A physical unclonable neutron sensor for nuclear arms control inspections

The paper addresses the challenge of generating sensor data that mutually distrustful parties can accept as authentic and truthful, with a focus on nuclear arms control inspections. Traditional solutions rely on cryptographic algorithms running on trusted hardware and tamper-indicating enclosures, but these have vulnerabilities: keys and raw data are susceptible to side-channel and extraction attacks; enclosures prevent legitimate inspection; and hardware trojans may be undetectable even with full access. In contexts without a common root of trust or trusted supply chains, establishing trustworthy sensors remains unresolved. The authors propose passive, non-electronic, physically unclonable optical media that intrinsically authenticate and protect data. Building on the concept of strong optical physical unclonable functions (PUFs), which yield unique, hard-to-model light-field responses to coherent optical challenges, they further integrate stochastic neutron sensitivity so that the PUF’s responses change irreversibly upon neutron exposure. This couples sensing and authentication: unchanged optical responses certify non-exposure above threshold, while exposure decorrelates responses, evidencing the presence of high-energy neutrons associated with fissile material. The work targets zero-knowledge-style inspections to determine presence/absence of fissile isotopes without revealing sensitive design information.

Sensor design and materials:

• Optical PUF medium composed of two scatterer populations embedded in a viscous, immiscible gel matrix: (i) superheated fluorocarbon droplets (~100 µm diameter) that vaporize into ~600 µm bubbles upon interaction with neutrons above a chosen energy threshold Eth; (ii) inert solid microspheres (~5 µm diameter) to increase optical scattering without affecting neutron sensitivity. Droplet composition and size set the neutron energy threshold and sensitivity; scatterer and matrix densities are matched for stability.

Operational protocol (arms-control context):

• Setup/enrollment (inspector only): Inspector fabricates detectors; privately measures a large set of optical challenge–response pairs (CRPs) by illuminating sensor at positions q and angles θ to record output speckle fields; converts outputs to bit strings via image processing/error-correction for reproducible comparison; transports enrolled detectors to host facility.
• Validation (host only): Host performs non-destructive checks (e.g., X-ray) for safety; may destructively assay a subset for assurance.
• Proof of non-irradiation (host + inspector): Host asserts the detector structure is unmodified (no neutron exposure above Eth). Inspector randomly selects m < n enrolled challenges and sends them; host measures and returns responses; equality with enrolled responses implies no structural change (no above-threshold neutron exposure); used challenges are then retired.
• Presence/absence test for fissile materials (zero-knowledge extension): Place object between a neutron source of energy Es and detector array with Eth such that Es < Eth. Both parties monitor source energy/fluence. After irradiation, detectors are scrambled to prevent substitution, then challenged to verify no bubbles formed. If no bubbles are detected (responses match enrollment), inspector accepts proof of absence of fissile material. Because transmission (Eth < E) is avoided, and different fissile configurations can yield identical bubble counts, the protocol reveals only presence/absence.

Experimental realization:

• Fabrication: Optical cuvettes (1×1×3 cm3) filled with emulsions of C4F8 droplets (~100 µm; ~4000 cm−3) and polystyrene microspheres of 5.2 ± 0.42 µm diameter at concentrations up to ~7×10^7 cm−3 in water-based gel. Multiple emulsion types with increasing optical depth τ prepared.
• Apparatus: 635 nm laser diode (beam diameter ~2.9 mm) scanned across sensor surface; far-field patterns collected with a 1280×1024 CMOS camera (5.2 µm pixels). Neutron exposure via Thermo Fisher P-385 DT 14 MeV generator in a shielded canal.
• Response processing: Output images converted to reproducible bit strings using Gaussian pyramidal transform followed by 2D Gabor transform (scikit-image). Similarity assessed via normalized Hamming distance (HD). Memory-effect-limited challenge space estimated from decorrelation vs. beam displacement and angular sampling.

Analysis and metrics:

• Challenge space size estimated from spatial (beam position) and angular degrees of freedom within memory effect limits; decorrelation threshold displacement measured experimentally.
• Neutron-induced decorrelation quantified as HD evolution over irradiation time and as a function of the number of new bubbles for emulsions with varying optical depth τ.
• Intra-/inter-distance distributions computed to assess reproducibility and uniqueness across detectors.

