Author Topic: Consequences of SiPMs in parallel  (Read 2344 times)

0 Members and 1 Guest are viewing this topic.

Offline LoveLaikaTopic starter

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: us
Consequences of SiPMs in parallel
« on: June 18, 2020, 06:31:13 pm »
I've been working on this for a little bit now, but I wanted to ask here to get some perspective. I'm working with silicon photomultiplier, and some circuits I've built work fairly well. My circuits consist of an SiPM (in this case, the MicroFJ-60035) with an op-amp in a transimpedance amplifier configuration (TI's OPA656). For reference, I've linked to two PDFs below: one showing common readout circuits and one of the datasheet of the SiPM. The TIA circuit I'm using is in the readout PDF, figure 13, left side.

Now, it works fairly well with a single SiPM, so I tried to put more SiPMs in parallel to construct arrays. However, when you put SiPMs in parallel, you increase the input capacitance and you increase the noise from each one when you connect them all to one TIA, and one output. I got some results with 4 SiPMs in parallel, but putting 16 SiPMs causes some issues. I'm not getting any output under the same test conditions. I was thinking that this may be due to the increased capacitance, but I thought the OPA656 would be able to handle that. For reference, the SiPM's datasheet states that the anode has a capacitance of 4.14 nF, so with 16 SiPMs, that's 66.24 nF. That seems quite big. Given what I'm trying to accomplish, is it better to find a different way to read out the SiPMs, or can I just simply change the op-amp to something more suitable given the large capacitance? If so, what op-amps would you recommend?


Readout: https://www.onsemi.com/pub/Collateral/AND9782-D.PDF

Datasheet: https://www.onsemi.com/pub/Collateral/MICROJ-SERIES-D.PDF
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
Re: Consequences of SiPMs in parallel
« Reply #1 on: June 20, 2020, 05:02:24 am »
I have not used SiPMs.  I see the ap note refers to another for paralleled SiPMs:  https://www.onsemi.com/pub/Collateral/AND9778-D.PDF  Page 10 shows using Schottky diodes to isolate the SiPMs from each other but OR their current pulses together.

Without the OR diodes, when I look at how they work I'm thinking that unless the opamp can keep its summing node at zero V during a pulse, the voltage developed will allow some of the charge to get lost in the paralled SiPMs' R and C. Do you have a capacitor across the feedback resistor with a value proprtional to the number of SiPMs?  It would seem that any opamp would have to be both fast and able supply current to the summing node and if the feedback capacitance is quite large, it will need more current.  Maybe you'll need a current buffer to augment the opamp if it cannot supply the current needed.

Were you using the SiPMs in normal or fast mode? Intuitively, I don't think that opamp is fast enough for fast mode.  I'm not familiar enough with currently available parts to suggest another.
« Last Edit: June 20, 2020, 06:51:28 pm by duak »
 

Offline LoveLaikaTopic starter

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #2 on: June 21, 2020, 04:42:51 am »
Thanks for replying. I see what you're talking about, though I wasn't using the Schottky diodes. I was just paralleling SiPMs together with the TIA circuit. I was using it with the normal signals, not the fast mode signals. Regarding what you said about the feedback capacitor, I was using the standard recommended value in the circuit, 3 pF. When you're talking about a current-buffer, do you mean a unity-gain buffer with a second op-amp? I think I can see how that may be useful in place of Schottky diodes, but it would have to be very fast. I know there are some specific op-amps developed for unity-gain applications, so I'm curious how throwing one in the mix will help. I'm not really familiar with op-amp characteristics myself. I tried other op-amps based on a single value, like the slew rate. The AD8014 has a really high slew rate, but it didn't turn out well. For a single TIA with a single SiPM, it kind of 'double-peaked' somehow, so I don't know much.
« Last Edit: June 21, 2020, 04:46:15 am by LoveLaika »
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
Re: Consequences of SiPMs in parallel
« Reply #3 on: June 21, 2020, 10:34:53 pm »
I haven't worked with SiPMs and, until I saw your post, didn't know they existed.  I see that Digikey has some for $40.  If they're sensitive to some types of particle radiation, I might be tempted to get one to fiddle with.   Anyway, the last photodiode amplifier I designed was about 15 years ago so I'm a bit rusty on the details of TIAs.  It worked with a medium area photodiode and had a bandwidth of a MHz or so.  Hopefully, some of that knowledge is applicable to SiPMs.

The capacitor across the feedback resistor in a TIA helps compensate for the photodiode (or SiPM's) capacitance to ground and generally prevents the opamp from oscillating.  Its value depends on the feedback resistor, the various capacitances on the summing node and the gain vs frequency characteristics of the opamp.  I would think that its value should increase proportionaly to the number of SiPMs in parallel.

About the opamp output current:  looking at figure 6 on p.4 of the J-series SiPM data sheet I see a peak voltage of 70 mV into a 10R sense resistor.   This gives a 7 mA peak photocurrent that must be balanced out by the TIA.  I'm assuming the current pulse is from a photon triggering a single photodiode - AND9770 shows that higher currents are possible when multiple photodiodes are triggered.   The opamp can deliver at least +/- 48 mA, some of which can drive a 50 ohm load and the remainder to balance the photocurrent.  Are there any conditions that require the opamp to deliver more than 50 mA?

