Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Photonic Physically Unclonable Function's Resilience to Multiple-Valued Machine Learning Attacks

Jessie M. Henderson, Elena R. Henderson, Clayton A. Harper, Hiva Shahoei, William V. Oxford, Eric C. Larson, Duncan L. MacFarlane, Mitchell A. Thornton

TL;DR

This work investigates this photonic PUF's susceptibility to Multiple-Valued-Logic-based machine learning attacks and finds that approximately 1,000 CRPs are necessary to train models that predict response bits better than random chance.

Abstract

Physically unclonable functions (PUFs) identify integrated circuits using nonlinearly-related challenge-response pairs (CRPs). Ideally, the relationship between challenges and corresponding responses is unpredictable, even if a subset of CRPs is known. Previous work developed a photonic PUF offering improved security compared to non-optical counterparts. Here, we investigate this PUF's susceptibility to Multiple-Valued-Logic-based machine learning attacks. We find that approximately 1,000 CRPs are necessary to train models that predict response bits better than random chance. Given the significant challenge of acquiring a vast number of CRPs from a photonic PUF, our results demonstrate photonic PUF resilience against such attacks.

A Photonic Physically Unclonable Function's Resilience to Multiple-Valued Machine Learning Attacks

TL;DR

This work investigates this photonic PUF's susceptibility to Multiple-Valued-Logic-based machine learning attacks and finds that approximately 1,000 CRPs are necessary to train models that predict response bits better than random chance.

Abstract

Physically unclonable functions (PUFs) identify integrated circuits using nonlinearly-related challenge-response pairs (CRPs). Ideally, the relationship between challenges and corresponding responses is unpredictable, even if a subset of CRPs is known. Previous work developed a photonic PUF offering improved security compared to non-optical counterparts. Here, we investigate this PUF's susceptibility to Multiple-Valued-Logic-based machine learning attacks. We find that approximately 1,000 CRPs are necessary to train models that predict response bits better than random chance. Given the significant challenge of acquiring a vast number of CRPs from a photonic PUF, our results demonstrate photonic PUF resilience against such attacks.
Paper Structure (9 sections, 4 figures)

This paper contains 9 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Background
  3. A Photonic PUF
  4. Brief Introduction to Machine Learning-Based Attacks for Conventional PUFs
  5. Attacking the Photonic PUF
  6. Modeling the PUF
  7. The Network, Training Phase, and Evaluation Phase
  8. Results and Discussion
  9. Conclusion and Future Work

Figures (4)

  • Figure 1: Overview of the PIC PUF layout. Note that the PUF described in this work uses only one of the two available outputs to generate CRPs.
  • Figure 2: The $24$-cell photonic PUF architecture.
  • Figure 3: Overview of the neural network architecture, the training phase, and the evaluation phase. The network accepts $D_c$-digit responses encoded using radix-$R_c$ and uses six layers (described in detail in the main text) to predict the $D_r$-digit response represented using radix-$R_r$. During training, the loss is computed as the mean-squared error between the ground truth response represented as $D_r$-digit, radix-$R_r$ value and the $D_r$-digit response prediction. During evaluation, the $D_r$-digit response prediction (which includes continuous or "Cont." values) is rounded. The resulting $D_r$-digit response comprised of integer values is then converted to a base-10 integer, which is converted to a $24$-bit response vector that is compared to the ground truth $24$-bit response to assess prediction accuracy.
  • Figure 4: Response prediction accuracy for $6$,$600$ models with challenges encoded using three different radices. Each reported accuracy is the likelihood of correctly predicting any bit in the test set averaged across validation folds and the four PUFs. Upward-pointing triangles indicate the maximum individual bit accuracy over all PUFs and folds for the given challenge/response radix combination. Downward-pointing triangles indicate the minimum accuracy for the same.