EAGER: Detecting and Avoiding Side-Channel Attacks with Security Conscious Prediction

多年来，计算机一直在使用一种名为“投机”的技术来获得良好的性能。计算机还实施安全策略，防止私人数据泄露给未经授权的实体。最近，研究人员了解到，投机可能会在无意中让私人信息被泄露。攻击者可以操纵推测，通过“旁路”传递数据，从而击败安全策略。这个项目将探索如何使用机器学习来检测计算机是否受到旁路攻击，并触发防御措施，以防止私人数据被泄露。这项工作将实现安全计算，同时保持投机的好处。预测器将接受培训，以检测系统是否受到攻击，从而提供对预测的置信度。预测器的输入将是微体系结构事件计数等特征。预测器将被离线训练，并在硬件中实现，以便在执行期间使用。使用来自真实系统和模拟系统的测量，将探索与恶意行为相关的功能。基于神经学习的预测器将利用这些特征进行训练。预测器将在微体系结构和电路模拟器中进行原型和评估。基于预测器置信度的缓解也将是原型。侧通道攻击威胁到继续使用投机来提供高性能。预计这项工作将能够继续使用对私人用户数据安全性高度自信的推测，同时继续保持当今移动、服务器和嵌入式应用程序所要求的迫切需要的性能水平。来自代表性不足群体的学生将被鼓励参与这项研究。这项研究将出现在德克萨斯农工大学的课堂教学中。项目代码和数据将在项目完成后至少两年内提供。该项目的产品，包括技术论文、代码档案和数据集，将在http://taco.cse.tamu.edu/secure/.This颁奖典礼上提供，这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智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

For many years, computers have been using a technique called "speculation" to achieve good performance. Computers also implement security policies that prevent private data from being revealed to unauthorized entities. Recently, researchers have learned that speculation can unintentionally allow private information to be leaked. An attacker can manipulate speculation to communicate data through a "side-channel," defeating security policies. This project will explore ways to use machine learning to detect whether a computer is under a side-channel attack and trigger defenses that will keep private data from being leaked. The work will enable secure computing while maintaining the benefits of speculation.A predictor will be trained to detect whether the system is under attack, providing a level of confidence in the prediction. Input to the predictor will be features such as counts of microarchitectural events. The predictor will be trained offline and implemented in hardware to be used during execution. Using measurements from real and simulated systems, features correlated with malicious behavior will be explored. Predictors based on neural learning will be trained with those features. The predictor will be prototyped and evaluated in a microarchitectural and circuit simulator. Mitigations based on the predictor confidence will also be prototyped.Side-channel attacks threaten the continued use of speculation to provide high performance. It is expected that this work will enable the continued use of speculation with high confidence in the security of private user data while continuing the much needed level of performance demanded by today's mobile, server, and embedded applications. Students from under-represented groups will be encouraged to participate in the research. The research will be featured in classroom teaching at Texas A&M University.The project code and data will be made available for at least two years following the completion of the project. The products of this project including technical papers, code archive, and datasets will be made available at http://taco.cse.tamu.edu/secure/.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.