This is half of a very nice transfer function plot - See if you can get COMSOL to give you the phase information too and plot that in tandem (e.g. https://en.wikipedia.org/wiki/Bode_plot).
Also, I'm mildly curious how the frequency shifts with support distance. Are you using material parameters for the supports, or treating them as an area of constrained geometry?
After playing around a bit more with Comsol, I realized and fixed some slight mistakes. I added in the phase information. The transitions in phase occur very rapidly and with resolution I was running at I couldn't get any points during the transitions. I used a loss factor of 1e-8 for the silicon (as measured here).
This simulation runs by considering the support points of finite size as fixed constraints.
I'm going to try and check effects of the support positions.
I also uploaded my model to the svn under: /trunk/CryoLab/comsol/cryo_cavity_sagging_transfer_functions/
I added the ability to loop over the support separation. I checked values from .5in to 3.5in. All of these were with L=4", R=1", and at 60 degrees.
The first plot shows each different support separation plotted together. The remaining plots are the bode plots for each individual support separation. Up until aout 10kHz, the transfer functions are pretty much the same besides the slightly different gains at low frequency.
I also committed these changes to the SVN.
Try fitting a function of the form:
I used the matlab function lsqnonlin to do this fitting. This function fits real valued functions so I had it fit both the real and imaginary parts. The first plot is done with tau=0 while in the second I allowed it to solve for tau. I used the DC gain as the initial guess for G, 1e8 for Q, the location of the maximum for f_res and 0 for tau. The version with non-zero tau would not converge on the default settings so I played with the parameters until it converged. Both solvers also ended up giving a warning that the value they converged to might not be a true solution. I tried playing around with the settings of the solver and it ended up giving results that were very clearly off. I might try to play around with it a little bit more later.
The fit with no tau agrees pretty well up until the higher order resonances kick in.
The fit with tau also agrees pretty well in magnitude for the first resonance however the phase is all over the place. As I said, the solver had issues with this so I might try a few things to make it work better.
Between the two fits, G and f_res stayed fairly constant between the 2 as expected. However Q decreased by an order of magnitude in the solution with tau.
Wrong: G / (1 - i f / Q - f^2 / f_res^2)
Correct: G / (1 - i f / f_res /Q - f^2 / f_res^2)
Relationship between Q and decay time
There is a way to do the complexfitting by converting C into 2xR
i.e. Conver a data set from
a1 + i b1
a2 + i b2
a3 + i b3
an + i bn
and instead of the complex function f(x)+i g(x), use the 2D fitting function
z(x) = kroneckerDelta(x,1) f(y) +kroneckerDelta(x,2) g(y)
Then the 1D complex fitting is mapped to 2D real fitting.
In any case, your transfer functions might not have sufficient resolution to allow us to do the Q fitting.
I tried again with fitting the correct transfer function. The fit is pretty much the same except with a slightly higher gain and resonant frequency. However, I noticed that the solver wasn't changing the value of Q. In the plot below, I initially guessed 5e7. I tried decreasing the initial guess of Q, but the solver wouldn't change it at all until I got to around 1e5. So as Koji said, there's probably not enough resolution to get a good fit for Q.
With the correct transfer function, I couldn't get the solver to converge anymore with a tau parameter I guess because the function it is trying to fit is slightly more complicated now.
The approach with changing a complex valued function to a 2-dimensional real valued function is what I have been doing for the fitting. I guess I didn't make that too clear in my original post.
EDIT: Evan pointed out to me that the complex pole pair should create a phase lag of 180, not a phase lead. I had used a Hermitian transpose (') instead of a standard transpose (.') on the COMSOL data which flipped the sign of the phase. For the fitted data, there was a flipped sign in the transfer function. It should be: G / (1 + i f / f_res /Q - f^2 / f_res^2). I updated the below plot with this in mind. Changing this didn't affect any of the parameters as both the data and the fitting function had reversed phase.
I investigated the sensing noise issue a little further. I measured the noise on the error signal without light on the PD, with light but no sideband modulation, and with sidebands. In both cases there was a total of 3.5 mW of optical power on the PDs and I confirmed the cavities to be far off any relevant resonances.
Obviously scattered light is still a HUGE problem. The figures show the 'quiet' state when the floating table has been left alone for a couple minutes. I'm going to change the hose on the vacuum pump to a more flexible one i bought a while ago and see if it gets better. All optics are already on 3/4 inch posts. We should have the new super-windows soon, which will be a nice feature. Using a buzzer might tell us where the scatterers are.
Other observations: 3.5 mW are enough to give the shot noise some clear separation from the PD noise. The noise levels seem to be a little higher in the East path, which is due to the resonant PD having more transimpedance gain.
I added all the sensing noises (ignoring scattered light) together for the two paths, where I assumed ~80% visibility in the cavities for the reduction in light power on the PD for shot noise.
We're not as clearly limited by noise in the LB1005 as I thought. In the next few days I'll measure the beat noise with better resolution and try to make sense of these numbers. I also still have to diagnose if something is up with the PDs and why they seem to be saturated so quickly.
I used the test inputs on the resonant PDs to see if the notches are still in the right place. I found that the overall transfer functions have qualitatively quite different shape above their resonances. It seems like the west path doesn't have much of a notch at all, and a secondary peak in its response between 300 and 400 MHz. I haven't worked with these PDs before, is this within the specs or is it likely that someting on the board is broken? Koji told me after today's meeting that care must be taken that the fast gain chip doesn't oscillate, maybe that's what's going on?
