Needing more than a spark test?

That's OK - I expect I will have hold of it eventually, and it is probably the only one I would ever want. I do still have the PMT to play with.
We stay with it, though, even for USA folk, it is the largest cost chunk, proportionately, of all the bits (so far). We may revise that idea when it comes to the costs of the physical constructions of what goes around the AM241 stuff, and how the radiation lands on some steel offered. You wanted some lead bits in there too!

Keeping an eye out for other silicon PIN diodes of reasonable area, (betrayed by their capacitance), is something we should do. As I understand it, one needs more than simple area. The other thing needed is thickness, so that the photon actually gets absorbed instead pf passing right through. Diffusion thickness PIN diodes could reasonably easily be made. Getting up a thick enough epitaxial layer, that was pure enough was apparently uneconomic by slow diffusion. The technology moved on, and purer silicon and GaAs became available. Larger area diodes with the required absorption characteristics are still a very minority thing - but maybe we can get lucky. All the amplifier circuits and software can stay pretty much the same.

I did stumble a bit on the simulations. A perfectly well formed exponential rise pulse looks OK at 100nA, but starts to look a weird mess at 2nA.

I have to be reasonably convinced a build is going to work, and very definitely, it has to get actually built and the hell wrung out of it before I would expect anyone else to run with it on my say-so. When I do get hold of it, it's going to be winter breadboard time. I will be getting in some A/D converter stuff, probably one that uses SDA to command transfer of full 8 or 16 bits parallel at a time into GPIO. That will be at the A/D speed maxed out. Probably enough to get, (at the fastest), about 20 samples out of a 4uS complete event. It means about 200kHz to 500kHz bandwidth amplifier, and 2Msps to 5Msps A/D converter. The actual events would likely be slower. If we were counting detected "pulses", it would be Hz, to maybe a few hundred Hz, or some few kHz, but when they happen, we hopefully know their energy.

What we have is between about 5,000 and 60,000 electrons building up in a "pulse", being spread into a capacitance of between 50pF and 300pF, (depending on reverse bias), while at the same time being "leaked away" by a 40Meg ohms noisy resistance, and some few pA into (or out of) the charge amp . That little tickle of electrons in and out, onto the (tiny) junction of the TIA amp, provokes a feedback current to force the voltage between the inputs to zero. The feedback current, through the feedback resistance around the op-amp, gives us the first stage TIA gain.

In your vision of the gadget, do you have any preferences as to what batteries it carries? My notion is to have some button cells, or small batteries, enough to work the sensitive front end. It would be convenient if that also provided the bias, which can be between 2V to 10V. I thought the "back-end", A/D current, and data transfers could be powered from the receiving computing device.
- - - - - - - - - - - - - - - -- - - - -
My machine "fun" stuff is more or less mothballed for the present.. The profiles for the outhouse/shop/hideout are going up. If I get very cheeky, I can nudge it to (in USA units) 18' x 12'. I may have to settle for 16' x 10', but I am perishing the thought!
It's odd, but who would have thought that an extra 2ft would seem like acres!

I see some of the other shops folk have on HM. Compared to what I hope for, even maxed out as hard as I can push my luck with it, they seem gargantuan!
 
I found some papers discussing the use of silicon photovoltaic cells as x-ray detectors but it wasn't obvious if their output is proportional to the x-ray photon energy. The other "minor" issue is that solar cells are BIG, so their capacitance and dark current would be huge by comparison to the X100-7.

I have done some comparisons between the X100-7 and silicon photodetectors offered by Thorlabs and Roithner and the one parameter that jumps out is the capacitance/mm^2 for them all. The X100-7 has much lower capacitance, roughly .85pF/mm^2 vs about 4pF/mm^2 for the others. This suggests that the intrinsic region in the X100-7 is much wider compared to standard PIN photodetectors, so they may not work very well for us as proportional x-ray detectors. Oh well....

Regarding your question regarding battery operation, that's way out on my development horizon. I'd just design or buy components that use relatively little power so it would be relatively easy to drop in a LiPO power pack and linears/switchers as needed to generate the needed voltages. The pocketgeiger is a good example of using a boost switcher to generate the voltages needed to bias the detector and run the amplifiers/comparators (but at this point we don't know if it's quiet enough for our needs, eh?).
 
Interesting about the capacitance/mm^2. As a feature, I read that opposite to you, in that I think it to be a good thing.
It lets you know that the intrinsic region is thicker, hence the lower capacitance/mm^2.
The total capacitance is then made up back to (normal) or larger, only by more area.
The thicker layer brings more photons to a halt - is what I am thinking.

The supplied absorption probability plot lets us know that very near all in the range 5keV to 100keV are being collected, and perhaps enough of the remainder down to 1keV or 2keV to be useful.

I think what we gather is proportional to the energy. It just happens to be delivered as an awkwardly shaped pulse area. The amplitude is only nicely analogue to the energy if the pulse shape looked straight sided triangular with a constant duration length, which it does not! That said, there is enough "sort of triangle bump" in there to allow the "pulse stretcher" & peak detector approaches to make something that approximates more energetic photons arrival.

