GadgetSci MCA

GadgetSci MCA

An open-hardware, 4096-channel multichannel analyzer that runs the whole spectroscopy pipeline on-chip. No PC required. Launching soon on Crowd Supply.

What it is

A pocket-sized multichannel analyzer: 72 × 87 × 18 mm, about 100 grams. It pairs with a SiPM detector — or any pulse source — and gives you 4096 channels of energy resolution.

The part that matters most: it does the full spectroscopy pipeline on the device. Pulse detection, peak reconstruction, energy extraction, and histogram accumulation all run on-chip. Most analyzers in this price range don't — they sample pulses and stream them to a host PC that does the real work in software.

That difference isn't academic. It means you can plug GadgetSci into a USB power bank, walk out the door, and come back with a complete spectrum. No laptop, no tether, no throughput ceiling imposed by a serial link.

Some specifics:

• 4096 channels, sustained count rate of at least 10,000 counts/s
• Preserves the full energy resolution of LaBr₃(Ce) scintillators — and has been run behind an HPGe. The analyzer won't be the bottleneck in your chain
• SMA input with selectable pulse polarity — it doesn't care what produced the pulse, so it works with detectors you already own
• Onboard software-configurable SiPM bias supply, 25–30 V for onsemi SiPMs. No external HV brick
• USB Type-C and 3.3 V UART
• Second USB port for an external battery — runs for days on a power bank

When you want a computer in the loop, the included GUI gives you real-time spectrum visualization, parameter configuration, and ASCII export for downstream analysis in ROOT, Python, InterSpec, PRA, or whatever you prefer. The UART path makes embedded integration straightforward — a rover, a drone, a fixed monitoring station, Home Assistant.

Why I built it

I work in a research environment where we build gamma radiation detectors. Characterizing them properly needs a high-quality data acquisition chain, and the institute has excellent ones. The problem is that they are always in use somewhere in the facility, and buying one for preliminary work is out of the question at institutional prices.

So the question was simple: could I build a cheap, good-enough acquisition system for preliminary testing and stop competing for time on the main instruments?

It didn't stay good-enough. Somewhere in the second year it became better than it needed to be, and at that point it stopped being a workaround and started being a product.

GadgetSci MCA is aimed at researchers in the same situation, at educators who want a real spectroscopy tool for teaching, and at serious hobbyists who have outgrown sound-card MCAs but can't justify a commercial instrument.

What you can do with it

Point it at a household smoke detector and watch the americium-241 line appear at 59.5 keV. At an old lantern mantle, and read the thorium-232 decay chain. At a piece of uranium glass, a bag of potassium-chloride salt substitute, a granite countertop, a piece of antique orange Fiestaware — each has a signature, and the whole point of a spectrometer rather than a Geiger counter is that you can tell them apart.

A Geiger counter tells you that. A spectrometer tells you what.

On the bench

Plug into a PC over USB-C and the included GUI shows the spectrum as it builds — counts accumulating channel by channel, photopeaks rising out of the background while you watch.

From there you can set and save the SiPM bias voltage, the threshold, and a detector alias — then export the whole spectrum as ASCII for analysis in whatever you already use: Python, InterSpec, anything that reads a column of numbers.

The device keeps building the histogram whether a PC is attached or not; a second USB port takes an external battery. The GUI is a window onto what the instrument is already doing, not the thing doing it.

That Eu-152 spectrum was taken through an HPGe. The SMA input doesn't care what produced the pulse, so the same pocket analyzer that rides a SiPM detector in the field can sit behind a germanium detector on the bench — which is exactly the situation this started from: a working acquisition chain, available, when the main ones are busy.

Take it into the field

A small ESP32-C3 bridge board connects GadgetSci to your phone. From there you get live spectra on screen, every measurement tagged with position, and a map history you can scroll back through.

That turns a bench instrument into a survey tool. Walk a site and watch the map fill in. Take it up a mountain and watch the cosmic contribution rise with altitude. Trace how background changes across a granite outcrop, an old mine tailing, a beach with monazite sand.

Tap any point on the track and you get the full record for it: date, time, latitude, longitude, altitude, radiation rate, and ground speed.

Ground speed matters more than it might seem. Counting statistics depend on dwell time, so a low reading taken at walking pace and a low reading taken from a moving car are not the same measurement. Logging speed alongside the rate lets you tell them apart afterwards, instead of guessing at your own data.

Everything stays on your phone. Download it whenever you want and do whatever you like with it — your own analysis, a plotting tool, a citizen-science project. No account, no cloud, nobody else holding your measurements.

How it works

SiPM pulses have fast rising edges — nanoseconds, not microseconds. Digitizing that waveform directly would demand sampling rates that have no business in a pocket instrument at this price.

So the front end does the hard part in analog. A SiPM detector is biased by the onboard configurable supply, and its output passes through a low-noise front end into a peak detect-and-hold stage that captures the pulse amplitude at full speed and holds it steady. That lets the ADC digitize a settled value instead of chasing a moving edge.

Getting that stage right was the hardest part of the design. I eventually built a test board carrying six different peak detector topologies and settled it by measurement rather than simulation.

From there, a dedicated onboard ADC digitizes the held value and feeds the RT1010 over a high-speed interface. Everything after that happens on-chip.

Once the samples are in RAM, the MCU takes over:

• Trigger and pulse detection — identify candidate pulses in the incoming sample stream
• Peak reconstruction — reconstruct the pulse from its samples
• Energy extraction — map the reconstructed pulse to an energy value using the calibrated response
• Histogram update — increment the corresponding bin in the 4096-channel histogram held in MCU RAM

All four stages run continuously while the next block of samples is already arriving. The Cortex-M7's floating-point unit keeps the math loop fast enough to sustain the pipeline in real time.

The MCU is an NXP i.MX RT1010 (MIMXRT1011) — a Cortex-M7 at 500 MHz in a part priced closer to mid-tier MCU territory than to high-end ones. That combination of clock speed and price is what made an on-device pipeline affordable in a product aimed at accessibility. MCUXpresso was mature enough that I spent my time on the actual problem rather than on vendor tooling.

The making-of

Two years, and a series of purpose-built test boards — each one made to answer exactly one question, including one carrying six different peak detector topologies, because I wasn't going to settle that choice on paper.

The product itself took two revisions. That's the payoff of testing separately: by the time the first integrated board existed, the hard questions had already been answered on boards built to fail.

Somewhere in the middle there was a vacation day spent seeing whether the entire system would fit in behind the detector itself. It did. Then came the overheating crisis, because real-time acquisition means nothing ever sleeps, and a sealed miniature volume has nowhere to put the heat.

That crisis is why the product has the shape it does. Splitting the detector from the acquisition system fixed the thermals — and turned the analyzer into a standalone instrument that works with any compatible SiPM detector, not just mine.

I'm publishing the full story in the Logs below, one episode at a time.

Open source

Full hardware design files under CERN-OHL-P, firmware and PC software under MIT, published at Crowd Supply campaign launch.

The audiences I most want to reach — physics educators, university labs, citizen scientists, detector hobbyists — benefit from being able to inspect, modify, and extend the tools they use. An open reference design for an RT1010-based instrument is worth more to the community than a closed one.

GadgetSci is also compatible with Pomelo detector modules in terms of communication protocol.

Launching soon on Crowd Supply

GadgetSci MCA is in pre-launch on Crowd Supply. If you want to know the moment the campaign goes live:

→ Pre-launch on Crowd Supply

If you're working on RT1010-based projects, building your own detector, or just curious about the design, I'd like to hear from you — leave a comment or reach out through the Crowd Supply page.