Artificial fingerprints engraved through block-copolymers as nanoscale physical unclonable functions for authentication and identification

doi:10.21203/rs.3.rs-4170364/v1

Download PDF

Article

Artificial fingerprints engraved through block-copolymers as nanoscale physical unclonable functions for authentication and identification

https://doi.org/10.21203/rs.3.rs-4170364/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Besides causing financial losses and damage to the brand's reputation, counterfeiting can threaten the health system and global security. In this context, physical unclonable functions (PUFs) have been proposed to overcome limitations of current anti-counterfeiting technologies. Here, we report on artificial fingerprints that can be directly engraved on a wide range of substrates through self-assembled block-copolymer templating for secure authentication and identification. We show that engraved nanopatterns are unclonable unique objects that endow high encoding capacity density while satisfying main requirements of PUFs, including high aging and thermal stability. Besides showing that these nanopatterns can be encoded in binary code matrices with high entropy and high uniqueness, we propose a strategy for robust authentication and identification in real-world scenarios based on computer vision concepts. These results can shed new light on the realization of PUFs embracing the inherent stochasticity of self-assembled materials at the nanoscale.

Physical sciences/Nanoscience and technology/Nanoscale materials/Molecular self-assembly

Physical sciences/Nanoscience and technology/Nanoscale devices

Counterfeiting represents a growing problem for society that has not only enormous economic implications, from manufacturing of goods to specialized technologies, but represents a threat for the entire security and health systems.¹ In the current era of Internet of Things (IoT), widely exploited anti-counterfeiting techniques based on tags such as graphical barcodes, watermarks and holograms realized through a deterministic process endow low complexity and high predictability, allowing easy forgery and cloning by counterfeiters.² Also, widely exploited software-based authentication/identification systems endow intrinsic vulnerability due to electromagnetic interference and cyber-attacks.³ To overcome main limitations of current anti-counterfeiting technologies, hardware encryption systems based on physical unclonable functions (PUFs) have attracted a growing interest.^4,5 These are physical one-way functions that enable the generation of a secret key based on the inherent physical characteristics of a physical system/device, where the uniqueness of the key relies on uncontrollable variabilities and high randomness of the physical system even when realized through the same fabrication process.⁶ In this framework, new perspectives for the realization of PUF can be enabled by exploiting the true essence of nanotechnology where the fabrication of materials and systems at the nanoscale is often hindered by unpredictable disturbances and fluctuations.⁷ This is because at this scale the influence of thermal and statistical fluctuations cannot be avoided, resulting in output material structures and properties that are not simply determined by known inputs. Leveraging the inherent stochasticity of nanotechnologies, a wide range of physical unclonable functions have been demonstrated by employing different taggants and detection methods. This includes random fractal structures,⁸ fluorescent, photoluminescent and chromic materials,^9–12 spontaneously formed core-cell nanoparticles,¹³ 3D printed and nanoimprinted materials,^14,15 plasmonic nanomaterials,^16–18 graphene-based devices,¹⁹ as well as metallic nanopatterns replicated from molecular self-assembly^20,21. These techniques mainly imply the realization of unique nanopatterned devices as anti-counterfeiting tags that can be affixed to products requiring protection, but that can potentially be removed or detached through appropriate chemical/physical treatments.

Here, we report on artificial nano fingerprints engraved in a wide range of materials as disorder based PUFs for secure authentication and identification. Besides proposing a strategy for user-independent extraction of nanopatterns without artifacts from micrographs, we show that by exploiting local features of nanopatterns it is possible to encode nano fingerprints in unique binary code matrices endowing high bit uniformity and entropy. Finally, we propose and test a computer vision-based strategy for robust authentication/identification through artificial fingerprints in real-world scenarios, opening new perspectives for the realization of PUF that utilizes the intrinsic unpredictability of self-assembling materials as a source of randomness.

Artificial fingerprints at the nanoscale engraved on target substrates

Since ancient Babylon, fingerprints have been exploited as unique identifiers to secure commercial transactions, as testified by clay seals attached to their business documents.²² In the era of Internet of Things, fingerprint matching is largely deemed to be a secure system for both authentication (i.e., matching a person’s biometric template) and identification (i.e., determining the identity of a person), as schematized in Fig. 1a. Here, the validation and verification of a person’s identity are based on unique physical characteristics – biometrics – of their fingerprints.²³ The security of this process is guaranteed by the fact that, although two or more fingerprints may share the same global features, there are currently no known pairs of fingerprints with identical pattern of local features, also called minutiae points.²⁴ Besides the individuality of the fingerprint, the premises for fingerprint authentication/identification rely on the pattern persistence, meaning that the characteristics of fingerprints must be stable over time.²⁵ Similarly, artificial fingerprints can be exploited as PUFs for the authentication and/or identification of a product in the supply chain (the workflow of PUF generation, database initialization and authentication is schematized in Fig. 1b). Inspired by fingerprint biometrics, we developed a process to directly transfer a nano fingerprint pattern into a wide range of target substrates, as schematized in Fig. 1c. This process is based on i) molecular self-assembly of block copolymers (BCPs) on the target substrate, ii) selective removal of one of the BCP phase, and iii) pattern transferring to the target substrate. Molecular self-assembly relies on phase separation of chemically distinct and thermodynamically incompatible polymeric components of BCPs that occurs randomly under thermal fluctuations, resulting, under proper conditions, in lamellar fingerprint-like patterns (Supplementary Note 1). After achieving the self-assembled pattern, one of the two polymers comprising the BCP can be selectively removed, resulting in the formation of a disordered nanolithographic mask. Figure 1d reports an example of self-assembled lamellar BCP showing fingerprint-like global features (fabrication details in Methods). Notably, these patterns are characterized by the presence of spatially distributed local features (defect points) closely resembling minutiae points of human fingerprints. In case of lamellar patterns, the defect taxonomy of both positive and negative phases includes terminal points, 3-way junctions and dots (Fig. 1e). It is worth mentioning that typical features of lamellar structures such as typical dimensions, periodicity, correlation length, defect density and line-edge roughness can be tailored by appropriate choice of BCP molecular weights and processing conditions.^26,27 The polymeric fingerprint-like pattern can be then transferred to the target substrate through selective reactive ion etching (RIE) or wet chemical etching, depending on the material of the target substrate. Examples of patterns successfully transferred to diamond, Si, quartz and SiO₂ target substrates are reported in Fig. 1f-i (details of pattern transfer on different substrates in Methods). Additionally, patterns have been demonstrated to be transferable on metallic (Cu, Cr, Co, W, CoCrPt, Permalloy),^28–33 dielectric (Si₃N₄),³⁴ semiconductor (ITO, graphene)^28,33 and flexible substrates^35,36. The ability to conduct the self-assembly process of BCP on various materials opens the potential for directly engraving artificial fingerprints at the nanoscale onto a wide range of devices and products, including screens and microchips.