The resonance frequencies were fine, but the notch locations were off by ~1MHz in both cases. I opened the PDs and their RF cages and corrected this. It required a little back-and-forth because closing the RG cage seems to change some capacitances, shifting the exact notch locations up by up to 50 kHz. The modulation frequencies are 32.7 MHz (65.4 MHz notch) for West and 33.59 MHz (67.18 MHz notch) for East.
The resonances were pretty much unchanged during the re-tuning.
I can't see the important part of the transfer function (f>100MHz)!
But I think it is quite possible. In any case you probably need some modification of the RF preamp.
You can look at the schematics of the same PD circuits at the 40m.
(Click each PD link like REFL33)
This modification at the 40m also involves the correction of the RF output impedance, which needs to be 50Ohm.
I'm still trying hard to understand my current sensing noise floor of about 10-16 m/rtHz (= 300 mHz/rtHz) before I go forward with replacing the test cavities with the dummy cantilever cavities. Long story short, I don't have any more great ideas at the moment, so I think I will move on with the payload recplacement very soon.
After lots of work (see replied-to post), I was able to all but rule out nonlinearities in the beat readout and intensity noise as the culprit. As quoted in the box at the bottom, my next move was to investigate the PDH demod setup and locking RFPDs to see if I could make any improvements there.
One interesting thing to note before moving on to the deliberate changes below is that, for perhaps the first time, the beat had drifted to a very low frequency of around 25 MHz last night. This afforded me a unique opportunity to rule out any weirdness in the downconversion step of the beat readout, where I typically use a Marconi to step the beat frequency from several hundred MHz down to ~10-20 MHz for piping into the RedPitaya. These direct beat measurements from last night showed no appreciable differences from the Marconi "superheterodyne" method. By this morning, the beat had shifted back up to ~55 MHz, near the high end of the RedPitaya range. I made one last direct beat measurement, then switched back to the superhet readout, and verified that---again---there was no change to the beat spectrum.
Switch to level-17 mixers
Last night, I swapped out the level-13 mixers I was using until that point for level-17 units, as Johannes said he saw some noise reduction in the same sensing-noise-limited band when switching to higher-level mixers in the past. As I mentioned in the last post, I was able to circumvent the need for amplifiers on the LO signals by simply switching which OCXO box outputs were used for what. (The "EOM" output spits out +25 dBm, while the "LO" output gives +14 dBm. I am currently only using between -6 dBm and 0 dBm on my fiber EOMs (see aside below), so the large amplitude from the OCXO box is not really necessary. The only difference between these two outputs is that one comes from the "CPL" output of a directional coupler, while the other comes from the main through port, so there's no reason they can't be used for the opposite purpose w.r.t. the output labels on the box.) I am now sending the "LO" output, with 20-dB attenuation (= -6 dBm as a result), to the fiber EOMs, and the "EOM" output, with 8-dB attenuation (= 17 dBm) to the new level-17 mixers.
After sweeping the error signals and retuning the demodulation phases, I re-locked the cavities with no problem, but observed that the new mixers had no significant effect on the beat noise.
Aside: I should note that, prior to the mixer exchange, I was using a -3 dBm drive to the EOMs. During my testing last week, I had reduced the EOM drive to this level from the 0 dBm level I had been using for months prior to this. I didn't notice a difference when I did this, so I left the extra 3 dB of attenuation in. After having switched the mixers out yesterday, I tried going back to 0 dBm, whereupon I found that the noise had marginally increased. In light of this, I went to an even further reduced EOM drive of -6 dBm, and saw that this made the noise marginally lower as compared to the -3 dBm drive state. Reducing the drive further (to -9 dBm or -12 dBm) resulted in a noise level at or higher than that with the larger 0 dBm drive. So, it seems like the noise is minimized with -6 dBm EOM drive---at least with the current level-17 mixers in place.
Locking with gold iLIGO RFPDs
This was actually the first thing I started working on Monday afternoon. I picked up the spare gold iLIGO RFPD that Gautam had spotted for me at the 40m and brought it back for testing. Its label said "? POP(22,110) ?" on it, so it seems there was some question as to what exactly it was. I put it in my optical setup and ran an AM TF on it, and it seemed like it had a decent response at 110 MHz. I then opened it up and attempted to modify the TF, but what I found inside was a complete mess of flying components, cut traces, and other fun features. I tried doing some reverse engineering, but eventually called it quits that night after hours of getting nowhere.
Luckily, I had another idea: simply borrow the PDs from the CryoCav experiment for an afternoon. Even more luckily, Johannes agreed to let me do this ;-)
This morning, I borrowed his PDs and put them into the REFL ports on my table (first in the E path only, then in both). In each case, I again swept the error signals and retuned the demodulation phase with the new PDs. Below is a plot comparing the beat noise in three cases: 1) the usual state with two PDA255s for locking, 2) the hybrid state with a gold PD on the E path but a PDA255 still on the W, and 3) the state with gold PDs in both paths.
As you can see, there is no appreciable difference between the three configurations. (NB: The reduced noise at lower frequencies in the all-PDA255 case is the result of the suspension being more rung up this morning than last night. The features are a combination of more motion showing up linearly at very low frequencies (near 1 Hz), as well as scattering between a few tens of Hz and ~150 Hz.)
