these materials so let’s combine the two and try to make oMG right now my students argue that I first came up with the acronym and then came of it reasoning for it so how do the gyroscopes work maybe in the last couple of minutes a gyroscope works there’s a mechanical proof mass in this case that is oscillating back and forth along the x axis right now if this entire board feels a rotation say about the axis out of the plane of the board so in this case say Omega Z you get a cross coupling term that couples the velocity along X with the rotation of the system about Z and that’s comes out as a Coriolis force a fictitious force that acts along the y axis because it’s just basically X cross Z and that acts as an acceleration of the outer frame the outer frame deforms due to that acceleration and you get some capacitive electrodes that can pick off that motion so it really consists of two parts there is an oscillator along the x axis there is an accelerometer along the y axis along the orthogonal axis and it has the capacitive sensor hopefully that is sensitive enough to detect the tiniest of motions along the y axis and that’s those are the exactly three things that I showed that opto mechanics was capable of producing very low decent low phase noise with more one or s square kind of slope oscillators you have a pretty decent accelerometer with very good sensitivity that can measure the Brownian noise floor of a mechanical system so we have all the building blocks the problem was over a minute this is all rectangular geometry and I just showed you all very nice and circular design so well you know there there are more mechanical mode families out there this one is pioneered by

felucca aziz group at Georgia Tech where this is using a shell mode that excites uses this as your dry mode and the diagonally vibe D generate more as a sense mode I will spare you the details of how this works or if you’re interested please read who are you Joe Hardy’s PhD thesis but needless to say these two modes can also be used and you can see this is the mode wineglass mode that identified with you earlier and so what we are doing with this is building something like this which is essentially sort of a 3d shelf like I promised it’s made of silicon dioxide and now you can imagine the wait a minute where is a waveguide well that’s that’s coming and we try to make a 3d gyroscope or OMG out of this so summarize my talk we designed a 1 gigahertz oscillator that has pretty decent phase noise and we showed that yeah we can actually launch up to make giggle oscillations with no flicker noise demonstrate decent accelerometer and now we’re using this oscillator with this accelerometer principle to try to make these 3d or mg devices and I’ll conclude my talk here and take any questions