Extraction of pattern morphology

As for human fingerprints, a critical step for exploiting nanoscale fingerprints as PUFs is represented by the automatic extraction of the fingerprint pattern including minutiae (defects) from input images. This process is crucial and may heavily rely on the quality of the input fingerprint micrographs. Since ridge structures in poor-quality fingerprint images can be not well defined, the pattern can be not correctly detected, and spurious defects can be created while genuine defects can be ignored. To increase the robustness of pattern extraction with respect to the quality of input fingerprint images, we adapted a fingerprint enhancement algorithm able to adaptively improve the clarity of ridge and valley structures based on estimated local ridge orientation and frequency³⁷ (Fig. 2a, details in Methods). This allows to obtain a binarized genuine fingerprint pattern that can be exploited to investigate BCP morphological parameters including defect localization (defect maps), line period, line width, line-edge roughness, line-width roughness, and correlation length (correlation maps)³⁸ through an automated analysis (Methods) An example of the process flow, from SEM image to fingerprint pattern enhancement and defect localization, is reported in Fig. 2b. A comparison of a binarized pattern obtained by fingerprint enhancement algorithm and other common binarization thresholding techniques on the same micrograph is reported in Fig. 2c (raw micrograph in Supplementary Figure S1). Here, the white phase of binarized images represents ridge lamellae structures, while the black phase represents valleys. Fingerprint pattern enhancement outperforms traditional thresholding techniques, allowing to obtain a binarized map with reduced noise and with a reduced number of artifacts (additional details in Supplementary Figure S2). Notably, whereas the effectiveness of traditional thresholding techniques relies on the image quality, contrast and brightness, fingerprint enhancement enables the automatized extraction of patterns even in case of poor quality and low contrast images (Supplementary Figure S3). Moreover, it eliminates the need for image-dependent adjustment of binarization parameters, thereby avoiding any potential bias from the user. As depicted in defect maps (Fig. 2c), the reduced number of artifacts in images processed with fingerprint enhancement results in a limited number of spurious defects. Also, this approach enables the generation of correlation maps with a decreased noise level (Fig. 2d). A statistical analysis, performed by considering SEM micrographs acquired on different areas of the same patterned sample, allows a quantitative evaluation on the effect of the binarization extraction of defects and correlation length of the structures. Figure 2d reports the defect density obtained by analyzing images binarized with the different techniques. As evident, fingerprint enhancement results in a lower mean defect density, quantitatively showing the possibility of reducing the counting of spurious defects through this binarization technique. Furthermore, larger defect distributions obtained by conventional binarization techniques are due to overestimation of defects in images not properly binarized through these techniques (details in Supplementary Figure S4). More in detail, the radar chart in Fig. 2e shows that conventional binarization techniques result in a higher overestimation of certain types of defects. Since defect density is inversely related to the dimensions of lamellae orientation domains,³⁸ fingerprint enhanced images result in a higher mean value of correlation length (Fig. 2f). All these observations show that binarization through the fingerprint enhancement algorithm allows to retrieve a genuine fingerprint pattern that, besides reducing artifacts, is user independent and can automatically extract relevant information even from poor-quality images and images acquired with different contrast/brightness conditions. These are fundamental aspects for correct authentication and/or identification of PUFs based on BCP nanopatterned substrates in real-world scenarios.

Binary code matrices encoding of artificial fingerprints

These artificial fingerprints can be considered as PUF where the nanopattern represents the unique response r when the surface is scanned with a focused beam of electrons representing the input challenge c. In this context, the uniqueness of the PUF response r = f(c) is guaranteed by the intrinsic randomness of the self-assembly process of pattern formation, thereby relying on the internal and uncontrollable manufacturing variability of the system to establish the unique input/output relation f(∙). Note that the unique response of the PUF pattern can be probed also through other imaging techniques, such as atomic probe microscopy (AFM).