I'd just design or buy components that use relatively little power so it would be relatively easy to drop in a LiPO power pack and linears/switchers as needed to generate the needed voltages. The pocketgeiger is a good example of using a boost switcher to generate the voltages needed to bias the detector and run the amplifiers/comparators (but at this point we don't know if it's quiet enough for our needs, eh?
Normally, I also do exactly as you would, but when it comes to signals as low as less than 100,000 electrons, it gets difficult! The battery option was not available to me, and making correct bias tees with LDO regulators fast enough to fight line noise while also having their own band-gap references not contribute noise made the voltage supply circuits more elaborate than the amp. Chemical volts seemed clean by comparison.

I mostly attempted with pHEMt FET front ends at 8-15GHz, having noise at about 40K equivalent. S-Band FETs had noise at about 28K. Of course, a FET with 12dB gain at 10GHz is going to have unstable, unusable huge gain at 10MHz, and worse at audio, but the noise issue for our FETs is the same. Any resistive bits at all in the circuits before the first device gain hike would always raise the noise floor.

This is why I gave consideration to a cell-powered front end, and relegate anything with clocks, shifting bits etc, to not sharing it's noise by using a different power source, which can have regulators etc. The sensitive places are the TIA amp, and the bias voltage. The noise model is still coming together. If it indicates our noise contributions are less than the TIA and X100-7, then OK, we don't care anymore, but everything so far indicates that our amp would work well enough to be able to make the same plot as in the data sheet.

Here is a first partly assembled noise model using PWL approximation. (Try not to laugh)! It worked, but I have since abandoned it for a more direct way of making the PIN diode contribution, and for design purposes, we don't need most of the rest of it anyway. So it went down to something much simpler.

Noise-PIN-1.png

- - - - - -
So far, we are still having fun with the Royal Mail depot. This morning only open between 08:00 and 10:00, mostly accessed remotely via website, and a dysfunctional phone access options system. The COVID thing has made it kind of difficult. In person it shall be, or better, or if I can get to the postie (postman) when he visits our porch. He is a friend, and can likely help.

[Edit: Via my brother-in-law, who has more phone stamina than me. Success! Paid by card. Will be delivered Saturday. :):)]
 
Interesting about the capacitance/mm^2. As a feature, I read that opposite to you, in that I think it to be a good thing.
It lets you know that the intrinsic region is thicker, hence the lower capacitance/mm^2.
The total capacitance is then made up back to (normal) or larger, only by more area.
The thicker layer brings more photons to a halt - is what I am thinking.

No, we're on the same page. A wider depletion region = better for x-ray detection. That's why I believe the photodiodes from Thorlabs & Roithner won't be as good as the X100-7 when it comes to x-ray detector efficiency.

That's some diode model you came up with! Does it slow down the simulations much?

I'm happy to learn that you have resolved your delivery problem, one issue resolved..
 
There is a world of difference between having a mathematical equivalent and making LTSpice calculate it's way through a model subcircuit with a waveform transient. A Spice noise analysis provides a plot of the noise the easy way .

The "hard" way.
There are those folk who would generate a squiggly waveform of "noise" using a behavioral sources driven by RAND( ) and RANDOM( ) to make stuff like you see on the floor trace of a spectrum analyser. They even make filtered versions of pseudo-random bandwidth-limited noise, "white", "pink", you name it! To be fair, there are some analyses that need this. They are needed for spread-spectrum systems where the waveform looks like noise, but it isn't really random, and the data can be lifted clear of the noise if a coherent synced up pseudo-random local signal is used. Also for some modulation simulations. We definitely do not need them. In fact, for design, we hardly need any of the noise models in the previous posting anyway!

I don't do this..
LTSpice-Noise-02.png

The "easy" way.
Don't bother with the waveforms. SPICE noise analysis is just numbers. My one is from LTSpice doing its noise analysis from the known Nyquist or Johnson noise in a resistor from the molecules thermally bashing about in there. (OK - maybe wiggling about a bit)!

Vnoise = √(4kTBR) k = 1.380649E-23 Boltzmann's Constant
T = Temperature K
B=Bandwidth Hz
R = Resistance Ohms

We choose 1nV/√Hz, so B=1.
We choose 300K. I normally use 290K, but 300K seems to be the "norm".
Solve for R, remembering that Vn^2 needs one to divide by 1e18 to get back to nV
We find a resistor 60.358 Ohms will deliver that 1nV/√Hz

To turn that into any noise level we like, corresponding to what we see on the specification sheet, e.g the exceptional LSK389 from Linear Systems at 1.5nV/√Hz. Too expensive, and hard to get in UK, but anyway.. we use a Voltage-Controlled Voltage source, and put in 1.5 as the multiplier.

Voltage-Noise-Model-2.png

V1 AC 0 is just a dodge to persuade the .noise directive there is something making decades of frequencies for the plot.
The hit on computation time is just a blink. Simple and fast, and not even needed most of the time

Of course, this noise model is already built into the supplied op-amp device SPICE models from LinearTech and the like. I would only use it externally for things like FETs and X100-7's, and even then, not needed for design unless the need is to play dark current noise against signal, which we do, temporarily, have to check out.

