AI-Driven Technology Fights Counterfeit Chips with Optical Detection

Researchers in the United States have made significant strides in combating the growing issue of counterfeit semiconductor devices by introducing a cutting-edge optical anticounterfeit method powered by artificial intelligence (AI).

Counterfeit chips have become a major concern within the semiconductor supply chain, with estimates suggesting the problem is valued at over $75 billion. To address this challenge, a team of experts at Purdue University has put forth an innovative optical anti-counterfeit detection technique that leverages AI to differentiate between natural aging processes and malicious tampering.

Their groundbreaking approach, known as Residual, Attention-based Processing of Tampered Optical Responses (RAPTOR), focuses on identifying tampering by analyzing intricate gold nanoparticle patterns that are intricately embedded on semiconductor chips.

Various methods have been employed in the past to ensure the authenticity of semiconductor devices, often involving the integration of physical security tags into the chip's functionality or packaging. Central to many of these techniques are physical unclonable functions (PUFs), which are unique physical systems that are challenging to replicate due to economic constraints or inherent physical properties.

Optical PUFs, which rely on the distinct optical responses of random media, have emerged as a promising solution. These systems are relatively easy to fabricate and offer rapid measurement capabilities, making them ideal for conducting tampering identification experiments. Metallic optical systems at the nano-scale have gained popularity for their strong scattering response at optical wavelengths, enhancing resilience during post-tampering assessments.

Despite the advantages of optical PUFs, achieving scalability and ensuring accurate discrimination between adversarial tampering and natural degradation remains a significant hurdle. Factors such as physical aging at elevated temperatures, packaging abrasions, and humidity impacts can complicate the detection process.

The robustness of the optical PUF technique extends to various adversarial tampering scenarios, including malicious package abrasions, compromised thermal treatments, and adversarial tearing. To develop their method, the research team curated a dataset of 10,000 images featuring randomly distributed gold nanoparticles, which were then processed using advanced image analysis techniques.

In their experiments, the team simulated tampering behaviors in the nanoparticle PUFs, considering both natural variations and deliberate adversarial tampering. RAPTOR prioritizes nanoparticle correlations across different samples, enabling it to effectively detect tampering instances with remarkable accuracy.

Overall, RAPTOR demonstrated exceptional performance, achieving a detection rate of 97.6 percent under worst-case tampering scenarios. This success surpasses previous methods and underscores the potential of advanced optical anticounterfeit technologies in safeguarding semiconductor devices against fraudulent activities.