For the record---unlike with the PDA255s---the lowest noise was achieved with no optical attenuation in the REFL paths before the gold PDs. Whereas I need significant (OD 1.0-1.5) attenuation to get this minimal noise level with the PDA255s, adding any such attenuation while using the gold PDs resulted in a higher lowest achievable noise. This latter behavior with the gold PDs is more in line with my intuition, since the reduction of the optical gain should increase the beat-referred sensing noise. I'm not sure if this tells me anything new about the weird attenuation effect with the PDA255s.
Conclusion & plan
I'm not sure what else there is to try at the moment that justifies continuing to hold off on the payload swap. One thing to note is that, with the recent (albeit slight) reducion in noise, it is clear that there are some features in the ~few-kHz region that make the picture a little less clear. For example, the noise at 2.5 kHz appears limited by these narrowband features, and not by the broadband floor anymore. Also, above ~4.5 kHz (i.e., above the highest newly resolved feature), the noise is well understood as residual laser frequency noise. Therefore, the only place where the noise remains anomalously high is in the narrow band between 700 Hz and 1.5 kHz. Here, it seems too flat to be the result of narrowband displacement features, and the noise floor mystery persists.
In light of the above, I don't think there's enough left to learn in this configuration to keep waiting on the cavity swap, so my plan is to move forward with that. Hopefully tomorrow, Johannes and I will bond a mirror to the second dummy cantilever (which I need to clean and bring over from the KNI), and then I can open up the chamber and start with the swap.
With the dummy cantilever cavities in place, I can continue with the noise hunting once I see the new displacement noise situation. One "free" thing to try is iterating the shape of the servo TFs to further reduce the laser frequency noise. (In fact, I can still do this some tonight as a last-ditch effort before switching out the cavities.)
More stuff I'm going to try:
MONDAY JULY 3rd, 2017
I read notes on how to operate the temperature diode. I took apart an old piece of aluminum equipment to use as a mount for the temperature sensors. On the left is the calibrated temperature diode, and on the right is the diode to be calibrated. Using the drill press, I carved a groove to place the uncalibrated diode. I have a circuit connecting the positive terminal of the uncalibrated diode to the crystats feedthrough, connecting the negative terminal to the positive terminal of the calibrated diode, then from the negative terminal of the calibrated diode to another electrical feedthrough. This configuration provides identical current for both diodes. I have two circuits connected in parallel, detecting the voltage drop across each diode (with negligible loss in current). I could not test this system because I could not find the connector for the electrical feedthroughs. The diodes are secured using clips provided with the calibrated diode. Insulating grease is used to shield the contacts of the diodes from the conductive aluminum base.
I had to replace two of the optics right after the faraday (both were crummy 1064 optics that I was using to pickoff ~1% of the 1550 beam for power monitoring). They had to go.
They were wedged, so this blew the alignment downstream far enough off that I was missing lenses.
I tried to just tweak the things upstream of these optics (which are 50:50 beamsplitters) to realign to the path, and was able to recover alignment in one of the arms.
Sadly, I noticed that I was getting serious distortion of the beam through the faraday. WIthout a 5 axis type mount, I'm afraid I need to realign everything downstream of the Faraday after aligning to it, which means realigning the cavities somewhat from scratch. This is disappointing - I will work at this tomorrow carefully and well rested so that I don't have to completely scrap my previous alignment work and rebuild the table.
Had some minor setbackage:
Dewar went soft b/c it ran out of cryogens overnight.
Ran out of cryogens because there was too much gas (He) in the insulating (outer) vacuum**
Too much gas in the outer vacuum because of the honking leak between the inner and outer vacuum
**will calc a data point for hold time vs pressure from this.
Once i let He into the outer vac, it takes quite a while to get back to low pressure:=> I either babysit the cryostat constantly filling it up for the duration of this run, or open it up and fix the leak.
After consulting with Warren, I decided to fix the leak rather than brute forcing back down and trying to live with it.
Borrowing the He leak detector from a very generous Schwab lab as soon as they are done with it (possibly as early as tomorrow). Gathering quotes for our own leak detector.
Backfilling with N2 to warm up, and will open the outer seal and start leak checking the inner as soon as I get the leak checker.
The leak is now huge. When backfilling the inner vac with N2, the outer vac came up to room pressure as well.
Following our discussion earlier when we realized that the AC electronics of the 1811 may be saturating, since our RF power (at the mod freq of 33.59 MHz) is near or over 55 microW, I added an additional 10dB attenuator before the EOM. The total attenuation is now 40 dB. To compensate for this I removed the 10 dB attenuator at the input of the LB1005.
I also added a 50 ohm terminator in parallel to the 20 dB attenuator at the analog input modulation port of the ITC 502 current driver. This is to ensure a more or less accurate 20 dB attenuation of the control signal since the ITC 502 input has an impedance of 10 kOhm.
Everything seemed to lock once again when the gain on the LB1005 was increased from 5 to 6.
We should note that the LB box driving into 50 Ohm is current limited at its output (assuming we don't narrow the voltage window using the trim pots on the back panel). It can supply up to +- 20 mA, so if we see control voltages approaching 1 V we are nearing saturation of the servo controller.
To run background scripts on cryoaux, e.g. python temperature control servos using systemd: There are several examples one can use as templates in /home/controls/services/. For background tasks it's recommendable to set up a logfile, check the python scripts that are called by the existing service files to see how it's done. When ready do the following:
I finished my cryostat setup. I built a suspension for the liquid nitrogen tank of the crystat in order to make setting up the GeNS easier. I glued the RTD to a disk with apoxy, and sealed the chamber. I was able to shoot in a 7mW red laser, and capture the reflecting beam using the below configuration of optics. Once I set up the electronics, this experiment will be ready to be run.
