Characterizing Laser Signal Injection and its Impact on the Security of Cyber-Physical Systems

Lasers can be used to inject adversarial-controlled signals into sensors used in cyber-physical systems. This capability is often unexpected, use physical mechanisms that were never considered, and exploits the blind trust in sensors. These laser signal injection attacks can allow adversarial influence or even control over a system's perception of the environment, leading to potentially harmful situations. The contributions of this work are the characterizations of laser signal injection in three cyber-physical contexts: LiDAR sensors used in autonomous vehicles, MEMS microphones in voice-controllable systems, and the sensors used in space systems. These characterizations include an in-depth investigation of attacker capabilities, the development of models to describe the vulnerability, and a description of the consequences on the relevant cyber-physical systems. The characterization of laser signal injection into LiDAR sensors used in autonomous vehicles builds upon previous research to define the capabilities of an attacker to influence LiDAR data. This work shows that false laser returns can be spoofed at a large number of angles and formed into specific shapes. These shapes can be used to trick object detection algorithms to register false objects, causing autonomous vehicles to make dangerous control decisions. This is followed by recommendations for future defenses by adjusting the object detection algorithms or making modifications to the LiDAR sensors to prevent laser signal injection. The characterization of laser signal injection into MEMS microphones used in voice-controllable systems is the investigation of a previously unknown vulnerability. This work shows that MEMS microphones are unexpectedly vulnerable to laser signal injection attacks due to a combination of photoelectric and photoacoustic phenomena. Models of these effects are presented, as well as a comparison of the relative contributions of these effects on the output of the microphone given different laser injection characteristics. Beyond this, there is also an investigation of the vulnerabilities of voice-controllable systems that rely on microphone data to perform autonomous actions, leading to new threat models. Defense recommendations for protecting the microphones at the system level or developing new light-resistant microphones are presented. The characterization of laser signal injection into sensors used in space systems is a preliminary investigation of future threat models. This work includes a preliminary attack surface analysis of sensors that are potentially vulnerable to laser signal injection. It goes over the current capabilities of a potential adversary as well as capabilities on which to focus future research to determine future vulnerabilities. An example case study on light-sensitive sun sensors is presented to show an example attack on space sensors. Beyond this, recommendations are made to set up test benches for future vulnerability research on satellites, as well as recommendations for future defenses to detect laser single injection. These characterizations are performed not just to present the attacks, but to fully understand the mechanisms that lead to vulnerabilities within these cyber-physical systems. This understanding is necessary to design future systems in a way that will be resistant to all forms of laser signal injection, allowing sensors and cyber-physical systems to be safer, more trustworthy, and more secure.PHDComputer Science & EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/177684/1/bencyr_1.pd