An open-source Bitcoin seed generator called Entropy32 Plus has demonstrated a method for converting radioactive decay events into standard BIP39 recovery phrases using an offline hardware device. Published in September 2026 by developer Alexander Higgins, the project’s repository includes firmware, circuit-board designs, fabrication files, and a printable enclosure. Higgins describes the device as an experimental, educational project and explicitly warns users not to trust it with substantial funds before an independent review.
Radioactive Decay as a Physical Entropy Source
Entropy32 Plus, nicknamed “The Universe Bifurcator,” captures pulses from a Geiger counter and measures the time between detected decay events. Radioactive decay occurs unpredictably at the level of individual atoms, and the project uses this timing as its physical source of randomness rather than relying solely on software-generated values.
The current design accepts pulses through a 3.5-millimeter connection from a GQ Electronics GMC-320S Geiger counter. According to the documentation, the counter can detect emissions from a radioactive source or background radiation. After receiving a pulse, an LM393 comparator converts the counter’s analog signal into a digital input, which passes to an ATmega328P microcontroller on a custom circuit board.
The firmware records each accepted event using a microsecond timer. It compares the latest interval with the preceding one, producing a one when the new interval is longer and a zero when it is shorter. Equal intervals are discarded. Measurements below 200 microseconds are rejected as possible glitches, while accepted comparisons continue until the device has collected 512 bits. Higgins states this method uses pulse timing because the precise arrival of an individual decay event cannot be predicted, though he does not claim that every collected comparison contributes one full bit of measured entropy.
SHA-256 Conditions the Raw Seed Data
Once the 512-bit pool is complete, Entropy32 Plus processes all 64 bytes through its bundled SHA-256 implementation. The firmware uses the first 16 bytes of the resulting hash for a 12-word phrase or all 32 bytes for a 24-word phrase. According to the project, hashing is intended to reduce visible patterns linked to detector dead time, count-rate changes, or correlations between measurements. While hash conditioning can spread available entropy across an output, it does not prove how much unpredictability was present in the original data.
The firmware includes a startup test for SHA-256 that calculates the hash of “abc” and compares the result with a known value. The device stops before collecting entropy if the answer is incorrect. Passing this self-test confirms only that the implementation returns the expected result for that test vector; it does not test the Geiger counter, assess pulse independence, or measure the minimum entropy entering the hash. The project collects twice as many comparison bits as the maximum 256-bit BIP39 input it produces, and until the physical source undergoes formal measurement, the repository does not establish how much usable entropy each comparison contains.
Mapping Conditioned Data to BIP39 Words
Following SHA-256 conditioning, the device applies the process defined in the BIP39 specification, which converts an approved entropy length into a mnemonic sentence selected from a standardized list of 2,048 words. For a 12-word phrase, Entropy32 Plus uses 128 bits of conditioned data and calculates a four-bit checksum. The combined 132 bits are divided into twelve 11-bit groups, with each group selecting one word. The 24-word setting uses 256 bits and an eight-bit checksum, producing 24 word indexes from the resulting 264-bit sequence.
As explained in crypto.news’ seed phrase guide, recovery phrases hold the information needed to restore wallet access. Anyone who obtains the correct words in their proper order can recreate the wallet’s keys. A seed phrase is not itself a collection of independently generated words; modern Bitcoin wallets derive many private keys from one master seed, meaning weakness in the original entropy can affect every address generated from it.
Entropy32 Plus compiles the complete BIP39 English wordlist into its firmware, occupying 13,117 bytes — roughly 43% of the microcontroller’s flash capacity. With the OLED interface, SHA-256 implementation, and collection logic included, the compiled program uses 30,006 of the chip’s 30,720 available bytes, leaving only 714 bytes unused.
Offline Design Keeps Seeds Away from Connected Devices
The published hardware files list no wireless communications hardware. Users operate the generator through two buttons and read the phrase from a 0.91-inch OLED screen. Generated word indexes remain temporarily in the microcontroller’s RAM. Once the user records the phrase, pressing both buttons opens a confirmation process that overwrites the word array and restarts entropy collection.
The firmware clears the original pool after copying it for processing, and temporary hash, checksum, and entropy buffers are overwritten after the BIP39 indexes have been calculated. According to the developer, the device does not save a recovery phrase to persistent memory. However, the open-source design does not include a secure element or a mechanism that proves the installed firmware matches the public code.
Physical isolation removes several network-based attack routes, but it does not verify that a device was assembled correctly or loaded with trusted software. A modified unit could display predetermined words or retain generated data without changing its outward appearance. Crypto.news’ wallet security guide recommends keeping recovery phrases away from phones, cloud accounts, photographs, and other connected systems. Entropy32 Plus’s screen-based process is consistent with offline transcription, provided the hardware and firmware are trusted.
Independent Entropy Testing Remains Unfinished
The repository states that the Geiger-based entropy source has not been tested under NIST Special Publication 800-90B, which covers the evaluation of physical entropy sources and includes separate procedures for independent and non-independent data. Higgins calls for recording raw inter-arrival times over a long period and examining them with NIST’s non-IID minimum-entropy estimators. The firmware comments warn that general statistical test results alone would not establish sufficient entropy.
“Please verify the entropy source you intend to use otherwise understand that you will be using the device at your own risk,”
the README states.
The developer recommends mixing the Geiger-derived data with another independently validated source before creating a wallet for valuable funds. As of September 13, no outside audit, NIST assessment, or public raw-event dataset was included in the repository.
Past Wallet Failures Underscore Testing Importance
Past wallet failures demonstrate why entropy testing extends beyond confirming that final words follow BIP39 formatting. Crypto.news reported that Kraken’s security chief called for independent wallet testing after a Coldcard seed-generation vulnerability. A related technical review explained how weak entropy can expose Bitcoin wallets without revealing which people use the affected devices; attackers can generate possible seeds, derive their public addresses, and search the blockchain for matches.
Crypto.news later reported that predictable seed generation remained exploitable after the underlying flaw had existed for years. Correcting faulty firmware cannot repair phrases generated earlier; affected users must create new wallets and transfer their assets.
Active Development Without Production Release
Entropy32 Plus remains under active development. Its commit history shows an initial commit dated September 4, followed by firmware, circuit-board, and enclosure changes. As of September 13, the repository had no tagged production release or published timetable for completing an independent security assessment.