I found an interesting paper from CERN about SiPMs and readout electronics: https://indico.cern.ch/event/164917/contributions/1417117/attachments/198508/278657/1-cdlt_Photodet2012.pdf

A schematic with circuit values and operating conditions and some images of the waveforms you're having trouble with will help us understand what you're seeing.

Cheers,
 

Offline jmelson

  • Super Contributor
  • ***
  • Posts: 2852
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #4 on: June 21, 2020, 10:50:41 pm »
Now, it works fairly well with a single SiPM, so I tried to put more SiPMs in parallel to construct arrays. However, when you put SiPMs in parallel, you increase the input capacitance and you increase the noise from each one when you connect them all to one TIA, and one output.
Right, we've gone back and forth with this at work.  My boss thinks that multiple SiPMs will be great because they collect more light.  And, in theory, the light goes up by a factor of N, and the dark current pulses also go up by N, but add in quadrature.  But, the killer is the capacitance.  That ruins the performance of the amplifier.  We were really hot to make a large detector system with hundreds of SiPMs, and I got a preamp that worked well.  The Sensl (now ON Semi) SiPMs have the fast output, but the dark current is horrible.  We moved over to Hamamatsu devices, and the dark current is almost an order of magnitude less.  But, we still had issues with the discharge/reset of the micropixels causing a long tail.  Also, it seemed that the tail became longer with greater pulse amplitude, therefore changing the pulse shape.  Since we were interested in doing pulse shape discrimination, a constant pulse shape was pretty important.

Jon
 

Offline LoveLaikaTopic starter

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #5 on: June 22, 2020, 03:21:19 pm »
Thanks for your reply, duak.  Your comments made me think about things that I never considered. I've attached a basic schematic showing what I'm working with. Sorry, I don't have any waveforms.

I never considered the op-amp output current limitations. So, using your calculations of 7 mA from 1 SiPM, assuming we put multiple SiPMs in parallel, we get more current (proportional to the number of SiPMs) that we to balance out. With 4 SiPMs, that's roughly 28 mA, and with 16, that's 112 mA, more than 2x the op-amp ratings. In a power supply, if your circuit draws too much current when at a fixed voltage, then the voltage being outputted will decrease in order to fall below the specified current output (at least that's what happens when I draw too much current from a bench power supply). Is it safe to assume the same here, that with so much current, the op-amp output voltage will more or less drop to low values?

I don't really have any conditions of the TIA about current delivery, just that I would like to preserve the signal waveform. I also never considered increasing the feedback capacitance of the feedback loop. That is something I will have to look into and simulate to see how it works.

Thanks for the link to the slides. Hopefully that will get me thinking on ways that I can see the limitations in my design.
 

Offline LoveLaikaTopic starter

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #6 on: June 22, 2020, 03:26:26 pm »
Thank you very much for your reply and your experience. I'm always trying to learn from mistakes and see where I can improve. Thinking about what I'm trying to do, the SiPM's capacitance really screws things up. I just didn't really consider it when I ran tests. Based on your experience, can a good pre-amp compensate for the increased capacitance and dark current from the SiPMs? If so, what characteristics did you look for when choosing a pre-amp?
 

Offline jmelson

  • Super Contributor
  • ***
  • Posts: 2852
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #7 on: June 22, 2020, 04:46:11 pm »
Based on your experience, can a good pre-amp compensate for the increased capacitance and dark current from the SiPMs? If so, what characteristics did you look for when choosing a pre-amp?
Yes, up to a point.  The idea is to make the preamp input impedance as low as possible.  We used an LMH6714MF as the first stage amp, with a 900 Ohm feedback resistor paralleled by a 1.8 pF capacitor.  This gets pretty good rise and fall times, but not good enough for a fast scintillator.  We ended up back with photomultiplier tubes.

Jon
 

Offline duak

  • Super Contributor
  • ***
  • Posts: 1048
  • Country: ca
Re: Consequences of SiPMs in parallel
« Reply #8 on: June 22, 2020, 10:01:46 pm »
It seems to me that the SiPMs are quite sensitive and that the choice of opamps may not help with noise if it's generated by the SiPMs.  However, If the noise is generated by the opamp, it can be made worse by increasing capacitance from the summing node to ground.   The following is my attempt to explain Noise Gain.  A TIA's DC closed loop gain or transresistance is the output voltage divided by the input current and in your circuit is simply the 50 ohm feedback resistor.  If you were to place a 50 ohm resistor between the summing node and ground, the transresistance wouldn't change but a characteristic called the Noise Gain would be doubled.  The closed loop voltage gain is doubled because the two 50 ohm resistors act as a 2:1 voltage divider BUT their Thevenin equivalent feedback resistance is now 25 ohms so the transresistance is the same.  Now, any voltage present between the opamp inputs or elsewhere in the opamp is doubled.   The SiPM capacitance does the same thing, but because it is a reactance instead of a resistance and forms a complex impedance we have to talk about the transimpedance.  Bottom line: because the capacitance reduces the summing node's impedance to ground it reduces the negative feedback from the opamp output and so doesn't cancel out as much of the opamp's noise as it did before. 