X100-7 noisies
Saturated bias may give only 50pF from the PIN, but the dark current is 4nA.
6V bias gives about 100pF, with the dark current at about 2nA
100pF is still OK for TIA stability and bandwidth.

This is why I am trying so hard to get at how many nanoamps 5,000 or 100,000 electrons will provoke, if dumped in a pulse fashion into the capacitance and other stuff connected to it, this over a time 4uS to 10uS. The currents will be close to competing with the dark current. I still don't fully trust my figuring in that area, but it won't be long now, and most of my choices are already made. Hanging it together in proto form and getting some trial numbers snatched would have me cheering.

Of course, you have your own SparkFun toy, and your switched gain amp (which I might borrow), and a PCB. If you are slick with PCB layout, and you like mine, you can lay out what I test. Alternatively, I can do layout (a bit slowly) , and send you the manufacture files (Gerbers, whatever). Maybe there is a low cost small quantity supplier in UK, just like yours. I haven't researched that yet.

But that is looking ahead. Tomorrow we (hopefully) get the SparkFun thingie delivered. I may not even be able to play with it right away.
 
Last edited:
Unfortunately, the output of the pocketgeiger's first amplifier is not brought out to a test point, so I can't easily measure the DC voltage (in order to get a handle on the detectors actual dark current). I think at some point I will have to replace the copper shield with thin aluminum foil, which will expose the LMC662 for doing some (careful) probing.

I looked up the LSK389 and its noise specs do look impressive. Linear Systems doesn't have an online store, they want an RFQ (which I didn't submit). When you say "expensive", what are we talking about? I can imagine that their minimum quantity is large enough to put even a $1 part out of our reach.
 
From memory, I think the price of the LSK389 would cost $4-67 each for SOIC 8L for 1000 pieces.
It gets to $6-70 each for the TO-71 6L package, also for 1000. They have also the LSK170
Ahh..Haa - I found this..
https://audioxpress.com/news/linear-systems-reintroduces-lsk389-ultra-low-noise-n-channel-jfet

They are much hyped, but are in fact a version of 2SK197 and its subsequent me-toos, revived from the previous century.
Mouser lists it as known, but not in their offerings.
Digi-Key also strikes out
Linear systems site does not offer them except by getting taking your details to get up an order for 1000+
Even if it could be had in singles, it needs to be something I can get in UK without a hassle
(I am not yet through importi in effect a single diode on a hobby board, treated as if it was a volatile military item!
When I then peek among more available JFETs, I find some with nearly similar performance.

If I need to use the FET-augmented circuit at all, those JFETs would have to be a tad more attractive to acquire and use.

I know I am peeking in deep at what, in the end, may turn out to be a minimalist, fairly low cost small circuit, but that's OK.
 
I scaled the current pulse so the simulated output of the pocketgeiger matched the real board, using one of its ~.4V output pulses as the target peak output. Integrating the current pulse and dividing the total charge by 1.6E-19 indicates that the pulse comprises about 35,000 electrons.
 
Mark - please run that by me again..
400mV amplitude pulses where? Do you work back through the gain to get the current pulse?

The work function for silicon is 4.6eV to 4.85eV.
If that is what it takes to knock every one of those electrons free to be gathered by the drift field.
.. and they all came from one photon starting out.

Hmm .. not sure.
Did it count a passing 161keV photon?
(It's late here - I may not be thinking clearly)!
 
Mark - please run that by me again..
400mV amplitude pulses where? Do you work back through the gain to get the current pulse?

The work function for silicon is 4.6eV to 4.85eV.
If that is what it takes to knock every one of those electrons free to be gathered by the drift field.
.. and they all came from one photon starting out.

Hmm .. not sure.
Did it count a passing 161keV photon?
(It's late here - I may not be thinking clearly)!
The 400mV pulses were coming out of the second linear amplifier. To get the overall gain, I used the simulation results to calculate Vout/(charge in) and then back-calculated what input charge was necessary to get a 400mV pulse.

I chose the 400mV pulses as most-likely candidates because they disappeared when I removed the americium source. The monster pulses still appeared. I haven't tried my lead shielding to see how that affects things. That's something I need to do anyway....

The absorption edge for silicon comes to about 1.2eV, which is about a 1.2um wavelength (this why silicon-based cameras crap out below about 1um). So its absorptivity isn't dictated by what's needed to knock an electron off into a vacuum, but what's needed to move carriers between a couple of energy bands, whose energy difference happens to be.....1.1eV!

I confess to some confusion regarding how the X100-7 would respond to a 161keV photon -- at that energy, silicon should be pretty much transparent, but, on the other hand, the photon could generate a large number of carriers despite incomplete absorption. The best approach may be an experimental one, to compare (say) zirconium and molybdenum, whose K-alpha lines are 15.478 and 17.43Kev respectively, and see how the pulse heights change. Zr and Mo can be found on ebay for not a lot of money.
 
Back
Top