Machine Learning Attacks on Optical Physical Unclonable Functions

[摘要] Traditional security algorithms for authentication and encryption rely heavily on the digital storage of secret information (e.g. cryptographic key), which is vulnerable to copying and destruction. An attractive alternative to digital storage is the storage of this secret information in the intrinsic, unpredictable, and non-reproducible features of a physical object. Such devices are termed physical unclonable functions (PUFs), and recent research proves that PUFs can resolve the vulnerabilities associated with digital key storage while otherwise maintaining the same level of security as traditional methods. Modern cryptographic algorithms rest on the shoulders of this one-way principle in certain mathematical algorithms (e.g. RSA or Rabin functions). However, a key difference between PUFs and traditional one-way algorithms is that conventional algorithms can be duplicated.Here, we investigate a silicon photonic PUF a novel cryptographic device based on ultrafast and nonlinear optical interactions within an integrated silicon photonic cavity. This work reviews the important properties of this device including high complexity of light interaction with the material, unpredictability of the response and ultrafast generation of private information. We further explore the resistance of silicon photonic PUFs against numerical modeling attacks and demonstrate the influence of cavity’s inherent nonlinear optical properties on the success of such attacks. Finally, we demonstrate encrypted data storage and compare the results of decryption using a genuine silicon PUF device the ;;clone” generated by the numerical algorithm.Finally, we provide similar analysis of modeling attacks on another well-known type of optical PUF, called the optical scattering PUF (OSPUF). While not as compatible with integration as the silicon photonic PUF, the OSPUF system is known to be extremely strong and resistant to adversarial attacks. By attacking a simulated model of OSPUF, we attempt to present the underlying reasons behind the strong security of this given device and how this security scales with the OSPUFs physical parameters.

[发布机构] Johns Hopkins University

[效力级别] Cryptography

[关键词] Physical Unclonable Function;Cryptography;One-Way Function;photonic PUF;scattering PUF;Physics