Brittany and I have been in the lab setting up ESD-driven actuation and moving towards cold Q measurements.
Pele now gets down to ~3e-3 torr, while Nalu (the smaller cryo) gets down to about 3e-5 torr. There is still a lot of Mylar shielding and adhesive inside the cryostat; this shielding was installed by the previous owners, so presumably was vacuum compatible. We are divided on whether the outgassing is what's limiting the vacuum, but one solution could be to add the charcoal to the container, it might help the vacuum especially when we add nitrogen.
We soldered leads to the ESD and a temperature sensor on the heat shield. We were working on feeding the ESD an AC signal in a way that does not send too much current through the ESD--this will be completed Wednesday afternoon. We also tweaked the JaNS mount, so it is slightly more stable.
As noted previously, we have both uploaded changes to the drawings for GeNS on the DCC [D1600472]. We can bring them for comments to the Thursday meeting, or we can find a few people individually. We may need to make a few fixes to the assembly after the modifications before Thursday.
22 of the 25 wafers we ordered from University Wafer will ship Wednesday or Thursday. I asked that they ship the remaining 3 when available.
We also modified the COMSOL model [D1700104] with new materials properties, temperatures, disk parameters, and are getting closer to believing we can predict the forest of modes on the FFT. We think we see the lowest butterfly around 560--does this make sense to people?
Look forward to more to come!
I will add a summary of where we are with the set-up to help guide the reader
Yesterday, we were conseratively ramped up sending in DC voltages from 0-150V. We did not see any response from the disk.
We applied AC signals to the high voltage supply with square waves from a function generator. Similarly, we did not see any response from the disk.
We verified that what we were seeing was the disk by actually knocking it off it's suspension system and completely losing the return beam.
So here are the drawing promised yesterday. I hope that any decent job shop can turn these into the parts we want.
First, a drawing with the 'holes' specifications and location
and then a specification of the geometery.
7 Inch fat posts
Coat hanger shaped thing for crane
Nothing new here. The shorter the better for enhancing thermal noise. A laser tuning range of 10-20 GHz sets a lower limit for the cavity length of ~1cm, which has an FSR of 15 GHz.
A design choice needs to be made for the mirror radii of curvature, maximizing the coating noise impact factor. For crystalline coatings the minimum roc is 10cm, while for amorphous coating it can be shorter. The graph shows the combined coating noise impact factor from both cavity mirrors as a function of roc. For the plano-concave curve one mirror is held flat, while for the concave-concave cavity both mirrors are changing. Only for mirrors with roc smaller than ~1.5cm the half-symmetric cavity shows a larger impact factor. Above the symmetric configuration is to be favored.
The short length with its big FSR results in an enlarged cavity linewidth. This affects the frequency discriminator negatively and we need to assert that sensing noise is not keeping us from reaching the coating noise limit. For an impedance matched cavity the optical reflective transfer function close to the resonance is 2*df/FHWM, with df being the frequency difference between cavity and laser, and FWHM the linewidth of the cavity. Using PDH this translates to a demodulated error signal (in terms of light power) of 4P0J0(m)J1(m)df/FWHM [W/Hz]. Assuming a modulation index of m=0.2 and an initial carrier power P0=1mW, the shot noise on the RPD contributes 2.3e-12 W/rt(Hz) of sensing noise. It is suppressed by the optical gain in the cavity, and reaching a frequency stability of df=1e-3 Hz/rt(Hz) (motivated by expected coating brownian noise levels) requires a cavity linewidth of less than about 175kHz. In a 1cm cavity this requires a Finesse of better than 85,000. A high finesse requires better, thicker coatings, which further enhances coating noise.
The round-trip losses that correspond to a finesse of 100,000 are 63 ppm, so we can aim for T=30ppm per mirror. A Ta2O5-SiO2 quarter-wave stack coating requires 17 quarter-wave layers to achieve this according to my calculations. The corresponding coating thickness is about 8.5 microns. The projected coating brownian noise level, again as a function of roc in a L=1cm cavity is shown in the following graph (at room temperature), calculated with the simple CBN noise model from Nakagawa et al. from 2001, gives an idea of the expected level of the frequency noise. The parameters used are stated in the figure.
My recommendation would be to fix the mirror rocs and then fine-tune the cavity length for coating noise levels and g-factor. I found that for roc=10cm a cavity length of 1.875cm yields g=0.66, which is a good place to be, with an FSR of 8GHz. Note that the high finesse relaxes any g-factor constraints, so it is not that essential to stick with values that are considered safe. I think that somewhere around there we can find our cavity configuration.
I had an optical bonded cavity for about 2 minutes today before the second mirror plopped off during handling, but I'm confident that I can repeat the bonding attempt tomorrow with more success.
As Chris suggested in a recent meeting, using some 3D-printed parts I made a rig to hold the cavity vertically so I can rest the second mirror to be contacted on top and move it around as I search for the least lossy position. I confirmed the orientation of the mirror wedges by shining a laser through the annulus - the deflection of the transmission is pretty significant and easily identifies the wedge (deflection points to the thickest part of the mirror barrel)
Getting the initial alignment on the input side was tricky because of the significant wedge and silicon's high index of refraction. I looked for dips in the reflected light to make sure I have something to see on the camera in transmission, and then used a lens for better visibility. Once I had visual feedback it was easy to identify the 00-mode and align the input beam to it. As before, I didn't really do any mode-matching, just placed an f=10 cm lens such that the reflected beam had roughly the same size on the viewer card as the incoming beam, which means that the wavefront matched the mirror curvature semi-well. The lens was also producing a waist within ~2cm of where it needed to be. With this I achieved ~70% visibility, which was plenty to lock to the cavity.
