2017 December 27

Holiday update on status

I have not yet stopped Run 66 - I've run a few tests and I thought I'd give an update. I've refined the estimate on the energy threshold and learned a bit about how to run stably, so with Anthony's input maybe we can make a final educated decision about whether to end this run.

As I've informed some of you, the problem with the DAQ event read rate has essentially been solved. I am now running with the modified trigger front-end program as the new default. There seems to be still some loss of events, less than 10%, which could be explored more deeply. One thing that's been learned is that it won't be easy to extract true rates from old data because the loss was energy dependent.

We tried reversing the QET bias of three of the 6 channels so that their pulses were inverted. This should help eliminate correlated noise from external sources when the pulses are added to create pTotal. But Nick did a quick check and saw no improvement in the randoms width in the pulse-flipped data. So it seems that any external pickup is negligible compared to our primary noise sources (e.g. HV noise, and maybe passive Johnson noise).

We had some discussions about the HV noise. A model of the HV noise in this detector is that it's mostly from some sort of low-level conduction of current through the crystal. One still unsolved mystery is that calculations of the HV-induced power at the TESs are much higher if you base it on phonon baseline shifts than if you base it on noise. It suggests that maybe some heating of the TESs occurs apart from the Luke phonons produced from the leakage current (thermal phonons, which would not show up as signal but would change the effective substrate temperature). It makes you wonder whether some of the leakage is surface current or something.

Nick decided to run some tests with HV bias on a subset of the Side 2 phonon channels, leaving the others floating. He tested biasing only pA (the outer ring), and all except pA. He measured the baseline shift and the noise increase between 0V and 70V in all three cases. He found that the lowest noise and HV-induced power occurred in the standard all-channels-biased case, and that the highest was in the pA-only case. This is not quite what we expected but it actually fits Noah's understanding and previous observations: when the area biased is smaller, there's a greater charge density on those channels and the maximum field is stronger, and leakage has been seen to be higher in these cases.

At any rate it's clear that we can't improve our noise situation by floating some of the Side2 channels.

By applying 8 μW power to the MC heater we could raise the Tower temperature to 55 mK. I took quick IbIs at this temperature and found the expected shift, and chose lower QET biases to put the TESs close to the point in their transitions we ran them at 21 mK. What we knew from Run 61 is that this would reduce the electrothermal oscillation issue.

The main purpose was to see if I could improve the noise threshold. There was some suggestion that right after biasing, the noise was lower - but as I'll describe below, it may not be a good measure to look at the noise at an early point after biasing. When comparing the threshold an hour after biasing, it was almost exactly the same in the low- and high-temperature cases. This is itself interesting - the TES dP/dI gains are certainly different, but the calibrated noise is the same, suggesting that the noise is mostly at the TES level - the noise is amplified as much as the signal. This would certainly be true for shot noise from HV leakage, and maybe also for passive Johnson noise.

There's an advantage to the higher temperature because it reduces the oscillations, so that for example it's easier to trigger on the center channel, which goes into oscillations after a muon - at low temperatures this produces trigger bursts, but at the high temperature we avoid them. The disadvantage of the higher temperature is that the TESs are more sensitive to variations in base temperature - the baselines can wander during data-taking, which I don't see at the low temperatures. So I've been running at the low temperature since completing this test.

While evaluating the noise threshold and also how things change with time after biasing (starting with a neutralized detector), I realized I needed to be careful with the calibration. Taking data after biasing in 10-minute increments, I saw that the 10keV line drifted upward somewhat over the first hour. I then made a time-dependent calibration, and evaluated the noise threshold versus time. It gets worst over the first hour then flattens out. The best threshold is right at the beginning, near 100 eV, but that's very short-lived. Over the full 2-hours the mean threshold is around 180 eV; over the first 30 minutes it averages around 140 eV.

Comparing our spectra to the CDMSlite paper, ours looks a bit like CDMSlite Run 1. This reminded me that they used a fiducial cut because for high-radius events the charge carriers hit the detector wall and don't experience full Luke gain. So I tried a p-inner cut, and it has a very strong effect. Plotting the pA partition versus energy shows that 1) at 10keV, there's a falloff in the peak for high-radius events, and distortion for very high-radius events, 2) most of the low-energy blob seen in triggered events has high pA partition for some reason, and 3) a cut on pA partition can have high efficiency for 10keV and 1keV events while cutting out much of the low-energy stuff. A cut with about 90% efficiency at 1keV produces a better noise threshold, reducing the full 2-hour average to below 150 eV and around 120-130 eV over the first 30 minutes.

A very serious is question is why so much of the “noise blob” events in triggered data are removed by the p-inner cut, and also most of the randoms are removed (over 90%, after some quality cuts to remove tails). It could be that there is sporadic noise that just happens to produce high pA-partition. Or it could be that a lot of tails aren't removed by my crude tail cut, and random traces during tails produce high pA-partition for some reason. It raises the question of whether this cut produces a big efficiency hit at low energies, but we have no evidence of that - the 10keV and 1keV efficiencies are both good with the p-inner cut applied.

Even with the improved noise threshold, presumably below the 0.16 keV line, I haven't quite been able to see that line. Partly that's statistics, so I wrote a sequence script to gather a lot of data all within 30 minutes after biasing. A full cycle of flash, cooldown, prebias, bias, data, unbias is about 75 minutes, so it was a 40% duty cycle. I've run several 12-series (6 live hours, 15-hours real-time) sequences this way. One very promising thing is that none of the phonon channels has railed in any of these sequences. So we are able to achieve very stable running.

I took the series with a low-threshold trigger on a couple of different channels that don't have oscillation problems. I've started to analyze these and still haven't quite seen the 0.16 keV line. We seem to be right on the edge of it. I've discovered more quality cuts and there may be a hint of the line - I'll give another update.

I will schedule and announce a meeting tomorrow. I know Nick will be here but I'm not sure who else - assuming we have a quorum we can discuss these results further.

Matt