A first approach to exploit nanopatterns as physical unclonable functions is to generate a cryptographic binary code response of the system based on local features of nanoscale morphologies, as similarly reported in previous works.^20,39 In case of BCP templated patterns, the nano fingerprint feature can be converted to a binary code matrix by defining each pixel of the matrix as 1-bit or 0-bit depending on the presence or absence of minutiae in the corresponding spatial location, respectively, as reported in Fig. 3a. In other words, the binary code map can be built based on the defect map (in this case defects of the positive phase were considered). This allows the realization of the corresponding binary code matrix reported in Fig. 3b. Here, we show that the selection of the code matrix pixel size cannot be arbitrary but should be based on morphological characteristics of the pattern. Indeed, given a pattern area, the pixel size should be selected to maximize the randomness and uniqueness of encoded binary matrices. Figure 3c reports the bit uniformity and fractional hamming distance (HD) (i.e., the percentage of bits that differs between two binary code matrices) between different binary code matrices as a function of the pixel size calculated by considering 200 different nanopattern micrographs (area of 2.38 × 2.38 µm²). As can be observed, the mean value of fractional HD of around 0.5 (meaning that code matrices are different and distinguishable) is achieved in correspondence of a pixel size of 238 nm when a bit uniformity of around 0.5 can be observed (i.e., when different binary code matrices have almost equal numbers of 0 and 1) (the relationship between fractional HD in between patterns and bit uniformity of patterns is reported in Fig. 3d). This is because bit uniformity is beneficial to achieve the maximum randomness,¹² as testified by the maximum pattern entropy approaching the ideal value of 1 when bit uniformity is close to 0.5 (Fig. 3e, fractional HD between different code matrices as a function of the entropy of patterns in Supplementary Figure S5). It is noteworthy that, a bit uniformity (and fractional HD) of 0.5 can be observed when the pixel size approximately coincides with the correlation length of BCP patterns (ξ = 207 ± 18 nm, details in Methods). It turns out that through appropriate selection of the pixel size of the binary code matrix (in this case 238 nm), it is possible to obtain binary code matrices with bit uniformity close to the ideal value of 0.5 (in this case 0.52 ± 0.06) (Fig. 3f, Supplementary Figure S6), with unit entropy close to the ideal value of 1 (Supplementary Figure S7), and to achieve a distribution of fractional HD between different binary code matrices with mean value of 0.50 ± 0.06 (Fig. 3g, fractional HD for each couple of binary code matrices in Fig. 3h). These results confirm the high randomness of the binarized code matrix obtained from local nanopattern features. In this context, the encoding capacity of the system c^p, defined as the number of possible responses exhibited by a random pattern where c is the number of responses for each pixel while p is the image area in pixels, is inherently related to the morphological properties of the nanopattern. The encoding capacity of the system can be therefore improved through two strategies: i) by increasing the image area (i.e., increasing the number of pixels without changing the pixel size) and/or ii) by reducing the pixel size of the binarized code matrices while maintaining bit uniformity and fractional inter-HD of ~ 0.5. Concerning the first strategy, Fig. 3i reports the maximum encoding capacity of binary code matrices (c = 2) as a function of the image area (by considering the selected pixel size of 238 nm). Interestingly, an encoding capacity of the same order of magnitude of the world population can be achieved by considering an image with area below 2 µm², while an encoding capacity larger than the number of atoms in the known universe is achievable by considering an image area larger than 12 µm², showing the high density of encoding capacity of the system. Concerning the second strategy that aims to increase the encoding capacity of the system for unit area, it can be achieved by proper selection of molecular weight of involved polymers and processing conditions to increase the defect density while reducing the correlation length of the system⁴⁰ (examples in Supplementary Figure S8). The authentication protocols based on cryptographic codes obtained from fingerprint patterns PUFs can be further refined by considering not only the presence of defects in the pixel area but also defect density (Supplementary Figure S9) and/or defect types (Fig. 3j, additional examples in Supplementary Figure S10).

Authentication/identification in real-world scenarios through computer vision concepts

