Collaborative Research: EAGER: IC-Cloak: Integrated Circuit Cloaking against Reverse Engineering

集成电路（ic）的设计和制造外包减少了大量电子设计公司的运营和维护成本，缩短了上市时间。然而，这种外包和利益导致复杂的验证制造集成电路。此外，逆向工程设计方面的安全威胁已经浮出水面。为了解决这种IC的逆向工程，本项目建议在IC布局中插入噪声，以最大限度地减少IC逆向工程成功的概率。噪声的插入也必须符合标准的IC设计流程要求，并通过验证和验证。为了实现这一目标，该项目将对抗性机器学习与IC设计流程集成在一起，以便在攻击者获得给定IC的扫描电子显微镜(SEM)/布局图像的情况下实现有效的IC设计保护。在第一阶段，该项目开发了一个代理机器学习模型来检测SEM图像中的门，然后在SEM图像的空间约束下孵化新的对抗性扰动。通过对抗性学习获得的扰动将被评估并嵌入到一些可用的开源标准单元库中，以保证与现有IC设计流程工具的兼容性。该项目的成功完成将产生一套与IC设计流程兼容的安全IC单元库。具体而言，该项目(i)开发了新的IC单元布局，尽管通过逆向工程获得了图像，但仍无法被对手识别；（ii）在空间约束下引入新的对抗性摄动生成。由于跨学科的性质，该项目的结果影响了IC设计和对抗性机器学习领域的广泛研究人员。图书馆单元的开发将在获得相关许可的情况下外包给学术界和工业界。该项目预计将生成多种类型的数据，包括IC布局、SEM图像和细胞模型。所有代码将使用Python和SystemC/Verilog（如果需要）编写。完整的功能和测试代码将被记录，并将通过GitHub提供。pi的网站（http://mymason.gmu.edu/~spudukot）将会新增一个下载源代码的页面。数据将在奖学金结束后至少三年保留在GMU、加州大学戴维斯分校和加州大学欧文分校。学位授予后，与学生研究工作有关的数据将保留四年。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Outsourcing the design and manufacturing of integrated circuits (ICs) has minimized the operating and maintenance costs for a plethora of electronic design companies with reduced time-to-market. However, such outsourcing and benefits lead to complex verification of the fabricated ICs. Additionally, the security threats in terms of reverse engineering the design have surfaced. To address such reverse engineering of ICs, this project proposes insertion of noise in the IC layout that minimizes the probability of success in IC reverse engineering. The insertion of noise must also meet the standard IC design flow requirements and pass the verification and validation. To achieve this goal, this project integrates the adversarial machine learning with the IC design flow to enable efficient IC design protection despite attacker obtaining the scanning electron microscope (SEM)/layout images of a given IC. In the first phase, the project develops a surrogate machine learning model to detect the gates in SEM images, followed by incubation of novel adversarial perturbations under spatial constraints on the SEM images. The perturbations obtained through the adversarial learning will be evaluated and embedded in some of the available open-source standard cell libraries, guaranteeing compatibility with existing IC design flow tools. Successful completion of the project will result in a suite of secure IC cell libraries that are compatible with IC design flows. Specifically, this project (i) develops novel IC cell layouts that cannot be identified by an adversary despite obtaining the images through reverse engineering; (ii) introduces novel adversarial perturbation generation under spatial constraints. Due to the interdisciplinary nature, the outcome of the project impacts a broad variety of researchers in the domains of IC design and adversarial machine learning. The development of library cells from this project will be outsourced with relevant licenses to academia and industry. The project is expected to generated multiple types of data including IC layouts, SEM images, and cell models. All the codes will be written using Python, and SystemC/Verilog (If required). Fully functional and tested codes will be documented and will be made available through GitHub. The website of PIs (http://mymason.gmu.edu/~spudukot) will add a new page for downloading the source codes. Data will be retained at GMU, UC Davis, and UC Irvine for a minimum of three years after conclusion of the award. Data related to students’ research work will be retained for four years after the degree is awarded.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.