I don't know what the optimal solution might be.  Intuitively, I'd consider the following:

1.) reduce the amount of the paralleled capacitance by using diodes (or common base amplifiers, see 2. following) OR'd together, as in the ap note.

2.) use a different type of TIA that is either less sensitive to the capacitance at its input or has an inherently low (ideally zero) input impedance. The first thing I thought of was a current feedback amplifier with its inverting input used as the summing node.  I don't know if a CFA is less sensitive to capacitance.  Two other devices are Current Mirrors and Current Conveyors.   I then thought of using a common base amplifier (CBA) using a VHF bipolar transistor with the input applied to the emitter and the output taken from the collector.  I see other designs using a CBA.   A CBA has a current gain of slightly less than 1 and can have a voltage gain >> 1 that primarily depends on the load resistance.  It can have a low input impedance and It's also very fast.  The next step would be to use one CBA transistor per SiPM and tie the collectors together to get a common output as per 1. above.

Here's a link to a paper on the CBA: https://arxiv.org/pdf/1806.02497.pdf
Here's another but for a particular application (circuit starting p.125) https://aisberg.unibg.it/retrieve/handle/10446/52227/107791/Binder_Fabris_DT.pdf

When looking for prior art I found this paper:
https://agenda.infn.it/event/9123/contributions/77294/attachments/56122/66254/5_-_SiPM_Bias_Amplifier_Circuit.pdf

Regarding the output current, do you expect to have light intensity high enough to generate more than 50 mA of photcurrent at any instant in time?  If so, then you will need a current buffer capable of that.  Otherwise the opamp will probably limit its output current and its output voltage will not increase any further.
Paraphrased as: That's all there is and there ain't no more.

Hope this gives you some ideas.
 

Offline LoveLaikaTopic starter

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #9 on: June 25, 2020, 03:49:23 am »
Thanks for the links. I'll look at them for some more ideas and play around with them.

If you don't mind me elaborating on the first one, I've been doing some experimenting with SPICE, following the app note you referenced with the SiPM's fast out and using diodes to sum them together. I used a model of the SiPM (sorry, can't share it due to reasons), but think of the SiPM's output as a pulse which turns on the diodes mentioned. Depending on the number of SiPMs activated, the respective voltages sum up at the output. Running it in SPICE, the output voltages (when compared to the SiPM directly) seem very small. I'm seeing around 40 mV for one SiPM. Perhaps this is the consequence of having diodes 'separating' them from each other.

Seeing the results in SPICE made me think more about amplification. With multiplexing, the signals are so small, and the OPA656 doesn't seem to work well in a non-inverting op-amp configuration. It's just interesting to think about.
 

Offline LoveLaikaTopic starter

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: us
Re: Consequences of SiPMs in parallel
« Reply #10 on: June 25, 2020, 04:05:33 pm »
Based on your experience, can a good pre-amp compensate for the increased capacitance and dark current from the SiPMs? If so, what characteristics did you look for when choosing a pre-amp?
Yes, up to a point.  The idea is to make the preamp input impedance as low as possible.  We used an LMH6714MF as the first stage amp, with a 900 Ohm feedback resistor paralleled by a 1.8 pF capacitor.  This gets pretty good rise and fall times, but not good enough for a fast scintillator.  We ended up back with photomultiplier tubes.

Jon

Thank you for your reply. If I may, I'd like to get your opinion on something. It's about the multiplexing scheme that Duak referred to:

http://www.sensl.com/downloads/ds/TN-Signal_Driven_Multiplexing_Method.pdf

I'm running some simulations in SPICE with some models (sorry, can't show the models due to reasons), but from what I understand, when you pulse SiPMs, they turn on/off the diodes to pass their pulse along, and it gets summed up along with the other voltages. This isolates the SiPMs from one another, and this also keeps their pulse shape, but this scheme reduces the magnitude of the pulse significantly, and when you pulse multiple SiPMs together, the timing gets a little slower. For example, if an ideal SiPM fast-out pulse produces a sharp pulse of 170 mV, then a multiplexed SiPM produces a same-shape pulse reaching 80 mV. It's also faster too.

Multiplexing like this seems to have advantages in speed, but the magnitude is significantly reduced. Question is, how do I magnify such a small signal?
 

Offline Marco

  • Super Contributor
  • ***
  • Posts: 7045
  • Country: nl
Re: Consequences of SiPMs in parallel
« Reply #11 on: June 25, 2020, 04:16:07 pm »
If so, what op-amps would you recommend?

Do you really need a TIA at all? It will represent a relatively small Rs and thus provide the fastest recharge time, but do you need that? The great thing about avalanche devices is precisely that they have enough gain you can often just use a resistor and simple voltage amplifiers, which is much simpler than TIAs.

Also why don't you use the fast output? With Rs = 0, you get fast recharge time too. Connect a common emitter amplifier to each fast output and connect the collectors together to sum the results for the individual SIPMs.
« Last Edit: June 25, 2020, 05:02:13 pm by Marco »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf