作者拆解一款直径约4厘米的AlphaHound α-β闪烁探测器，称其灵敏度高于Ludlum 44-9，且体积更小。文中展示拆开过程（四颗内六角螺丝）并简要评价其内部结构。
I've recently acquired this tiny contamination monitor:
Just 4 cm wide!
It's more sensitive then a Ludlum 44-9 despite being smaller then it's pancake style G-M tube.
After removing four hex screws, the AlphaHound easily comes apart:
Oooo
This is very nice: Many similarly sized devices are difficult or impossible to open without damaging them. If it ever breaks, it won't be hard to get inside.
The top half has the buzzer, display and buttons. It does have some SMD components, but it's just voltage regulators and decoupling capacitors:
The display is a Crystalfontz CFAL128128A0-015W monochrome OLED:
Neither the display or the PCB are mounted to anything: They are held in place by pressure. Because of this, the back side of the PCB must be blank to avoid breaking the OLED display:
Wow, such component density.
The buttons live on a tiny daughter board:
These were a relatively late addition to the design, and are connected to the main PCB with a long ribbon cable. Unlike everything else, this board is actually screwed in to the case:
The case itself is 3D printed stainless steel, which is a reasonable choice for small volume products. However, the resulting metal is porous and hard to clean. (it's still an improvement over plastic in my book)
The black tape is my doing: This detector was one of the first (of this version) made and it had a loose screen: The tape takes up just enough space to keep things tight.
The bottom half connects to the top with a short ribbon-cable:
There are no components under the battery: all the detector's electronics are contained within the tiny 3x2 cm section above it.
The detector is hidden underneath the board:
To keep out stray light, the scintillator is mounted in a ring of black rubber, which makes contact with black foam glued to the PCB and mylar. When assembled, the foam is compressed and creates a light-proof seal against the rubber.
The scintillator is a sandwich of two different materials: Silver dopped zinc sulfide painted onto polyvinyltoluene mixed with an organic phosphor (EJ-212).
The zinc sulfide detects alpha particles, and the plastic scintillator detects beta. Alphas will produce a bright flash with a slow decay, and betas produce a much faster and dimer flash. The detector takes both of these factors into account to tell the difference between the two types of radiation.
The MICROFC-60035-SMT-TR photodiodes are very special: Instead of being a single photodiode, these SiPM's have an array of tiny reverse biased diodes:
In practice, the capacitors are connected to a low-Z output.
Each diode is run above its usual breakdown voltage, but they don't start conducting immediately. However, once a free electron-hole pair is created, the electron is accelerated by the electric field and slams into silicon atoms. These collisions are energetic enough to liberate more electrons: causing exponential "avalanche" breakdown.
A single photon is enough to make the diode start conducting.
It's a similar principle to a G-M tube, just for visible light. Just like a G-M counter, the diode includes a quenching resistor which causes the voltage to drop once the discharge starts. This resets the photodiode so it can continue detecting light.
These detectors have quantum-limited performance > 1 gigahertz bandwidth: something that's ordinarily super difficult to do.
A single avalanche diode isn't able to measure the intensity of a light flash, but the SiPM contains thousands of them: The amplitude of the output pulse depends on how many diodes are triggered, which is proportional to the brightness of the light.
There's also a tiny LED which is used for a self test: If the SiPMs are able to pick up a dim LED flash, they should be able to pick up particle events.
Ok, back to the board:
A map of the hound
The microcontroller is the ATSAMD21G18, a 32-bit ARM processor capable of running at up to 48 MHz. That might sound slow, but it's actually quite powerful for an embedded system: It doesn't have to run chrome.
The second largest chip is an ADXL335 accelerometer. In earlier versions, this was used to control the device, but is being phased out due to it's high cost.
Most of the other chips are too small to have a full part number printed on, but they are mostly voltage regulators, comparators and opamps.
The top left has a very standard boost converter:
This converts 3.3 volts into ~30 volts which is used to run the photodiodes.
I don't currently have a way to strip off the conformal coating covering it, so I can't trace out the pulse processing circuit. However, I'm quite confident it uses a peak detector circuit to measure the height of the pulse:
Theoretical pulse detection scheme: Don't look too closely.
This is a safe assumption because the microcontroller simply isn't fast enough to measure the 100 nanosecond scale pulses: The ADC is only able to measure a voltage every ~3000 nanoseconds.
The pulse shape discrimination is likely done by using an opamp integrator to time how long the pulse stays over a given threshold:
Theoretical PSD scheme: Don't look too closely.
This method produces similar pulse scatter plots to the real detector — including the distinctive curve of the alpha cluster — and is relatively simple...
... but I don't know if this is actually how it works.
This section will be updated soon™.