EAGER: SARE: Real-Time Learning and Countering of Side-Channel Emissions to Enable Secure RF and Analog Microelectronics

传感器和物联网(IoT)的快速增长具有改变社会、经济和提高生活质量的潜力。位于极端云边缘的许多物联网设备以无线方式收集和传输敏感信息，以进行远程计算。然而，敏感信息可能会通过旁路泄露，包括功耗和电磁(EM)排放。这些无线设备容易受到黑客攻击或利用的漏洞，已成为公众对物联网安全和安全的主要担忧。然而，现有的最先进的网络安全方法主要集中在软件和数字模块上。安全措施尚未集成在射频(RF)和模拟域中，以验证信号和电源或抑制旁通道发射。为了弥补这一差距，该项目将开发一种全面的自我测试方法，其中包括纳米级电磁传感设备、可重新配置的射频电路和机器学习算法，以直接在射频和模拟前端检测威胁和对抗恶意攻击。结合新兴的材料、器件、电路和系统概念，该项目旨在开发射频/模拟领域的内置威胁检测和反应方法，在不降低性能的同时实现良好的能效。这项工作的结果有可能对安全电子和电信行业产生重大影响。能源和空间的最小使用可以使能源受限的无线设备具有片上检测和反应系统，以在第一线快速预测和应对恶意攻击。此外，该项目将为不同级别的学生(K-12、本科生和研究生)和普通公众提供一个独特的机会，了解无线设备的安全漏洞及其对策。该项目的目标是通过跨学科研究提高射频/模拟前端无线设备的安全性。将开发低功率和低电压的片上传感器来收集数据，用于分析功率和电磁信号行为。为了感知磁场的微小变化并向机器学习电路提供信息，将开发一种纳米级异质结构，将其与新型自旋扭矩器件单片集成在一起，这些器件可以用作坚固的高保真电磁传感器并嵌入到互连中。将开发一种传感电路，以实现高精度的电磁测量，同时消除对外部杂散场干扰的任何响应。将研究低功耗电荷共享模数转换器以执行机器学习算法。为了快速预测和检测潜在的攻击，机器学习电路将被训练来学习系统在不同操作方案下的正常行为，从而使这个内置的检测系统能够实时准确地预测各种对手。将开发学习算法和可重构的体系结构，以形成一个闭环的快速威胁检测和反应系统。将利用线性分类器集成的增强方法来以最小的硬件成本来提高检测精度。可编程纳米级振荡器将采用新兴的跳频材料和各种切换方案，以提高对篡改旁通道攻击的抵抗力和弹性。当检测到威胁时，抑制机制将被激活以立即减轻损害。这一裁决反映了NSF的法定使命，并已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。

Rapid growth of sensors and Internet of Things (IoT) has the potential to transform the society, economy, and improve the quality of life. Many IoT devices at the extreme cloud edge collect and transmit sensitive information wirelessly for remote computing. However, the sensitive information can be leaked through side channels, including power consumption and electromagnetic (EM) emissions. The vulnerability of those wireless devices to hacking or exploitation has emerged as a major public concern over IoT security and safety. Nevertheless, existing state-of-the-art cybersecurity approaches are mainly focused on software and digital modules. Security measures have not been integrated in the radio frequency (RF) and analog domains to verify signal and power sources or to suppress the side-channel emissions. To bridge the gap, this project will develop a holistic self-testing approach incorporating nanoscale electromagnetic sensing devices, reconfigurable RF circuits, and machine-learning algorithms to detect threats and counter malicious attacks directly at the RF and analog front-end. Combing emerging material, device, circuit, and system concepts, this project aims to develop a built-in threat-detection-and-reaction approach in the RF/analog domain without degrading the performance while achieving good energy efficiency. The results from this work have the potential to make a significant impact on the secure electronics and telecommunication industry. The minimal usage of energy and space can allow the energy-constrained wireless devices to have an on-chip detection-and-reaction system to rapidly predict and counter malicious attacks in the front line. In addition, the project will provide a unique opportunity for students of different levels (K-12, undergraduate, and graduate) and the general public to learn the security vulnerability of wireless devices and its countermeasures.The goal of this project is to enhance security in wireless devices at the RF/analog front-end through interdisciplinary research. Low-power and low-voltage on-chip sensors will be developed to collect data for analysis of power and EM signal behaviors. To sense small changes of magnetic fields and inform the machine-learning circuits, a nanoscale heterostructure will be developed to monolithically integrate CMOS circuits with novel spin-torque devices that can be utilized as robust high-fidelity EM sensors and embedded into interconnects. A sensing circuit will be developed to achieve high-accuracy EM measurement while eliminating any response to the external stray field disturbance. Low-power charge-sharing analog-to-digital converters will be investigated to perform machine-learning algorithms. In order to rapidly predict and detect the potential attacks, machine-learning circuits will be trained to learn the normal behavior of the system with different operation schemes so that this built-in detection system can enable accurate predictions of a variety of adversaries in real time. Learning algorithms and reconfigurable architectures will be developed to form a closed-loop fast threat-detection-and-reaction system. Boosting approaches for ensembles of linear classifiers will be exploited to enhance the detection accuracy with minimal hardware costs. Programmable nanoscale oscillators with emerging materials for frequency hopping and various switching schemes will be exploited to increase resistance and resilience to tampering side-channel attacks. When a threat is detected, the suppression mechanism will be activated to mitigate the damage immediately.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.