• Integrated sensing-authentication: The optical PUF’s response inherently certifies the sensor’s structural state. Unchanged responses imply no neutron exposure above threshold; neutron interactions irreversibly produce bubbles that decorrelate responses, thereby signaling exposure.
• Large challenge space: Based on measured decorrelation after ~0.025 mm transverse beam displacement and angular sampling within the optical memory effect, a single sensor supports ~2.5×10^16 distinct challenges yielding uncorrelated responses. Exhaustively measuring all CRPs would take on the order of a year even at one CRP per nanosecond, exceeding realistic inspection timelines.
• Neutron-induced decorrelation: For detectors with optical depth τ > 4 (≈16 scattering events per photon), responses to a fixed challenge become fully decorrelated (HD ≈ 0.5) after the formation of as few as 1 to 9 bubbles. Time-series measurements show HD rising to ~0.5 during 14 MeV exposure.
• Sensitivity to challenge perturbations: Responses are highly sensitive to small beam position changes; average decorrelation occurs after ~0.025 mm lateral displacement on the sensor surface, consistent with strong PUF behavior.
• Zero-knowledge property for inspections: With Es < Eth and photon insensitivity, the protocol reveals only presence/absence of fissile materials. Simulations indicate different uranium objects (e.g., a solid sphere with U-238 fraction x238 and a thick shell with lower x238) can produce identical bubble counts under the same conditions, preventing disclosure of configuration or isotopics.
• Uniqueness and reproducibility: Computed intra-/inter-distance distributions are characteristic of optical PUFs, supporting that each sensor is unique and valid challenges are reliably recognized.
• Security against cloning and modeling: Exact physical cloning is impractical due to stochastic emulsification and microsphere placement; even with perfect 3D knowledge, simulating light transport for cm3-scale media would require ~10^26 operations (~years per cm3 at exascale). No successful machine-learning modeling attacks on optical PUFs beyond the optical memory effect have been demonstrated; current transmission-matrix/deep-learning approaches do not predict responses beyond that limit.

The study demonstrates that passive, non-electronic, optically complex PUF media can provide verifiable sensor security without reliance on trusted electronics or post hoc cryptography. By coupling neutron sensitivity to the optical PUF’s scattering structure, the medium serves simultaneously as a sensor and authenticator: if responses match enrolled values, inspectors can trust both the structural integrity (no above-threshold neutron exposure) and the honesty of reported data. This decouples trust in readout hardware from the sensor medium and addresses supply-chain and tamper concerns that have hampered arms-control verification. In inspection scenarios, the protocol enables zero-knowledge presence/absence determinations of fissile material while preventing leakage of design-sensitive information. The measured large challenge space, rapid neutron-induced decorrelation to HD ≈ 0.5, and strong uniqueness support protocol soundness and completeness under stringent adversarial models. Security analysis indicates physical cloning and response prediction remain infeasible with current capabilities, even for resourceful state-level adversaries, due to manufacturing stochasticity and the computational intractability of modeling mesoscopic light transport beyond the memory effect.

The work introduces and validates a new class of verifiably secure, physical-unclonable neutron sensors for high-assurance applications such as nuclear arms-control inspections. The passive, non-electronic optical PUF medium inherently authenticates measurement data and evidences neutron exposure through irreversible structural changes, enabling zero-knowledge presence/absence tests without trusted readout. Experiments confirm strong PUF behavior, large challenge spaces (~2.5×10^16), rapid neutron-induced decorrelation to HD ≈ 0.5 after few bubbles for optically deep media, and sensor uniqueness. These results suggest a practical pathway to authenticating inspector-provided equipment and expanding verification options in future arms-control agreements. Future research directions include increasing optical complexity (e.g., submicrometer scatterers at higher concentrations), exploring alternative matrix materials for stability, and leveraging quantum one-way functions and secure quantum communication to develop quantum-secure, unclonable radiation sensors.

• Field validation: The study is a proof-of-principle with laboratory experiments; performance under realistic field conditions and extended custody chains remains to be demonstrated.
• Stability and materials: Long-term pre- and post-irradiation stability depends on matrix properties; while prior work indicates months-scale stability without microspheres, different matrices may trade reusability for stability. Alternative scatterers (e.g., ZnO, SiO2) had compatibility issues.
• Zero-knowledge constraints: The protocol’s privacy relies on Es < Eth and photon insensitivity to avoid transmission information and photo-neutron confounds; careful source selection and monitoring are required.
• Attack surface: Although cloning and modeling attacks appear infeasible, advances in high-resolution imaging, manufacturing, and machine learning could reduce margins; comprehensive red-teaming is advisable.
• Throughput and CRP management: Enrollment requires generating and securely managing large CRP sets; used challenges must be retired, which depletes the list over repeated proofs.