The comparison of a binarized code matrix with a database code matrix, as required for authentication/identification, implies no misalignment of the image acquired by the end user with the corresponding image stored in the database. Even small translations, rotations and/or deformations of the end user image can cause incorrect authentication/identification due to a different conversion of the fingerprint pattern to the binary code matrix. Indeed, these transformations can cause bit flips in the related binary code matrix due to the different spatial localization of defects (Supplementary Figure S11). Here, we show that the robustness of the authentication/process in real-world scenarios can be enhanced by exploiting matching algorithms based on computer vision concepts, without the need of extracting a binary code matrix from the nanopattern. For this purpose, we synthesized 100 different PUF nanopatterns engraved on a SiO₂ substrate, and we built a database consisting of micrograph images of the corresponding fingerprint patterns acquired by a first scan (details in Methods, Supplementary Figure S12). To obtain a set of images used for testing (test set), in a second scan we collected micrographs of the same PUF patterns. Authentication/identification were tested by comparing an image from the test set with a database image from the database, where the genuine nanopattern is obtained through the fingerprint enhancement algorithm discussed before. Noteworthy, the obtained binarized lamellar patterns inherently endow bit uniformity close to 0.5 (0.496 ± 0.008) and unit entropy close to 1 (values evaluated on the database set of images, details in Supplementary Figure S13). For a given pair of previously binarized images, which we will refer to as database and test image, respectively, we aim at determining whether they represent the same nanopattern. As expected in real-world scenarios, slight shifts, rotations, and different magnifications are present between the database and the test set image of the same nanopattern. Given a database/test pair of images, a set of matching key points between the two images are computed through a Scale-Invariant Feature Transform (SIFT) algorithm (other methods such as Speeded Up Robust Feature, SURF, algorithm and Oriented FAST and Rotated BRIEF, ORB, were tested and give similar results) and a FLANN (Fast Library for Approximate Nearest Neighbors)-based matcher. Then we select a few points providing the best matches (we tested between 10–15, not significantly affecting the results), and use those as a guide to build a homography transformation to map the test image on top of the database one. Examples of feature matching and overlapping of images by considering two images of the same nanopattern (example of successful matching) and on two images of different nanopatterns (example of non-successful matching) are reported in Fig. 4a and b, respectively. If the two images are taken from the same nanopattern and thus only differ by slight deformations (e.g., rotation, shift, etc.), the homographic transformation automatically corrects for these effects, so that the resulting image should overlap with the database one, as can be evaluated by superimposing the test image on the top of the database image through the XOR operation (Fig. 4a, Supplementary Figure S14). On the other hand, if the two images are from unrelated samples, the transformation obtained from the matching algorithm will be unable to return an image with a reasonable overlap with the database (Fig. 4b, Supplementary Fig. 15). The quantification of the superposition is performed by considering the fractional HD calculated on the whole binary image obtained by superimposing the test image on top of the database image through XOR operation (details of the matching algorithm in Methods). Figure 4c reports the fractional HD matrix obtained by comparing images from the test set with the database. As can be observed, HD values closer to 0 can be observed while comparing images of the same PUF (i.e., the test image has a small number of bit values that differs from the database one after the homographic transformation), while HD values of approximately 0.5 can be observed while comparing images of different PUFs. A clear separation of intra and fractional inter-HD distributions can be observed in Fig. 4f, further showing the robustness of the authentication/identification protocol (mean values and std. dev. of inter and intra-HD distributions are 0.14 ± 0.01 and 0.500 ± 0.003, respectively). Here, a HD-based decision threshold of 0.45 can be identified (Supplementary Figure S16). It is worth noticing that the extraction of the binarized fingerprint pattern from the micrograph image through the fingerprint enhancement algorithm provides noise robustness to the authentication process, a crucial aspect for exploiting image based PUFs for real-world applications. In this context, clear separation of fractional intra and inter-HD distributions can be observed also by considering a test set of images acquired by different users operating with different equipment (details in Methods, Supplementary Figure S17 and Supplementary Figure S18), further corroborating the robustness of the authentication/identification protocol.

An analysis of the stability shows that PUFs can be correctly authenticated/identified even after 6 months when the sample was left in normal ambient conditions (Fig. 4d and g), revealing long-term reliable operation. Furthermore, the high thermal stability of PUFs was demonstrated by exposing samples to the harsh conditions of 200°C for 30 minutes, where the PUF also experienced a high heating rate of ~ 15°C s^–1 and a high cooling rate of ~ 0.8°C s^–1. Fractional intra and inter-HD distributions after thermal treatment still show clear separation (Fig. 4e and h). Details on intra and fractional inter-HD mean values and standard deviations evaluated for each test are reported in Supplementary Table 1. In all these cases, successful authentication/identification of all PUF patterns without any false positive or false negative have been achieved by exploiting the decision threshold of 0.45 previously evaluated through the comparison of the test set of images with the database.

Results show that morphological characteristics of artificial fingerprints realized with a bottom-up approach through BCP self-assembly can be exploited for the realization of physical-disorder-based tags for authentication/identification. While global features of the nanopattern, such as correlation length and defect density, can be deterministically controlled by structural and processing parameters (e.g. molecular weights of the BCP, processing temperature and time), each nanopattern is a unique entity with peculiar local features thanks to the randomness arising from thermodynamic instabilities during the self-assembly process. In this framework, results show that the coexistence of deterministic and stochastic features in the same nanopattern can be further explored for increasing the encoding capacity density of the system, as revealed by the intrinsic relationship between the pixel size of the encoded binary code matrix, its bit uniformity and entropy, and the correlation length of the nanopattern. Besides the shown possibility of using additional morphological features such as defect types to further increase the encoding density, additional morphological parameters such as the line-edge roughness, line-width roughness and/or the 3D morphology obtainable through scanning probe microscopies can be explored (example in Supplementary Figure S19). Indeed, since it cannot be excluded a priori that a nanopattern can be cloned by using advanced and expensive lithographic tools with resolution < 10 nm (even if this is unlikely), line-edge roughness, line-width roughness, and the 3D morphology of systems are unlikely to be reproducible. In this context, it is worth noting that recent advancements in the synthesis of BCP with covariant properties allow for on-demand tuning of morphological characteristics (e.g., periodicity, line-edge roughness), long-range order and defectiveness.⁴¹