Using one of the SiFi fiber-integrated EOAMs I modulated the laser power with a square wave and was able to perform optical ringdown measurements on the short cavity. Due to the systematic error I described in elog 2014 I estimate that there is about a 5 ppm uncertainty on the round-trip loss when using the EOAM to switch the light (a square wave step LARGER than the half-wave voltage sends out-of-phase light into the cavity, which actively depletes the cavity field, driving the loss estimate up). The reverse is unfortunately also true, if the beam is not extinguished entirely the loss will seem lower because the input light still pumps the cavity field. For every ringdown I therefore first minimize the off-state power using a pick-off PD, but doing this several times with no change to cavity alignment in between I found that the loss estimates I obtain have a spread of ~5 ppm. In comparison, for back-to-back ringdowns without changing anything the estimates are well within 1 ppm of each other.
On the first attempt I measured 140 ppm roundtrip loss, which is not far from the best I've seen before (137 ppm) and within the systematic error. I was curious though if tapping on the mirror can move it into a better position. Not really. I tapped several times, which did move the mirror (had to realign the input light) but I didn't see a significant decrease or increase (minimum was 138 ppm, maximum 144 ppm). I previously measured T=65 ppm on the HR coating, so 130 ppm are from mirror transmission.
Using my fingers on the rim of the mirror (AR coating had to be off so I could do interferometry) I pushed the top mirror down in a likable position. Applying slight lateral and rotary forces while pushing there was no slipping, so I thought it was a job well done and took the cavity out of the mount. I shook it a little and turned it upside down, the mirror held, and I placed it back in the storage mount, new mirror facing up. I then though it would be nice to get a picture that shows both mirrors attached, but when I took it back out and tilted it the freshly applied mirror plopped off, fortunately not falling far, and landing on the AR coating side, which I inspected and found no damage.
I was trying to first contact mirror and spacer and do it again, but wasn't patient enough to let the FC dry sufficiently - and it ripped while peeling. I spent the rest of the day re-applying first contact, but couldn't get it all off in one piece. In the end I succeeded, but then saw some residue on the bevel I will continue this quest tomorrow with some fresh FC from Downs.
I also heat-sonicated all the parts for the cavity-baking today using the new bath in the PSL lab. I confirmed that the complete jig holding a cavity will fit into the little baking oven, so I'll do a baking dry-run with only the jig parts tomorrow as I re-assemble the botched cavity.
The photocurrent in terms of the phase of the beat signal is:
I_PD = I_DC + I_RF sin(\phi)
where \phi is (omega1-omega2)*t.
The discriminator, current per radian, is
dI/d\phi = I_RF cos(\phi) ⇒ I_RF for \phi = 0
So a fluctuation in photocurrent causes a interpreted fluctuation in phase as:
D\phi = (d\phi/dI) DI.
Phase fluctuations are related to frequency fluctuations as f * D\phi = Df, so
Df = f / I_RF * DI
for shot noise, DI = sqrt(2*I_DC*e) (ignoring cyclostationary corrections).
so Df = ( f / I_RF ) * \sqrt( 2 * I_DC * e)
Assuming I_RF = I_DC = 1mA we get:
Df = 1.8e-6 ( f / 100Hz ) Hz/rt(Hz)
Seems low enough!
On Friday cleaning, we vacated the east optical table in QIL. The Si scatterometer was disassembled and the Si block was moved and stored to the cryo lab.
I managed to align close enough to the Si cavity to see flashes:
More details to follow!
Below is a picture (+ beams) of what I put on the table to do cavity scans. The mode matching lenses are what Nic called out in elog:603.
Cavity flashes can be seen in the video in elog:605
I spent a little while trying to get a clean cavity scan in reflection (I want to see the dip in power from the 00), but got nothing which resembled a lorentzish dip. (Noisy crap)
I scanned frequency via the laser current, using the ITC510 unit to control the diode.
I will post sweeps when I can get them off this fine floppy disk format onto my computer and matlab them into readable plots.
The relevant time constant for filling the cavity with light is:
so if we pass through the cavity at 5MHz/ms, we are sweeping through the cavity on the order of its filling time
Sweeping at 400MHz: 400MHz * ms / 5MHz = 80 ms. Since it's a triangle wave, period = 160 ms, or f _sweep > 6Hz
I do not understand the behavior, I am sure there is something simple I am missing.
I am naming the cavities by their mirror serial numbers. We have:
I have convinced myself that Cavity1934 is good, but think that 1621 might need a redo. The story:
Assuming zero loss on the mirrors, I recall we designed the cavities to have ~20k finesse. As mentioned before, this corresponds to a ~13us fill time (just effective path length).
I repeated the process for cavity1621, and was able to get flashes / some sort of "ok" alignment, and when I sweep VERY FAST and average like the Dickens, I get a cavity pole of ~22 MHz, or a finesse of 33.
Either I am missing something, or I need pop off the mirrors, clean everything, and reassemble cavity1621. I will crowdsource ideas for what I could be doing wrong shortly.
