Scholarlink

<p>Cloud services and the internet of things, which operate in dynamic multi-user and multi-device environments, require scalable, real-time authentication. However, conventional static key management systems lack flexibility and security. Here we report on a security system based on physical unclonable functions that uses chaotic vertical-cavity surface-emitting lasers as entropy sources for key generation. The system offers response rates above 500 Gbps with an energy consumption below 1 pJ per bit per laser emitter. To match its high-throughput performance, we developed a convolutional neural network model for real-time authentication that performs dynamic key matching with near-zero false positive rates. To enhance security and reduce communication overhead, an adversarial generative framework is incorporated to protect key transmission and improve resistance to model inversion attacks. We also designed a three-dimensional co-packaged hardware structure for the physical unclonable functions that supports compact integration and flexible deployment with an estimated energy consumption of 2.04 pJ per bit.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Physical unclonable functions based on chaotic vertical-cavity surface-emitting lasers for dynamic authentication

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Zhican Zhou,
  • Hang Lu,
  • Nakul Nandhakumar,
  • Omar Alkhazragi,
  • Xiangpeng Ou,
  • Heming Lin,
  • Tien Khee Ng,
  • Boon S. Ooi,
  • Yating Wan

摘要

Cloud services and the internet of things, which operate in dynamic multi-user and multi-device environments, require scalable, real-time authentication. However, conventional static key management systems lack flexibility and security. Here we report on a security system based on physical unclonable functions that uses chaotic vertical-cavity surface-emitting lasers as entropy sources for key generation. The system offers response rates above 500 Gbps with an energy consumption below 1 pJ per bit per laser emitter. To match its high-throughput performance, we developed a convolutional neural network model for real-time authentication that performs dynamic key matching with near-zero false positive rates. To enhance security and reduce communication overhead, an adversarial generative framework is incorporated to protect key transmission and improve resistance to model inversion attacks. We also designed a three-dimensional co-packaged hardware structure for the physical unclonable functions that supports compact integration and flexible deployment with an estimated energy consumption of 2.04 pJ per bit.