The July 2022 issue of IEEE Spectrum is here!

Close bar

DIY Gamma-Ray Spectroscopy with a Raspberry Pi Pico

But beware a hardware glitch in the Pi’s RP2040 chip

5 min read
A hand places a tube protruding from a square circuit board against a tea cup.

Radioactive minerals can be identified in a surprising number of places, including old ceramic glazes.

James Provost

The global semiconductor shortage has made life tough for anyone using microcontrollers, with lead times now sometimes quoted in years. But there has been one bright spot: the US $4 Pi Pico, a microcontroller based on the new RP2040 chip. Not only does the RP2040 have lots of compute power, it hasn’t suffered the kind of shortages afflicting other chips. So when I decided to build a cheap DIY scintillating gamma spectrometer, it was the natural choice—although I didn’t realize I’d find myself navigating around teething problems of the sort that often affect a first-generation integrated circuit.

My interest in gamma-ray spectroscopy came from my physics studies. I find it fascinating that you can get so much information out of a single device. A gamma-ray spectrometer can be used like a Geiger counter with much better sensitivity, but unlike a Geiger counter, you can identify the exact composition of any gamma-emitting radioisotopes down to the picogram (or less). I started thinking about creating my own gamma-ray spectrometer when I saw the high price of even the cheapest commercially made devices. I wanted to see if I could make it easy and affordable to build a spectrometer.

The first step was to choose the scintillator at the heart of the spectrometer. In a nutshell, a scintillator measures both the energy and intensity of a flux of gamma rays, thanks to a transparent crystal. A gamma ray produces a free electron in the crystal, and this electron’s energy is proportional to the gamma ray’s. As the electron moves through the crystal, it excites atoms. The atoms, in turn, emit visible photons, with the total number of photons emitted proportional to the energy of the exciting electron. Thus, by counting the number of photons, you can gauge the energy of the original gamma ray. Counting how many gamma rays you detect over time gives you the radiation’s intensity, and looking at the energies of the gamma rays gives you a spectral fingerprint of a radioisotope.

The photon signal must be amplified to be detectable. Historically, this was done using a photomultiplier vacuum tube, but silicon photomultipliers (SiPMs) have become more common, and for my project they have a number of advantages, particularly in eliminating the need for a high-voltage power supply.

You can buy various used scintillator crystals on eBay fairly cheaply: I purchased a small sodium iodide crystal, 18 millimeters in diameter and 30 mm long, for about US $40. It came with a photomultiplier tube, which I removed and replaced with my SiPM, wrapping the assembly in black tape to prevent external light from leaking in and triggering the sensor.

3 printed circuit boards and a cylindrical crystal in a metal tube.A Raspberry Pi Pico [left] provides both compute power and the gamma-ray spectrometer’s analog-to-digital converter. A carrier board [middle] provides power and an interface with the silicon photomultiplier [top right] and scintillating crystal [bottom right] that react to gamma rays.James Provost

The scintillator and SiPM plug into a carrier board, which has a DC/DC boost converter to convert 5 volts into the 29.3 V the SiPM needs. The carrier board also hosts the Pico microcontroller along with some other supporting circuitry, including an amplifier that increases the output voltage of the SiPM to a level that the Pico’s built-in analog-to-digital converter (ADC) can detect.

On paper, the Pi Pico’s ADC looks very good. But there’s a flaw lurking in it.

The ADC in the Pico’s RP2040 chip is a critical component, and on paper it looks very good.
It has 12-bit resolution and can take measurements between 0 and 3.3 V at a rate of 500 kilosamples per second. But there’s a flaw lurking in the RP2040’s ADC.

I didn’t realize it existed until I started taking test spectra, writing software for the Pi that breaks up the ADC’s readings into 4,096 channels and counts the number of events in each channel over time. I noticed one channel kept reporting very high count values, creating a thin spike in my spectra. Puzzled, I took a four-hour background radiation measurement and discovered there were four problematic channels where the signal spiked unrealistically.

I started searching for what could be causing this and discovered I was not the first to run into problems with the ADC. A great website by Mark Omo—an EE who took it upon himself to investigate the problem—provides a detailed analysis, but in summary the issue is this: Ideally, an ADC chops the voltage range it can measure into an identically sized sequence of steps, producing a linear relationship between the input voltage and the numeric measurements it outputs. Of course, no ADC has a perfectly linear response across its measurement range, but the RP2040 has four spots where input voltages produce a wildly nonlinear response. This was the source of the mystery spikes in my spectra.

Two charts: top chart is labelled "Household ionization smoke detector." The X axis ranges from 0 to 10,000 counts, and the Y-axis ranges from 0 to 600 kiloelectron-volts. It shows two sharp peaks labeled Am-241. The bottom chart is labeled \u201cTea cup with uranite (pitchblende) glaze\u201d, with an x axis ranging between 0 and 200 counts with a Y-axis ranging between 0 and 1500 keV. A sharp peak can be seen labeled Th-234, with broader peaks indicating the presence of U-235, Pb-210, and Bi-234.Radioactive minerals are more common than many people think: some sample spectra captured with the DIY detector and the isotopes responsible for their signatures. The boxes beneath each spectrum show the raw, uncalibrated data.James Provost

Until the RP2040 is revised to fix this glitch, there’s not much you can do about it directly. Fortunately, with 4,096 channels, I could afford to employ the simplest software fix—just throwing away the measurements in the affected channels—without affecting the quality of the overall spectrum significantly.

Controlling and getting data from the spectrometer can be done via a USB interface (which also provides the power needed to operate it). I wrote software that can accept serial commands to, for example, put the spectrometer into Geiger counter or energy-measurement modes, or upload a histogram of all the measurements taken since the last power-up. You can write your own code to communicate with the spectrometer, or use a Web app I created that also plots spectra. (A link to the Web app, along with full build details and PCB files, is available on GitHub.)

For the future, I hope to make the spectrometer hardware capable of using a wider range of SiPMs and scintillators, so that people can use whatever detectors they can find. I hope you join me in this fascinating hobby!

This article appears in the July 2022 print issue as “DIY Gamma-Ray Spectroscopy.”

The Conversation (1)
Bhavesh Kakwani23 Jun, 2022

This is awesome! I remember having to decode SiPM signals using an FPGA when I was in university...I'm sure it was much harder than using a Pico 😅 even despite the ADC issue the author faced. Thanks for sharing, great read

The First Million-Transistor Chip: the Engineers’ Story

Intel’s i860 RISC chip was a graphics powerhouse

21 min read
Twenty people crowd into a cubicle, the man in the center seated holding a silicon wafer full of chips

Intel's million-transistor chip development team

In San Francisco on Feb. 27, 1989, Intel Corp., Santa Clara, Calif., startled the world of high technology by presenting the first ever 1-million-transistor microprocessor, which was also the company’s first such chip to use a reduced instruction set.

The number of transistors alone marks a huge leap upward: Intel’s previous microprocessor, the 80386, has only 275,000 of them. But this long-deferred move into the booming market in reduced-instruction-set computing (RISC) was more of a shock, in part because it broke with Intel’s tradition of compatibility with earlier processors—and not least because after three well-guarded years in development the chip came as a complete surprise. Now designated the i860, it entered development in 1986 about the same time as the 80486, the yet-to-be-introduced successor to Intel’s highly regarded 80286 and 80386. The two chips have about the same area and use the same 1-micrometer CMOS technology then under development at the company’s systems production and manufacturing plant in Hillsboro, Ore. But with the i860, then code-named the N10, the company planned a revolution.

Keep Reading ↓Show less