[EDIT: I no longer trust any of the red text - I was using the ITC510 to do the sweeps, and now believe that it was responsible for the crappy ungrokkable transient behavior I saw. I moved to using the ITC510 *just* for temperature control, and Rich's nice current driver to do the current supply / sweeps, and was rewarded with things that looked like transmission peaks]
IF we have measured the reflectivity of the mirrors, there's no reason for the Finesse to be anomolous; the amount of unforseen losses that we get from dirty surfaces is not large compared to the transmission of the mirrors in a F=20k cavity.
Take a look at the 40m measurements of the PMC finesse or the measurements of the RefCav finesse which Yoichi did a while ago.
I redid the scans for both cavities today with a brief break for the crazy air leak.
They look non-flaky now: I blame the ITC510 sweep mechanism / current noise for the previous mickeymousery. I will fit and post them tomorrow.
Attached are the cavity scans.
I used a function generator to make a 100Hz 1Vpp triangle wave, and drive the modulation input of the current drivers.
The calibration of the sweep is:
(Sweep speed) x (current modulation input) x (laser diode current to frequency)
(1 Vpp / 5e-3 s) x (1 mA / V) x (0.31 pm / mA) x (1.94e14 Hz / 1550nm) = 7.76e9 Hz / second
I took a few traces for each cavity, and fit an Airy function to each one using fminsearch. Relevant MATLAB code:
load('TEK00003.CSV'); %y-values for scope trace
[3 1e8 1.1e-4*16]);
The X(2) in the above is the coefficient of Finesse, its just 1/sqrt(F)*FSR to get the cavity HWHM.
The 16.2 is: pi x 7.76e9 Hz / sec x 1 / (FSR = 1.5GHz)
For cavity 1621, the four measurements of HWHM that this gives us are [1.5519e5 1.5570e5 1.6110e5 1.5525e5] Hz
I will use the mean of these as my 1st measurement of the cavity pole.
For cavity1621: f_pole = 1.57e5 +/- 2.9e3 Hz
I repeated the process with cavity1934
The five HWHM measurements for cavity1934 are [1.6388e5 1.4913e5 1.4889e5 1.5713e5 1.5277e5] Hz
For cavity1934: f_pole = 1.54e5 +/- 6.3e3 Hz
The variance in the 2nd set of measurements was a bit bigger.
I have no idea what the systematics are here, or why the sweeps are asymmetric. I do not believe that these are actually 2-6% numbers, but I think "good enough" is the word of the day.
I will put them in the cryostat and close up today
Entered lab around Tue Nov 24 13:24:57 2020 to finish photographing Zach's cantilevers.
some things about cameras, and in particular the FinePix F300 EXR
For storing lab photos in W Bridge, you can use our shared google acct instead so that we all have access to it (see chat for secrets)
Thanks, the photos are now on the shared drive.
I prepared the cavities for bonding: All relevant surfaces were prepared with clear first contact. I 3D-printed some parts to help me identify the orientation of the wedge and center the first mirror on the spacer, but I made the tolerances a little too tight. I'm having new versions printing overnight and will attempt the first round of bonding tomorrow.
Based on the pictures shown in elog 1985, I have identified mirrors 1,2,5, and 6 as the ones I want to make into cavities. I will bond mirrors 1 and 5 first, and then figure out a good way to make a cavity and bond the second mirror in place while monitoring the cavity loss. I have as basic idea but need to print out some more parts to see if it works that way.
measured the Si cavity dimensions. All the mechanics was/is designed for a 2"x4" cavity, but the cavity is smaller and we might have to modify things...
hole at optical axis: 0.53"
venting hole: 0.25"
I wanted to give a visual outline of the process that we have been using to do silicon fab at the KNI.
Until now, we have been performing this process on individual rectangular sections of Si (recall that we sent two 6" wafers out to be diced a while back---see CRYO:1250). The first and second prototypes (CRYO:1260 and CRYO:1264, respectively) showed some imperfections in both 1) the etch pattern (i.e., the 2D pattern that defines the end blocks of the cantilever) and 2) the etch depth uniformity (i.e., the final surface roughness of the etched region). We suspected that either one or both of these are the result of cleanliness issues while handling the individual rectangles. After talking with Shiuh Chao from Taiwan, it seems that a better option is to perform almost all of the process on an entire wafer, then mechanically scribe out the individual cantilevers just before the main etch. There will be more details on that at the end of this entry.
The following is a photo walkthrough of a test run I did on a practice wafer on Friday. This wafer is just a single-side polish 3" wafer that was given to me by a KNI staff member, and the mask I used (see below) was the one Justin and I made for the individual rectangles. I did not cut the sections out before etching, so this run was never going to make usable cantilevers; instead, I just wanted to test out the "whole wafer at a time" scheme and see how it affected the etch quality. For this, I only chose to etch from the top (polished) side, as opposed to our standard, double-side etch.
Spin and bake
This is the step during which the photoresist is applied to the wafer. Beforehand, the wafer is cleaned using a 1:1:40 solution of ammonium hydroxide, hydrogen peroxide and water.
