Gap-enhanced Raman tags for physically unclonable anticounterfeiting labels

The study addresses the growing global problem of counterfeiting, which spans luxury to everyday goods and poses economic and public health risks. Conventional anticounterfeiting techniques, including holograms, barcodes, and molecular tags (DNA, peptides, polymers), rely on deterministic fabrication processes that can be replicated, creating fundamental security vulnerabilities. Physical unclonable functions (PUFs) offer an alternative based on inherent, stochastic physical randomness that is practically impossible to clone. The research aims to develop and validate an optical, Raman-based PUF system using gap-enhanced Raman tags (GERTs) that enables large encoding capacity, rapid and repeatable readout, and robust authentication through point-by-point (pixel) comparison, thereby providing a secure anticounterfeiting solution.

Prior approaches include optical security features and chemically generated PUFs with large state spaces due to stochastic synthesis. Optical readouts via fluorescence or Raman are attractive for being non-contact and fast. SERS-based tags offer high multiplexing due to narrow spectral linewidths and unique molecular fingerprints, allowing demultiplexing methods (e.g., classical/least squares) to expand encoding capacity and reduce spectral cross-talk compared to fluorescence. Previous PUF implementations include fluorescent nanowire tags and microstructured fingerprints with pattern recognition. Chemical PUFs using luminescent nanomaterials, upconversion particles, DNA/peptide/polymer tags demonstrate feasibility and large capacities but still face limitations. GERTs, consisting of a metallic core–shell with embedded Raman reporters in nanogaps, provide strong enhancement, near-infrared excitation compatibility, high photostability, and environmental robustness, making them promising for repeated, fast authentication.

• Synthesis of GERTs: Core–shell Au nanoparticles with a ~1 nm nanogap embedding thiolated aromatic Raman reporters were synthesized by growing an Au shell on reporter-decorated Au cores via Au3+ reduction. Ten reporter molecules were used: 1,4-BDT, 4-NBT, 4-MBT, 2-M-5-NBI, 2-M-6-NBT, 2-NBT, 2-CBT, 4-CBT, 4,4′-BPDT, and 2-NT. TEM confirmed uniform gaps and sizes (~44–80 nm depending on reporter). Extinction spectra showed single visible LSPR peaks (500–650 nm). All GERTs produced strong SERS under 785 nm excitation with narrow, unique spectra.
• Fabrication of PUF labels: Aqueous GERTs were drop-cast onto SiO2 substrates with metal fiducial marks. Labels used either one type (e.g., 4-NBT) or multiplexed mixtures (three or ten types). For ten-type labels, concentrations were adjusted per tag to balance SERS intensities.
• Readout: Confocal Raman mapping (LabRAM XploRA INV, 785 nm, ~3 × 10^5 W cm−2, 60× objective) acquired spectra with mapping resolutions of 2×2, 10×10, and 50×50 pixels. Standard STAGE mode (stage motion) used 10–50 ms exposures per pixel. High-speed DuoScan/SWIFT mode used galvo mirrors and line-wise data handling to achieve ~0.7 ms per pixel and 6 s total for 50×50 maps.
• Preprocessing: Spectra smoothed (Savitzky–Golay), baseline-corrected using a fourth-degree polynomial and asymmetric truncated quadratic cost function; SiO2 spectrum included as reference in fitting.
• Demultiplexing (for multi-type labels): Non-negative least squares (NNLS) fitting modeled each pixel’s spectrum as a non-negative weighted sum of the ten pure reference spectra plus low-order polynomials for residual background. Weights per tag provided component intensities for each pixel.
• Digitization (encoding): Raw intensity matrices per selected bands (or NNLS-derived per-tag intensities) were standardized via Z-score (mean 0, SD 1) to mitigate system fluctuations. A global search optimization determined common thresholds for binning into binary {0,1} or quaternary {0,1,2,3} levels per pixel, maximizing within-label reproducibility and between-label disparity. Encoded data are matrices recording pixel positions, intensity level, and, for multiplexed labels, tag type, forming 3D encoding across position, tag type, and intensity.
• Authentication: Defined a similarity index I as the fraction of matching pixel codes between two digitized matrices (pixel-wise comparison). For one-type labels (4-NBT, 50×50), 100 labels were each scanned three times. Seventy labels formed a training set for threshold optimization; thirty labels formed a test set for validation. Similarity distributions for same-label vs different-label comparisons were evaluated to set preliminary error margins (decision thresholds) for authentication.
• Practical demonstration: PUF labels were formed on Scotch tape (adhesive side) using 4-NBT GERTs, then transferred to paper. Raman mapping at 50×50 showed strong tag signals with negligible background from tape/paper, demonstrating deployability on product surfaces.

