Bi₂O₂Se-Based Bimode Noise Generator for the Application of Generative Adversarial Networks.

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Generative Adversarial Networks (GANs) rely on noise for unique and robust image generation.
• Hardware-based noise generation offers greater unpredictability than software methods by incorporating physical entropy sources.
• Bismuth oxy-selenide (Bi2O2Se) is a promising material due to its electronic and optoelectronic properties.

Purpose of the Study:

• To demonstrate bimode Bi2O2Se-based noise generators for GAN applications.
• To explore the potential of Bi2O2Se in creating diverse and secure generated images.
• To analyze different physical noise sources integrated with Bi2O2Se.

Main Methods:

• Fabrication of bimode Bi2O2Se-based noise generators.
• Characterization of noise in photodetector (black current) and memristor (random telegraph noise) modes.
• Application of Markov chain with K-means clustering to analyze noise states and transitions.
• Evaluation of GAN performance using inception score and Fréchet inception distance.

Main Results:

• Successful demonstration of Bi2O2Se as a material platform for hardware noise generation.
• Identification and characterization of distinct noise modes (black current and random telegraph noise).
• Quantification of noise states and transition probabilities using advanced statistical methods.
• Validation of the generated image properties using standard GAN evaluation metrics.

Conclusions:

• Bi2O2Se is a versatile material for developing hardware noise generators for GANs.
• Hardware noise generation using Bi2O2Se enhances the unpredictability and quality of GAN-generated images.
• This approach offers a pathway to more secure and diverse AI-driven image synthesis.