Differently from previous works where nanoscale tags are affixed to the substrate of interest,^{8,12,15,18–20,42} the here proposed artificial fingerprint is directly engraved on the target substrate and, thus, cannot be removed (unless the substrate is mechanically scratched, but this will leave marks that can be considered as a trace of counterfeiting). As an important consequence, the PUF robustness and environmental stability relies on the physical/chemical/mechanical properties of the engraved substrate. This means that tailored choice of the substrate would enable the realization of PUF devices working even when exposed to harsh conditions (high/low temperatures, exposure to chemicals, mechanical agents, radiations, etc.) as required, for example, for space applications. In addition, these PUFs can be included in conventional tags such as QR codes in the context of a multiple-level security strategy. For anticounterfeiting applications, nanometer-scale artificial fingerprints can be exploited for secure authentication/identification by taking into advantage also that these tags are difficult to localize without an a-priori knowledge of their location. While on one hand the requirement of an electron microscope to challenge the PUF with a scanning focused electron beam can represent an obstacle (despite recent advances in low-cost and portable SEM equipment), on the other hand the requirement of a specialized equipment can further enhance the security of the authentication/identification process in a multiple-level security scenario. Furthermore, it is worth noticing that BCP-templating is compatible with CMOS technology,⁴³ allowing the integration of these PUF devices with conventional electronics. Note that the possibility of realizing fingerprints directly on chips to assess their legitimacy can be crucial to face back doors and hidden functionalities of fraudulent clone chips that, in the framework of global shortage of chips, can threaten global security.⁴⁴ All these characteristics enable artificial fingerprints based on BCP-templated substrates to comply with basic requirements for commercial applications, taking advantage of the intrinsic low-cost of BCP templating on a large scale that easily allows mass production.

In conclusion, we show that artificial fingerprints for counterfeiting applications can be engraved on a wide range of target substrates including - but not limited to - diamond, Si, SiO₂ and quartz to realize disorder-based PUFs readable through electron microscopy. By proposing a user-independent strategy for successfully extracting the nanopattern even from low-quality micrographs, we show that engraved nanopatterns can be encoded in binary code matrices endowing high bit uniformity and uniqueness. In this context, we show the importance of an appropriate encoding of nanopattern in the corresponding binary code matrix depending on morphological nanopattern features, including defects and correlation length. In addition, we propose a computer vision-based strategy that enables robust authentication/identification processes, as demonstrated by testing 100 PUF devices. These results shed new light on the realization of PUF that embraces the inherent stochasticity of nanoscale self-assembled materials as a randomness source.

Fabrication of self-assembled artificial fingerprint patterns by block copolymers

The surface of different materials (silicon, silicon oxide, quartz, diamond) were functionalized by RCP grafting process to promote the perpendicular orientation of lamellar BCP nanostructures. First, the substrates were cleaned in an ultrasonic bath in acetone followed by isopropyl alcohol, and functionalized by O₂ plasma treatment at 130 W for 20 min. Then, a solution of RCP (18 mg in 2 ml of toluene) was spincasted for 60 s at 3000 rpm onto the functionalized substrates. The grafting process was performed in a rapid thermal processing (RTP) machine Jipelec JetFirst200 at high temperature (T_a = 290°C) for an annealing time (t_a) of 300 s, in a N₂ environment with a heating rate of 15°C s^–1. Automatic cooling to room temperature was set to 240 s. The non-grafted polymeric chains were then removed by sonication in toluene for 6 min, resulting in a final grafted RCP layer thickness of ∼7 nm, as measured by spectroscopic ellipsometry (alpha-SE ellipsometer from J.A. Wollam Co.). BCP solution (18 mg in 2 ml of toluene) was then spincasted over the RCP functionalized surface at 3000 rpm for 60 s resulting in a total BCP thickness of 35 nm. The self-assembly was promoted by RTP at 230°C for 600 s in a N₂ environment with a heating ramp of 15°C s^–1 and automatic cooling to room temperature for 240 s. Selective removal of the PMMA phase of self-assembled BCP was achieved by exposure of the samples to ultraviolet radiation (5 mW cm^–2, λ = 253.7 nm) for 180 s followed by isotropic O₂ plasma etching (40 W for 30 s).

The synthesis of α-hydroxy ω-Br polystyrene-stat-polymethyl methacrylate (PS-stat-PMMA) random copolymer (RCP) with M_w = 14.60 kg mol^–1, styrene fraction (f_PS) of 0.59 and polydispersity index (PDI) of 1.30, is described elsewhere.⁴⁰ Lamellar-forming polystyrene-block-polymethyl methacrylate (PS-b-PMMA) BCP with M_w = 66 kg mol^–1, f_PS = 0.50 and PDI = 1.09, was purchased from Polymer Source Inc. and used without further purification. Toluene (99.8% anhydrous), acetone (99.5% anhydrous) and isopropyl alcohol (98% anhydrous) were purchased from Sigma Aldrich. Polymethyl methacrylate (PMMA) A4 positive electron-beam resist was purchased from MicroChem. Methyl isobutyl ketone (MIBK) developer was purchased from KemLab.

Engraving artificial nano fingerprints on target substrates and materials

The pattern transfer of the fingerprint pattern onto the substrate surface was performed by reactive ion etching (RIE) with different chemistries depending on the etched material. Silicon etching was achieved by SF₆/C₄F₈ RIE with ICP plasma of 750 W and RF table of 35 W. Silicon oxide and quartz etching was achieved by CHF₃/Ar RIE and power of 35 W. Diamond etching was achieved by O₂ RIE with ICP plasma of 2000 W and RF table of 200 W.

Morphological characterization of nanopatterns

The micrographs of artificial fingerprints were acquired by FEI Inspect-F field emission gun scanning electron microscope (FEG-SEM) using an Everhart-Thornley secondary electron detector (ETD). Additionally, a dual-beam SEM FEI Quanta 3D FEG was employed for authentication/ identification of artificial fingerprints, enabling the collection of the SEM micrographs through alternative instrumentation.