• Encoding capacity: For a one-type GERT label (50×50 pixels), binary encoding yields 2^2500 ≈ 3.8 × 10^752; quaternary yields 4^2500 ≈ 1.4 × 10^1505. For ten-type GERT labels, capacity increases to (2^10)^2500 ≈ 5.6 × 10^7525 (binary) and (4^10)^2500 ≈ 3.2 × 10^15051 (quaternary), exceeding suggested minimum practical capacity (10^300).
• Readout performance: High-speed DuoScan/SWIFT reduced acquisition to ~0.7 ms per pixel and 6 s total for 50×50 maps, versus 20 min in STAGE mode (10 ms/pixel) where mechanical motion dominated total time.
• Authentication robustness: For one-type 4-NBT labels (50×50), similarity indices between repeated scans of the same label were high (about 94% for binary and 84% for quaternary). Similarity between different labels was much lower (e.g., around 57% binary and 30% quaternary in examples), yielding clear separation. Across 100 labels (training and test), distributions of I for same vs different labels were well separated with gaps of ~20% (binary) and ~30% (quaternary). Preliminary error margins suggested: 85% (binary) and 70% (quaternary).
• Spectral demultiplexing: NNLS effectively decomposed mixed spectra into per-tag maps; some bleed-through observed at pixels with exceptionally bright particles due to partial band overlap, indicating a need for more distinct reporters or improved algorithms.
• Practical deployment: Tape-based labels on paper exhibited strong SERS from GERTs and negligible background, supporting practical application and handling.

The study demonstrates that GERT-based Raman PUF labels meet key PUF criteria: identifiability, unpredictability, and physical unclonability. Three-dimensional encoding across pixel position, tag type, and intensity yields enormous theoretical capacity, ensuring vanishingly small collision probability and strong resistance to cloning. High photostability and near-infrared excitation compatibility of GERTs enable repeated, reliable readouts. Authentication via pixel-wise comparison provides fast decision-making with low false positives, and experiments show clear separation between similarities of same vs different labels. While real-world factors (unequal state probabilities, readout noise, system drift) reduce practical capacity, the demonstrated capacities far exceed minimum recommended levels, providing a robust margin. High-speed DuoScan/SWIFT scanning significantly shortens readout time, enhancing practicality. The tape-based demonstration indicates compatibility with product surfaces and supply-chain workflows. Overall, the approach offers a secure, scalable anticounterfeiting solution leveraging chemically generated randomness and multiplexed Raman fingerprints.

This work introduces and validates Raman PUF labels constructed from gap-enhanced Raman tags, combining stochastic nanoparticle distributions with multiplexed, narrow-linewidth spectral fingerprints to achieve extremely large encoding capacities and robust authentication. The labels can be read rapidly with a confocal Raman system using high-speed scanning modes, and digitized for reliable, repeatable verification with clear separation between genuine and non-matching labels. Practical demonstration on transferable tape suggests suitability for deployment across supply chains. Future research should focus on accelerating readout (e.g., line or multipoint illumination, direct Raman imaging), enhancing demultiplexing (more spectrally distinct reporters, advanced unmixing algorithms), improving physical robustness with protective layers, optimizing global thresholds and error margins across diverse readers, and developing secure transmission protocols to prevent data substitution attacks.

• Readout speed, while improved (6 s for 50×50), still lags some electronic PUFs and may not yet meet high-throughput manufacturing needs.
• NNLS demultiplexing can suffer from spectral bleed-through when bands overlap and from exceptionally bright particles; more distinct reporters or improved algorithms are needed.
• Reproducibility is affected by instrument instabilities (laser power drift, detector efficiency changes, alignment shifts) and SERS signal fluctuations, reducing similarity from 100%.
• Real encoding capacity is lower than theoretical due to unequal state probabilities and readout errors; thresholds and error margins require careful tuning and may vary between readers.
• Potential vulnerability in data transmission if digitized labels are intercepted/substituted; requires additional cryptographic or coding layers.
• Physical robustness of bare NP patterns to environment and handling may require protective coatings or encapsulation for field use.