To date, we have been using a photoresist called ProTEK, which is actually a new-ish product that was designed to eliminate some steps with traditional wet etching (more on that later). To use this, a special primer must first be applied before the actual photoresist. In each case, the wafer is placed on a "spinner", which holds the center of the wafer via vacuum seal, the primer or photoresist is pipetted onto the central region, and then the wafer is spun at 2000 rpm for 1 minute, in our case:
After each application of either the primer or the photoresist on a side, the wafer must be baked on a hot plate. The primer requires a 1-min bake at 110 C followed by a 5-min bake at 220 C; the photoresist requires a 2-min bake only at 110 C. Since, per the standard procedure, the wafer must physically be placed on a hotplate (with a sacrificial wafer in between if so desired), I chose to apply the coatings in the following order:
I felt that was the best way to protect the more crucial top surface. Here are a couple photos of baking. On the left, the bottom side is up with the primer baking (top side is face down on the sacrificial wafer with its already-baked primer protecting it); on the right, the top side photoresist is baking. You can see some thin-film interference from the minute thickness variations.
After both sides are coated with photoresist, the wafer is ready for photolithography. This is accomplished on a Karl Suss MA6 mask aligner machine. This device aligns a photomask to the wafer (with sub-micron precision, if necessary), and then performs controlled UV exposure on the aligned system to define the etch profile.
So far, we have been using a high-resolution transparency print as our mask. This is a common technique in prototyping, and there is a glass blank in the lab that is used to hold the transparency (with scotch tape). The mask is loaded onto a holder and held with vacuum:
The holder is then slid into the machine, and then a drawer underneath holds the wafer, also via vacuum (there are different holders for each wafer size, and there are also adapters to hold smaller chips). The wafer, which is not yet installed, is held within the orange circular region on the plate below:
After loading, various parameters are set, including the contact type (in our case, the mask and wafer are brought into brief contact, then backed off 30 um) and the exposure type (we use 40 seconds of 365 nm exposure at the standard luminosity, which is fixed). Then, a microscope is used to adjust the position of the wafer under the mask via micrometers with 3 degrees of freedom (i.e., X, Y and theta). This is usually a very precise process, but in our case we can actually get by just looking at the alignment by eye. After everything is aligned, the exposure is run. For this, the microscope caddy moves out of the way and the UV source slides in and does its business. Here's a shot of the whole machine:
After exposure, the lithography pattern is only barely visible to the eye under certain light.
To complete the process, the exposed wafer must then be developed to remove the photoresist from the regions where etching is desired. This is done by swashing the wafer in ethyl lactate repeatedly for 10 seconds at a time, blow drying with an N2 gun in between. Four or five cycles is usually enough, and after this is done the pattern is clearly visible:
It is not clear to me what the gunk is in the regions to be etched (these are the central rectangles here). Repeated washing in the developer does not remove it any further, and it may somehow be related to the final etch quality. More on that later.
Now the wafer is ready for etching in KOH. When I did this with Justin, we set up a beaker on a hotplate with a magnetic stirrer, but the KNI has a full immersion etch bench that is much more convenient. The wafer is held in a wafer holder and dipped into the bath, which is heated to around 80 C and pumped to keep flow:
The etch proceeds at a rate of around 50 um/hour, perhaps faster if the solution is at a higher temperature. For this test, I let it run for about 2.5 hours before coming back to pull it out.
After etching, the photoresist must be removed using a piranha etch, which is just a concentrated solution of sulfuric acid and hydrogen peroxide. I think I didn't perform this at a high enough temperature, so in the following pictures there is still some photoresist on the wafer.
As you can see, the 2D etch definition is much better than in the first two prototypes I made with Justin (see the links at the top of this post). That is, the rectangular regions for the blocks at the end and for the thin cantilever section are very well defined, as compared to those previous trials with the individual rectangular pieces of silicon. So, I am concluding that those imperfections were the result of cleanliness/handling issues with the small pieces, and that they will not be present if we work with a whole wafer at a time.
On the other hand, the etch uniformity (final surface roughness) is still very bad. It is about as bad as the first prototype, and actually much worse than the second one (which itself was bad). From the 2nd pair of photos above, you can see that the unetched regions---though discolored from the remaining photoresist that still needs to be removed---have maintained their optical polish, while the etched regions are dull, even in areas that seem macroscopically uniform. This leads me to think that this is a problem with the actual etch process. It's interesting to note that the very-nonuniform-etch areas don't really correspond to the gunky areas I saw immediately post-development (see above).
In the last pair of photos, you can see some regions of the backside that were unintentionally etched. This is probably due to some photoresist having clung to a surface on which it was resting during any part of this process.
The conclusions are thus: 1) Working with a whole wafer solves a lot of problems, and 2) our etch process is not good.
A major caveat is that, apparently, the ProTEK photoresist was expired. It could therefore be that, with fresh ProTEK, we'll be in good shape. However, we may be better off cutting our losses and just going with the traditional oxide/nitride mask. This is the method used by Shiuh Chao's group in Taiwan, which has produced nice cantilevers, and is a straightforward process that I can develop quickly at the KNI. This process is summarized in this PDF:
I have discussed this with Melissa Melendes, the KNI staff member who has been helping me, and we can begin this on Tuesday. There are only two pieces of machinery that I need to get accustomed with to perform this process:
In addition, I'll need to make a new mask (one that has an array of identical cantilever patterns on a single wafer). I can make another transparency, which involves making another image and going to a printshop to have a high-resolution transparency printed, or I can just go ahead and make a nice glass chromium mask, if I know what I want. I'm told this can be done relatively easily using a mask maker at the KNI, though I will need to get some training on it.
For now, I will probably use another practice wafer and test out the oxide/nitride scheme to see how the etch improves. On the plus side, I performed the entire process you see here in a single afternoon by myself, so my proficiency is increasing. It will take a little time to learn to use the machines mentioned above, and having to use them in the process will make the whole fab take a little longer, but not by too much. Stay tuned...
