Food-Grade Physically Unclonable Functions

Counterfeit products in the pharmaceutical and food industries have posed an overwhelmingly increasing threat to the health of individuals and societies. An effective approach to prevent counterfeiting is the attachment of security labels directly on drugs and food products. This approach requires the development of security labels composed of safely digestible materials. In this study, we present the fabrication of security labels entirely based on the use of food-grade materials. The key idea proposed in this study is the exploitation of food-grade corn starch (CS) as an encoding material based on the microscopic dimensions, particulate structure, and adsorbent characteristics. The strong adsorption of a food colorant, erythrosine B (ErB), onto CS results in fluorescent CS@ErB microparticles. Randomly positioned CS@ErB particles can be obtained simply by spin-coating from aqueous solutions of tuned concentrations followed by transfer to an edible gelatin film. The optical and fluorescence microscopy images of randomly positioned particles are then used to construct keys for a physically unclonable function (PUF)-based security label. The performance of PUFs evaluated by uniformity, uniqueness, and randomness analysis demonstrates the strong promise of this platform. The biocompatibility of the fabricated PUFs is confirmed with assays using murine fibroblast cells. The extremely low-cost and sustainable security primitives fabricated from off-the-shelf food materials offer new routes in the fight against counterfeiting.

to the total number of cells from left to right is 6.1%, 6.4%, and 7.9%, respectively.This rate is 5.8% for the control group.

Extraction of binary keys:
MATLAB was used to extract binary keys from images.The acquired images were decomposed by image processing.Each image was then used to create bitmaps.Von-Neumann de-biasing was applied.The threshold was applied on images with a pixel size of 2752 × 2208.The image size was then reduced to 16×16 pixels.This image was then digitized to obtain a 256-bit key consisting of 0-bits and 1-bits.This procedure was repeated for all images (Figure S2).To Table S1.Key extraction steps

Uniformity
The uniformity for each key was calculated with the following formula.
n: the number of bits rl: lth binary bit (0 or 1) of an n-bit response from a key

Uniqueness
Uniqueness was calculated using 31 keys obtained from the images presented in Figure S11.

Classic von Neumann Debiasing
The following procedure was used for debiasing • Consider the key is composed of consecutive pairs of bits • Discard the pair of bits in the case of bits consisting of 11 or 00 • Retain the first bit, in the case of bits consisting of 10 or 01       Cost analysis of the Food-Grade PUFs:

Figure S2 .
Figure S2.Optical microscopy images of the surfaces following the spin-coating of dispersions.The weight/volume ratio of corn starch to water a) 1/5, b) 1/10 and c) 1/100.Scale bars are 100 μm.

Figure S3 Figure S4 .
Figure S3 Optical microscope images of the substrates before and after the transfer process.a) The silicon substrate before the transfer process, b) Silicon substrate after transfer of the starch particles to the gelatin substrate.c) Gelatin substrate after the transfer process.Scale bars are 100 μm.

Figure S5 .Figure S6 .
Figure S5.Imaging demonstration with a handheld microscope.a) Photograph of the handheld microscope (Dino-Lite).b) Bright field and c) fluorescence images acquired from PUF labels via the handheld microscope.Scale bars are 200 µm.

Figure S8 .
Figure S8.XRD patterns a) and TG curve b) of ErB, CS and CS@ErB.

Figure S9 .
Figure S9.Fluorescence microscope images of live and dead cells.Green fluorescent cells are alive, whereas red cells are dead.For the images, the ratio of the number of damaged/dead cells

Figure S10 .
Figure S10.Biocompatibility against reactive oxygen species generated by light exposure as a function of CS@ErB concentration.a) Cell viability, b) optical microscope images.
further verify the randomness of the extracted bits, National Institute of Standards and Technology (NIST) randomness test suite was used.The keys are considered random when the p-value ≥0.01.
number of keys from different chips HD (Ri, Rj) : Hamming distance between chips i and j Normalized Hamming distance = Hamming distance divided by the number of bits, (  ,  )  In the uniqueness formula, the sum of normalized Hamming distances is divided by the total number of comparisons of among different chips, HD mean value for different samples σ: the SD of the different samples

Figure S11 .
Figure S11.The definition of markers to determine the exact location of the regions, where optical and fluorescence microscope images are taken.The marker was fabricated by physical vapor deposition of ZnO through a stencil mask.

Figure S13 .
Figure S13.Representative images acquired from a selected region by varying illumination conditions: exposure in the range of 250 ms to 600 ms, gamma value 1.The keys extracted from these images were used in the calculation of HDINTRA.Scale bars are 100 µm

Figure S15 .
Figure S15.Long-term stability test of a PUF sample.Bright field, fluorescence images and corresponding binary keys extracted from fluorescence images of a sample stored for 4 months.Data zones of the images have been intentionally left to indicate the elapsed time.Fluorescence images taken at the same exposure time (600 ms).The similarity ratio between the bit sequences generated before and after the test is 98.8%.

Figure S16 .
Figure S16.Daylight stability of PUFs.Fluorescence microscope images corresponding extracted binary keys before (left) and after 1 day exposure to day light.Scale bars are 100 μm.The similarity ratio between the keys before and after the test is 96%.

Figure S17 .
Figure S17.Abrasion stability of PUFs.Fluorescence microscope images and corresponding binary keys before (left) and after the abrasion test.Specifically, the sample was glued under the weight of 200 g and moved 30 cm against an aluminum foil.Scale bars are 50 μm.The similarity ratio between the keys before and after the test is 96%.

Table S2 .
The NIST randomness test results for keys generated from food-grade PUFs.