Fingerprint enhancement algorithm for nanopattern binarization

The fingerprint enhancement algorithm exploited for pattern binarization was adapted from the algorithm described in ref.³⁷ by exploiting the fingerprint_enhancement python library. Parameters of the enhance_Fingerprint function were adapted on the basis of the pitch of the lamellar structures, obtained from the Fourier transform of the corresponding nanopattern image, and the resolution of the SEM image. The fingerprint enhancement binarization was compared to standard binarization techniques implemented through auto-local thresholding methods with default parameters of ADAblock.³⁸

Characterization of artificial nano fingerprints properties

Spatial coordinates, types, and density of topological defects of binarized nanopatterns, as well as the analysis of correlation length, was performed through the automated defect analysis tool ADAblock described in ref.³⁸, modified to be integrated into an automated analysis process including the binarization through fingerprint enhancement algorithm. The analysis of Fig. 3 was performed over 200 nanopattern SEM images with an original area of 2.64 × 3.08 µm² acquired over different areas of the sample, while the correlation length of nanopatterns were evaluated by considering 11 images with an area of 4.14 x 3.56 µm² of the same nanopatterned sample (here, only images with a reduced magnification were used to have a higher grain size/image dimension ratio as required for the analysis of correlation length). The binary code matrices have been obtained from 200 SEM images of nanopattern micrographs by considering a central square area of the SEM image of 2.38 × 2.38 µm². The analysis of Fig. 2 was performed over about 30 images for each binarization technique, in fact, a couple of these images were discarded. The defect density reported in Fig. 2e is expressed as the median value of the density for the different defect types evaluated over about 30 images.

The bit uniformity (${p}_{i}$) of the i-th binarized code matrix, defined as the distribution of 1s and 0s in the bit stream, have been calculated through the equation:

$${p}_{i}=\frac{1}{n}\sum _{k=1}^{n}{R}_{i,k}$$

where ${R}_{k}$ is the binary bit response of the $k$-th pixel of the $i$-th binarized code matrix, while $n$ is the number of pixels of the code matrix. A value of 1 means that all responses ${R}_{i,k}$ of the $i$-th pattern are logic-1. The unit entropy of the i-th binarized code matrix have been evaluated through the Eq. 1⁹:

$${E}_{i}= -\left[{p}_{i}{\text{log}}_{2}{p}_{i}+(1-{p}_{i}){\text{log}}_{2}{p}_{i}(1-{p}_{i})\right]$$

Fabrication of PUF devices based on artificial fingerprints at the nanoscale

Nanoscale PUF devices with dimensions of 3 µm x 3 µm were realized by creating a metallic frame surrounding the area of interest of the previously nanopatterned substrate, as shown in Supplementary Figure S12. For this purpose, the area of interest was defined by electron-beam lithography (EBL) in a dual-beam SEM FEI Quanta 3D FEG on PMMA A4 positive-tone resist on the patterned SiO₂ substrate. After the development with MIBK:IPA (3:1) developer solution, a double layer of Ti/Au (2 nm and 3 nm of thickness, respectively) was deposited by RF sputtering followed by a lift-off process under sonication.

Computer vision-based matching algorithm

The matching algorithm has been implemented by exploiting computer vision algorithms of OpenCV python library. As the first step, the image was cropped of 1.04 µm on each side of the SEM image to remove the Au frame of the PUF device that can interfere with the binarization process. The resulting image has an area of 1.47 x 2.06 µm² or 1.49 x 2.07 µm² on which to perform the following analysis. Then, the homographic transformation, required for overlapping the test image on top of the database image, was calculated through SIFT (SURF or ORB were tested to give similar results) and a FLANN-based matcher from OpenCV python library. Note that, after performing the homographic transformation, the transformed image (no longer binary) is binarized to 1 and 0 values through a thresholding technique, where the threshold was chosen halfway between the minimum and the maximum values. Then, quantification of the superposition was performed by calculating the fractional HD between considered images (calculation was performed over 987 x 708 or 996 x 717 pixels). Overlapped images in Fig. 4a and b have been obtained by superimposing test image after homographic transformation and the database image with XOR operation between the two. Note that XOR operation returns a 0-pixel value if the corresponding pixel of the test and database image are characterized by the same bit value, 1-pixel value otherwise. In these terms, an overlapped image with a high number of 0 values represents good overlapping between test and database images. Note that performing fractional HD is nothing but performing the XOR operation over two images, then summing all pixel values of the obtained overlapping image and normalizing over the number of pixels. Given the sharp nature of the binarized images, we typically obtain very clean cancellations when a good match is found, consistently leading to a clear separation of the fractional HD between good and bad matches.

Tests of PUF robustness

The thermal stability of PUF devices was tested by subjecting the samples to a heat treatment by the rapid thermal annealing at 200°C for 30 minutes. The annealing was performed in a RTP Jipelec JetFirst200 in N₂ environment and heating rate of 15°C s^–1 which puts the device into even harsher conditions when compared to conventional furnace heating. Then, an automatic cooling to room temperature was set to 240 s. Aging stability was evaluated by collection of the SEM images of the same set of devices after 6 months from its fabrication, by storing devices at normal ambient conditions.

Author contributions

I.M, C.M, F.F.L and G.M. conceptualized the idea and designed experiments. I.M performed sample preparation and experimental characterization. C.M. developed codes for image analysis and characterization of PUF performances. I.M. and C.M. supported by F.F.L. and G.M. performed data analysis. M.F. supported experimental activities. S.C. supported code development. F.F.L and G.M. provided funding and supervised the research. I.M., C.M., F.F.L., G.M. wrote the manuscript. All authors participated in the discussion of results and revision of the manuscript.