Edmund Optics has some Si optics (lenses and windows) which we can buy for doing some simple testing of polished surfaces (~120$ ea.):
I saw these, but had been avoiding them b/c of the cost. I'll go ahead and buy a handful (6) for testing. If I don't use them, we can probably find a use down the road.
The most recent measurements on the Taiwan-sourced Glasgow-style cantilver (see CRYO:1213) are encouraging, but the best Q measurement at low temperature is still a couple orders of magnitude worse than what is theoretically achievable, and about one order of magnitude worse than our conservative clamp loss estimates. Also, I've done some measurements on other modes (that have different expected clamp loss contributions due to the relative strain energy ratios) to try and sort out what is going on, with little success. Finally, some modes---including the 2nd bending mode at ~650 Hz---exhibited very low Q for no known reason.
One thing I thought about is that, since the Taiwan cantilever did not fit in the groove that was built into the block for the Glasgow-style cantilevers and therefore is just sandwiched between the two large pieces making up the clamp (see CRYO:1211), the clamp is likely pushing down at somewhat of an angle, which could lead to all sorts of non-idealities. Since the other Si samples we have lying around are roughly the size of the clamping region of this cantilever (~300-500 um), I opened up the cryostat today and reclamped the cantilever using a spare broken-off 300-um-thick cantilever piece as a spacer on the other side:
Pumping it all back down, I immediately measured Qs a bit higher than what we saw last time around at room temperature. The last measurement I made before leaving was tau ~ 135 s ==> Q ~ 46000, though it had been increasing up to that point, likely from the residual pressure, which was at ~10-3 Torr when I left. Compare this with the Q of ~14000 from the last time around, though admittedly I did not record the pressure at which this was measured.
I went into the lab with Justin today to make our second prototype resonator. I am writing the procedure down here from memory. I will correct/update this as we continue.
Note: of all the steps below, only the photolithography is done at KNI. The rest are done within the Painter labs.
Photoresist "spin and bake" (must be done twice: once per side)
Photolithography (this must also be done once for each side)
Etch (to be continued...)
We want our input coupling mirrors to have the same coatings as the mini-mirrors that will be attached to the cantilevers. So, I have sent six blank silicon substrates to John Tardif at Coastline so that he can include them in the coating run.
These are 1-m concave / plano 1" silicon mirrors. They were originally provided by Coastline and they are still in their original packaging.
With the necessary parts in hand, we succeeded in getting the SiFi table floating last night.
We followed the instructions in the TMC setup guide. The air supply configuration is as designated there:
Of note is that this system uses special tubes that have restricting orifices inside between each valve and isolator. These short, black tubes are marked with a red rubber band around each, and are located in the setup as prescribed above. There are 4 in all.
I couldn't take a new measurement last night because the cavities were swinging like crazy, but I should have one soon. I'm not sure what to expect, but it would be great if there was some extra isolation of environmental noise in the target (mid-audio) band.
Johannes and I shook the table somewhat as we investigated what we needed to buy to get the table floated,
The package arrived today, but I forgot to take to the lab. It's in my office. We can set it up tomorrow if you want.
Cryo lab and QIL cymacs have been running without glitches since before the holidays on the SiLabs 5340 timing boards.
QIL was unstable at first, until a DS345 was added to regenerate the 10 MHz reference at the far end of the cable. (We have plenty of DS345s on the shelf now, but if we ever wish to free this one up, we could apply some of the tips in LT design note 514 at the Si5340 input.)
For future reference, this is what was done to the boards:
The input is a 10 MHz and ~14 dBm sinusoid. (This is derived from the CMOS clock output of the Jackson Labs LC_XO GPS disciplined oscillator using a tee network suggested by Wenzel, then fanned out by a Symmetricom/Datum 6502 chassis.) Outputs are two complementary 65536 Hz 3.3V LVCMOS clock signals (sufficient to trigger the TTL inputs of the General Standards ADC and DAC).
The ClockBuilder-generated register configuration is attached and python scripts are in the QIL CDS target directory under qil-timing.
We need to get the beat readout rebuilt now that we have changed (fixed) the alignment the alignment in the cryostat.
Nic and myself noticed that one of the paths (laser diode 68 a.k.a. West path) was much harder to lock than the others.
Investigating further, the error signal for this cavity was tiny (~10x) smaller than the error signal for the other cavity.
I followed this for a while, and found that we were back in the state where: WHEN WE STRESS THE BOARD WHICH IS ATTACHED TO THE BUTTERFLY MOUNT, THE MODULATION DEPTH DRASTICALLY CHANGES.
Rich discovered / theorized that the capacitors on the little PC boards (SMA to butterfly pins) which go to the diode were cracked. It was difficult to not stress the board given the construction, so he replaced the 1nF and 3nF SM capacitors with leaded caps. The modulation depth shot through the roof.
I am trying to take Rana's advice to give myself some sort of manly gamma for my sidebands.
I have a broadband EOM from Thorlabs:
Using the ZHL RF amp (one of the heatsunk ones) from R.abb, I could get ~24 dBm into the EOM, which gave me gamma~0.02.
It looks like I need some sort of resonant circuit to do better, so I am taking Zach's circuit (ATF elog#1248) and cleaning it up so that it works for my application.
Components in hand:
2 x tunable inductors
1 x small inductor (3.3 uH measured on the 878A LCR meter with pokey-probes)
Going hunting for a 1.62 uH inductor from R.Abbot (This would give me 35 MHz resonant frequency)