RAPTOR technology outperforms traditional tamper detection methods by up to 40%.
Purdue University engineers led by Alexander Kirdyshev have developed an optical counterfeit detection method, RAPTOR, that leverages deep learning to identify unauthorized tampering of chips used in semiconductor devices. Their research was published in the peer-reviewed journal Advanced Photonics. (DALL-E 3 OpenAI image)
WEST LAFAYETTE, Ind. — Researchers at Purdue University’s School of Engineering have developed a patent-pending optical method to detect counterfeit chips used in semiconductor devices.
Purdue’s method, called RAPTOR (Residual Attention Based Processing of Tampered Optical Responses), leverages deep learning to identify tampering, improving on previous methods that struggle with scalability and distinguishing between natural degradation and adversarial tampering.
Alexander Kirdyshev, the Elmore Family Professor in the Department of Electrical and Computer Engineering, led the team that published the research in the peer-reviewed journal Advanced Photonics.
“Our plan offers a great opportunity for the adoption of deep learning-based anti-counterfeiting methods in the semiconductor industry,” he said.
Kildishev has disclosed RAPTOR to Purdue Innovates’ Office of Technology Commercialization, which is filing patent applications to protect the intellectual property. Industry partners interested in developing or commercializing RAPTOR should contact Will Buchanan, assistant director for business development and licensing in Physical Sciences, at wdbuchanan@prf.org inquiring about track code 70652.
Shortcomings in detecting counterfeit chips
Kirdyshev said the semiconductor industry has grown to a $500 billion global market over the past 60 years, but it faces the dual challenge of a severe shortage of new chips and a proliferation of counterfeit chips, posing a greater risk of failure and unwanted surveillance.
“The latter has inadvertently created a $75 billion market for counterfeit chips, endangering the safety and security of multiple sectors that rely on semiconductor technology, including aviation, communications, quantum, artificial intelligence and personal finance,” he said.
Kirdyshev said several technologies have been developed to verify the authenticity of semiconductors and detect counterfeit chips.
“These techniques primarily leverage physical security tags embedded into the chip’s features or packaging,” he said. “Central to many of these methods is a physically unclonable function (PUF) – a unique physical system that is difficult for an adversary to replicate due to economic constraints or inherent physical properties.”
Optical PUFs, which exploit the distinct optical response of random media, are particularly promising.
“But there are significant challenges in achieving scalability and continuing to accurately distinguish between adversarial tampering and natural degradation such as physical degradation at high temperatures, packaging wear, and the effects of humidity,” Kirdyshev said.
Purdue University’s RAPTOR Creation
Kildishev and his team drew inspiration for RAPTOR from the capabilities of deep learning models.
“RAPTOR is a novel deep learning technique that analyzes patterns of gold nanoparticles embedded in chips to identify tampering,” he said. “It is robust against adversarial tamper signatures such as malicious packaging scratches, improper heat treatment, and adversarial tearing.”
Yuheng Chen, a doctoral student in Kirdyshev’s group, said RAPTOR uses distance matrix validation of gold nanoparticles.
“The gold nanoparticles are randomly and uniformly distributed on the chip sample substrate, but their radii are normally distributed. Characterization by dark-field microscopy creates a unique database of randomly positioned dark-field images,” he said. “Gold nanoparticles can be easily measured using dark-field microscopy. It is a readily available technology that can be seamlessly integrated into any stage of the semiconductor manufacturing pipeline.”
“RAPTOR uses an attention mechanism that prioritizes nanoparticle correlations between pre- and post-tamper samples before passing them to a deep convolutional classifier based on residual attention. It takes nanoparticles in descending order of radius and builds a distance matrix and radius from pre- and post-tamper samples,” said Blake Wilson, a graduate student from Kirdyshev’s group.
Validation of Purdue University’s RAPTOR
The Purdue team tested RAPTOR’s counterfeit detection capabilities by simulating tampering actions on the nanoparticle system, including natural variations, malicious adversary tampering, thermal fluctuations, and varying degrees of random Gaussian transformation of the nanoparticles.
“RAPTOR demonstrated the highest average accuracy, correctly detecting tampering in 97.6% of the distance matrices under worst-case tampering assumptions,” Wilson said. “This outperforms traditional methods such as Hausdorff, Procrustes, and average Hausdorff distance by 40.6%, 37.3%, and 6.4%, respectively.”
Kirdyshev said the team plans to collaborate with chip packaging researchers to further innovate the nanoparticle embedding process and streamline the certification procedures.
“Right now, RAPTOR is a proof of concept that shows the huge potential of AI in the semiconductor industry,” he said. “Ultimately, we hope to translate this into a mature industry solution.”
Other members of the RAPTOR team include Alexandra Boltasseva, the Ron and Dotty Garvin Tonjes Distinguished Professor of Electrical and Computer Engineering, Vladimir Shalaev, the Bob and Anne Burnett Distinguished Professor of Electrical and Computer Engineering, and current and past students Daksh Kumar Singh, Rohan Ojha, Jaxon Pottle and Michael Bezick.
The team is supported by the U.S. Department of Energy’s Quantum Science Center, the National Science Foundation and the Elmore ECE Center for Emerging Frontiers at the Intersection of Quantum and AI.
About Purdue Innovates Office of Technology Commercialization
The Purdue Innovates Office of Technology Commercialization operates the most comprehensive technology transfer program of any major research university in the United States. Services provided by the office support Purdue’s economic development initiatives and benefit the University’s academic activities through the commercialization, licensing and protection of Purdue’s intellectual property. In fiscal year 2024, the office reported signing 145 agreements, signing 224 technologies, receiving 466 invention disclosures and granting 290 U.S. and international patents. The office is managed by the Purdue Research Foundation, a private, nonprofit foundation established to advance the mission of Purdue University. For more information, please contact otcip@prf.org.
About Purdue University
Purdue University is a public research institution that demonstrates excellence at scale. Ranked among the top 10 public universities in the U.S. and with two universities ranked in the top four, Purdue discovers and disseminates knowledge with unmatched quality and scale. More than 105,000 students learn at Purdue in a variety of ways and places, with nearly 50,000 learning face-to-face at the West Lafayette campus. Committed to affordability and accessibility, Purdue’s main campus has frozen tuition for 13 consecutive years. With the first integrated urban campus in Indianapolis, the Mitch Daniels School of Business, Purdue Computing, the One Health initiative and more, Purdue never stops for its next big leap at https://www.purdue.edu/president/strategic-initiatives.
