An open-source hardware project is exploring an unusual way to generate Bitcoin wallet recovery phrases: measuring the unpredictable timing of radioactive decay.
Entropy32 Plus, developed by Alexander Higgins and published in September 2026, converts timing differences between events detected by a Geiger counter into raw binary data. The device then processes that data locally with SHA-256 before turning part of the resulting output into a standard 12-word or 24-word BIP39 recovery phrase.
The project, also called “The Universe Bifurcator,” is designed to operate offline. According to its published architecture, it contains no Wi-Fi or Bluetooth hardware and does not permanently store the generated recovery phrase.
That design creates an unusual combination of physical randomness, dedicated electronics and standard Bitcoin wallet formatting. But the developer is also explicit about what the project has not yet established: the quality of its physical entropy source has not undergone NIST SP 800-90B evaluation or independent security review.
Entropy32 Plus is therefore presented as an experimental and educational device rather than a finished security product for protecting substantial funds.
From Geiger Counter Pulses to Binary Data
The foundation of Entropy32 Plus is the timing of individual radiation detection events.
Radioactive decay occurs unpredictably at the level of individual atoms. Instead of depending exclusively on a software-generated random value, the device uses the timing between Geiger counter pulses as its physical input.
The current design accepts pulses through a 3.5-millimeter connection from a GQ Electronics GMC-320S Geiger counter. According to the project documentation, that counter can detect either radiation from a radioactive source or normal background radiation.
When a pulse arrives, an LM393 comparator converts the analog signal from the counter into a digital input. A custom circuit board then passes that signal to an ATmega328P microcontroller.
The firmware records accepted events with a microsecond-resolution timer and calculates the interval between successive detections.
Rather than directly converting every measured interval into a numerical value, the software compares one interval with the previous interval.
If the newest interval is longer, the firmware records a one. If it is shorter, it records a zero.
When two intervals are equal, that comparison is discarded.
The firmware also rejects intervals below 200 microseconds because the project treats them as possible glitches.
This process continues until 512 comparison bits have been collected.
Why the Device Collects 512 Comparison Bits
Entropy32 Plus deliberately gathers more comparison bits than the maximum amount ultimately used for a BIP39 phrase.
The system collects a 512-bit pool, equal to 64 bytes, before applying its SHA-256 implementation.
For a 24-word BIP39 phrase, the device eventually uses 256 bits of conditioned output. A 12-word phrase requires only 128 bits.
The project therefore collects twice as many comparison bits as the maximum 256-bit input it ultimately supplies to BIP39.
That does not mean the device has demonstrated 512 bits of entropy.
Higgins does not claim that every comparison between radiation-event intervals represents one complete bit of independently measured entropy. That distinction is important because a sequence can contain many bits of data without containing the same amount of unpredictability.
Determining how much usable entropy exists in the physical source requires formal analysis of the measurements themselves.
At present, that analysis remains incomplete.
SHA-256 Conditions the Physical Measurements
Once 512 comparison bits have been gathered, the firmware passes all 64 bytes through its bundled SHA-256 implementation.
For a 12-word recovery phrase, Entropy32 Plus uses the first 16 bytes of the resulting hash. For a 24-word phrase, it uses all 32 bytes.
The purpose of this hashing stage is conditioning.
According to the project, SHA-256 can reduce visible patterns associated with factors such as detector dead time, variations in count rate or correlations between measurements. Conditioning distributes the available entropy across the hash output.
However, hashing does not establish how much unpredictability existed in the original measurements.
If the input contains less entropy than assumed, applying SHA-256 does not prove that the resulting 256-bit value possesses 256 bits of real entropy.
That limitation is one of the central unresolved questions surrounding the experimental device.
A Built-In SHA-256 Test Checks the Implementation
The firmware does include a startup test for its SHA-256 implementation.
When the device starts, it calculates the SHA-256 hash of the text “abc” and compares the result against the known expected value.
If the result does not match, the system stops before beginning entropy collection.
This check verifies that the firmware produces the expected SHA-256 result for that particular test vector.
It does not verify the Geiger counter, measure the independence of radiation pulses or calculate the minimum entropy contained in the source data.
The test therefore checks one component of the processing chain rather than validating the complete random-number-generation system.
Turning the Hash Into BIP39 Recovery Words
After the SHA-256 stage, Entropy32 Plus converts the conditioned data into the familiar BIP39 recovery-phrase format.
BIP39 uses a standardized list containing 2,048 English words.
For the 12-word option, the device takes 128 bits of conditioned data and adds a four-bit checksum. The resulting 132-bit value is divided into twelve groups containing 11 bits each.
Every 11-bit group identifies one entry in the 2,048-word BIP39 list.
For a 24-word recovery phrase, Entropy32 Plus uses 256 bits plus an eight-bit checksum. That creates a 264-bit sequence, which is divided into 24 separate 11-bit indexes.
This means the words themselves are not individually generated random selections.