Competing interests

The device configuration and implementation method of artificial fingerprints are currently under patent filing (Italian priority application number 102024000006367). Patent applicants: Politecnico di Torino, Istituto Nazionale di Ricerca Metrologica. Inventors: I.M, C.M., F.F.L and G.M.

Acknowledgements

Part of this work was supported by the European project MEMQuD, code 20FUN06. This project (EMPIR 20FUN06 MEMQuD) has received funding from the EMPIR programme co-financed by the Participating States and from the European Union’s Horizon 2020 research and innovation programme. Part of this work was supported by the European project OpMetBat, code 21GRD01. The project has received funding from the European Partnership on Metrology, cofinanced from the European Union’s Horizon Europe Research and InnovationProgramme, and by ParticipatingStates. Part of this work has been carried out at Nanofacility Piemonte INRiM, a laboratory supported by the ‘‘Compagnia di San Paolo’’ Foundation, and at the QR Laboratories, INRiM. Part of this work was supported by the European Union─NextGenerationEU under the National Recovery and Resilience Plan (NRRP), Mission 04 Component 2 Investment 3.1|Project Code: IR0000027–CUP: B33C22000710006–iENTRANCE@ENL: Infrastructure for Energy TRAnsition aNd Circular Economy @EuroNanoLab. The authors acknowledge Dr. Bruno Torre (Politecnico di Torino) for his kind support on AFM measurements.

Fink, C., Maskus, K. E. & Qian, Y. The economic effects of counterfeiting and piracy: A review and implications for developing countries. World Bank Research Observer 31, 1–28 (2016).
Organisation for Economic Co-operation and Development. Global Trade in Fakes. (OECD, 2021). doi:10.1787/74c81154-en.
Shamsoshoara, A., Korenda, A., Afghah, F. & Zeadally, S. A survey on physical unclonable function (PUF)-based security solutions for Internet of Things. Computer Networks vol. 183 107593 Preprint at https://doi.org/10.1016/j.comnet.2020.107593 (2020).
Pappu, R., Recht, B., Taylor, J. & Gershenfeld, N. Physical One-Way Functions. Science (1979) 297, 2026–2030 (2002).
Gao, Y., Al-Sarawi, S. F. & Abbott, D. Physical unclonable functions. Nature Electronics vol. 3 81–91 Preprint at https://doi.org/10.1038/s41928-020-0372-5 (2020).
Arppe, R. & Sørensen, T. J. Physical unclonable functions generated through chemical methods for anti-counterfeiting. Nat Rev Chem 1, 0031 (2017).
Aono, M. & Ariga, K. The Way to Nanoarchitectonics and the Way of Nanoarchitectonics. Advanced Materials 28, 989–992 (2016).
Sun, N. et al. Random fractal-enabled physical unclonable functions with dynamic AI authentication. Nat Commun 14, 2185 (2023).
Ma, T. et al. Dynamic wrinkling pattern exhibiting tunable fluorescence for anticounterfeiting applications. Nat Commun 11, 1811 (2020).
Ren, W., Lin, G., Clarke, C., Zhou, J. & Jin, D. Optical Nanomaterials and Enabling Technologies for High‐Security‐Level Anticounterfeiting. Advanced Materials 32, (2020).
Abdollahi, A., Roghani-Mamaqani, H., Razavi, B. & Salami-Kalajahi, M. Photoluminescent and Chromic Nanomaterials for Anticounterfeiting Technologies: Recent Advances and Future Challenges. ACS Nano 14, 14417–14492 (2020).
Chen, P. et al. Programmable Physical Unclonable Functions Using Randomly Anisotropic Two-Dimensional Flakes. ACS Nano 17, 23989–23997 (2023).
Li, Q. et al. Physical Unclonable Anticounterfeiting Electrodes Enabled by Spontaneously Formed Plasmonic Core–Shell Nanoparticles for Traceable Electronics. Adv Funct Mater 31, (2021).
Hu, S. et al. Three-Dimensionally Printed, Vertical Full-Color Display Pixels for Multiplexed Anticounterfeiting. Nano Lett 23, 9953–9962 (2023).
Zhang, J. et al. An all-in-one nanoprinting approach for the synthesis of a nanofilm library for unclonable anti-counterfeiting applications. Nat Nanotechnol 18, 1027–1035 (2023).
Gu, Y. et al. Gap-enhanced Raman tags for physically unclonable anticounterfeiting labels. Nat Commun 11, 516 (2020).
Huo, Y., Yang, Z., Wilson, T. & Jiang, C. Recent Progress in SERS‐Based Anti‐counterfeit Labels. Adv Mater Interfaces 9, (2022).
Lu, Y. et al. Dynamic Cryptography through Plasmon‐Enhanced Fluorescence Blinking. Adv Funct Mater 32, (2022).
Dodda, A. et al. Graphene-based physically unclonable functions that are reconfigurable and resilient to machine learning attacks. Nat Electron 4, 364–374 (2021).
Kim, J. H. et al. Nanoscale physical unclonable function labels based on block copolymer self-assembly. Nat Electron 5, 433–442 (2022).
Yang, G. G. et al. Intelligent block copolymer self-assembly towards IoT hardware components. Nature Reviews Electrical Engineering 1, 124–138 (2024).
Crawford, H. The Sumerian World. (Routledge, 2013). doi:10.4324/9780203096604.
Marasco, E., Albanese, M., Patibandla, V. V. R., Vurity, A. & Sriram, S. S. Biometric multi‐factor authentication: On the usability of the FingerPIN scheme. SECURITY AND PRIVACY 6, (2023).
Weaver, A. C. Biometric Authentication. Computer (Long Beach Calif) 39, 96–97 (2006).
Pankanti, S., Prabhakar, S. & Jain, A. K. On the individuality of fingerprints. IEEE Trans Pattern Anal Mach Intell 24, 1010–1025 (2002).
Bates, F. S. & Fredrickson, G. H. Block Copolymer Thermodynamics: Theory and Experiment. Annu Rev Phys Chem 41, 525–557 (1990).
Kim, H.-C., Park, S.-M. & Hinsberg, W. D. Block Copolymer Based Nanostructures: Materials, Processes, and Applications to Electronics. Chem Rev 110, 146–177 (2010).
Choi, J. W., Kim, M., Safron, N. S., Arnold, M. S. & Gopalan, P. Transfer of Pre-Assembled Block Copolymer Thin Film to Nanopattern Unconventional Substrates. ACS Appl Mater Interfaces 6, 9442–9448 (2014).
Lane, A. P. et al. Directed Self-Assembly and Pattern Transfer of Five Nanometer Block Copolymer Lamellae. ACS Nano 11, 7656–7665 (2017).
Tu, K.-H. et al. Universal pattern transfer methods for metal nanostructures by block copolymer lithography. Nanotechnology 26, 375301 (2015).
Cheng, J. Y. et al. Formation of a Cobalt Magnetic Dot Array via Block Copolymer Lithography. Advanced Materials 13, 1174–1178 (2001).
Choi, D.-G. et al. Magnetic nanodot arrays patterned by selective ion etching using block copolymer templates. Nanotechnology 15, 970–974 (2004).
Tseng, Y.-C., Peng, Q., Ocola, L. E., Elam, J. W. & Darling, S. B. Enhanced Block Copolymer Lithography Using Sequential Infiltration Synthesis. The Journal of Physical Chemistry C 115, 17725–17729 (2011).
Park, M., Harrison, C., Chaikin, P. M., Register, R. A. & Adamson, D. H. Block Copolymer Lithography: Periodic Arrays of ~1011 Holes in 1 Square Centimeter. Science (1979) 276, 1401–1404 (1997).
Park, S., Lee, D. H. & Russell, T. P. Self‐Assembly of Block Copolymers on Flexible Substrates. Advanced Materials 22, 1882–1884 (2010).
Kim, J. Y. et al. Flexible and Transferrable Self‐Assembled Nanopatterning on Chemically Modified Graphene. Advanced Materials 25, 1331–1335 (2013).
Lin Hong, Yifei Wan & Jain, A. Fingerprint image enhancement: algorithm and performance evaluation. IEEE Trans Pattern Anal Mach Intell 20, 777–789 (1998).
Murphy, J. N., Harris, K. D. & Buriak, J. M. Automated Defect and Correlation Length Analysis of Block Copolymer Thin Film Nanopatterns. PLoS One 10, e0133088 (2015).
Porti, M., Redón, M., Muñoz, J., Nafría, M. & Miranda, E. Oxide Breakdown Spot Spatial Patterns as Fingerprints for Optical Physical Unclonable Functions. IEEE Electron Device Letters 44, 1600–1603 (2023).
Ferrarese Lupi, F. et al. Tailored and Guided Dewetting of Block Copolymer/Homopolymer Blends. Macromolecules 53, 7207–7217 (2020).
Feng, H. et al. Optimized design of block copolymers with covarying properties for nanolithography. Nat Mater 21, 1426–1433 (2022).
Kiremitler, N. B. et al. Tattoo‐Like Multi‐Color Physically Unclonable Functions. Adv Opt Mater 2302464, (2023).
Liu, C.-C. et al. Directed self-assembly of block copolymers for 7 nanometre FinFET technology and beyond. Nat Electron 1, 562–569 (2018).
Villasenor, J. & Tehranipoor, M. Chop shop electronics. IEEE Spectr 50, 41–45 (2013).

Yes there is potential Competing Interest. The device configuration and implementation method of artificial fingerprints are currently under patent filing (Italian priority application number 102024000006367). Patent applicants: Politecnico di Torino, Istituto Nazionale di Ricerca Metrologica. Inventors: I.M, C.M., F.F.L and G.M.

SupplementaryMaterial.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Artificial fingerprints engraved through block-copolymers as nanoscale physical unclonable functions for authentication and identification

Status:

Version 1

Abstract

Figures

Introduction

Results

Artificial fingerprints at the nanoscale engraved on target substrates

Extraction of pattern morphology

Binary code matrices encoding of artificial fingerprints

Authentication/identification in real-world scenarios through computer vision concepts

Discussion

Conclusions

Methods

Fabrication of self-assembled artificial fingerprint patterns by block copolymers

Engraving artificial nano fingerprints on target substrates and materials

Morphological characterization of nanopatterns

Fingerprint enhancement algorithm for nanopattern binarization

Characterization of artificial nano fingerprints properties

Fabrication of PUF devices based on artificial fingerprints at the nanoscale

Computer vision-based matching algorithm

Tests of PUF robustness

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1