They represent an encoded form of the underlying entropy and checksum.
That distinction matters because modern Bitcoin wallets can derive many private keys from a single master seed. If the entropy used to construct the original phrase is weak or predictable, the consequences can extend to every wallet address derived from that seed.
The BIP39 Wordlist Pushes the Hardware Near Its Limits
The complete English BIP39 wordlist is compiled directly into the device firmware.
According to the repository, the list consumes 13,117 bytes, roughly 43% of the microcontroller’s flash memory.
The entire compiled program, including the OLED interface, SHA-256 code and entropy-collection logic, occupies 30,006 bytes of the ATmega328P’s 30,720 available bytes.
That leaves only 714 bytes unused.
The figures illustrate how tightly the project fits its functionality into the selected microcontroller.
The device uses a 0.91-inch OLED screen to display information and recovery words, while two physical buttons provide user interaction.
An Offline Architecture Limits Network Exposure
One of the main characteristics of Entropy32 Plus is its physical separation from network-connected systems.
The published design contains no wireless communications hardware. Users read the generated recovery phrase directly from the OLED screen rather than transferring it to another device electronically.
Generated word indexes are held temporarily in RAM.
After recording the phrase, the user can press both buttons to begin a confirmation procedure that overwrites the word array and restarts entropy collection.
The firmware also clears the original entropy pool after copying it for processing. Temporary entropy, checksum and hash buffers are overwritten after BIP39 indexes have been calculated.
According to the developer, the recovery phrase is not written to persistent storage.
This architecture reduces several network-based exposure paths, but offline operation alone does not establish that the entire system is trustworthy.
Offline Does Not Mean Automatically Verified
The open-source design does not include a secure element or a mechanism capable of proving that the firmware running on a particular device is identical to the publicly available code.
That creates an important distinction between a transparent design and a verified physical unit.
A modified device could theoretically display predetermined recovery words or retain generated information while appearing externally identical to a correctly assembled unit.
Similarly, physical isolation does not confirm that the hardware was built correctly.
The project’s open-source repository allows the firmware, circuit-board files, fabrication data and printable enclosure to be inspected, but inspection of public files is different from verification of a specific assembled device.
The project therefore places significant responsibility on the person building or inspecting the hardware.
Formal Entropy Validation Is Still Missing
The most important unresolved issue is the amount of entropy provided by the Geiger-based source.
The repository states that the system has not been evaluated under NIST Special Publication 800-90B, which provides methods for assessing physical entropy sources.
Higgins recommends collecting raw inter-arrival timing measurements over an extended period and evaluating them with NIST non-IID minimum-entropy estimators.
The project also warns that passing general statistical randomness tests would not, by itself, prove that sufficient entropy exists.
No independent audit, NIST assessment or public dataset of raw radiation-event measurements was included in the repository as of Sept. 13.
The developer therefore warns users to verify the source themselves and says they use the device at their own risk.
For wallets holding valuable funds, Higgins recommends combining the Geiger-derived data with another independently validated entropy source.
Why Weak Entropy Matters for Bitcoin Wallets
The project’s caution reflects the importance of entropy in cryptocurrency wallet generation.
A recovery phrase can follow the BIP39 format perfectly and still be unsafe if the source used to create it is predictable.
Previous wallet-security problems have demonstrated how failures in random-number generation can create exploitable seeds.
Attackers do not necessarily need to know which person uses a vulnerable device. If the generation process is predictable enough, possible seeds can be created, their corresponding public addresses derived and those addresses compared against activity visible on the blockchain.
Fixing faulty firmware later also does not repair recovery phrases created before the correction. Users of affected phrases would need to generate new wallets and move their assets.
This is why Entropy32 Plus distinguishes between producing correctly formatted BIP39 words and proving that the original physical source contains sufficient unpredictable information.
An Open Experiment Rather Than a Finished Security Product
Entropy32 Plus remains under active development.
Its repository shows an initial commit dated Sept. 4, followed by modifications to the firmware, circuit board and enclosure.
As of Sept. 13, the project had no tagged production release and no published timetable for an independent security assessment.
Its significance lies in the way it connects a physical phenomenon with Bitcoin key generation: unpredictable radiation-event timing is measured, converted into comparisons, conditioned with SHA-256 and mapped into standard BIP39 words without requiring a network-connected device.
At the same time, the project’s documentation repeatedly preserves an important boundary between an interesting entropy source and a validated security system.
The hardware can produce 12-word and 24-word Bitcoin recovery phrases. Its firmware can process 512 comparison bits offline. It can erase temporary data and avoid persistent seed storage.
What has not yet been demonstrated is precisely how much minimum entropy the Geiger-derived measurements provide.
Until independent testing addresses that question, Entropy32 Plus remains what its developer describes it as: an experimental educational project that demonstrates a distinctive way of turning radioactive decay into Bitcoin wallet recovery data without claiming that the resulting system is ready to secure substantial funds.
