In the past couple of decades, many in-vehicle features have been invented and deployed in order to make modern vehicles which not only safer and more reliable but also connected, smarter, and intelligent. Meanwhile, vehicular ad-hoc networks (VANETs) are proposed to provide communications between vehicles and road-side stations as the foundation of the intelligent transportation system to provide efficient and safe transportation. To support these updated functions, a large amount of electronic equipment has been integrated into the car system. Although these add-on functions around vehicles offer great help in driving assistance, they inevitably introduced new security vulnerabilities that threaten the safety of the on-board drivers, passengers and pedestrians. This has been demonstrated by many well-documented attacks either on the in-vehicle bus system or on the wireless vehicular network communications. In this dissertation, we design and implement several hardware-oriented solutions to the arousing security issues on vehicles. More specifically, we focus on three important and representative problems: (1) how to secure the in-vehicle Controller Area Network (CAN), (2) how to secure the communication between vehicle and outside, and (3) how to establish trust on VANETs.

Current approaches based on cryptographic algorithms to secure CAN bus violate the strict timing and limited resource constraints for CAN communications. We thus emphasize on the alternate solution of intrusion detection system (IDS) in this dissertation. We explore monitoring the changes of CAN message content or the physical delay of its transmission to detect on the CAN bus. We first propose a new entropy-based IDS following the observation that all the known CAN message injection attacks need to alter the CAN identifier bit. Thus, analyzing the entropy changes of such bits can be an effective way to detect those attacks. Next, we develop a delay-based IDS to protect the CAN network by identifying the location of the compromised Electronic Control Unit (ECU) from the transmission delay difference to two terminals connected to the CAN bus. We demonstrate that both approaches can protect the integrity of the messages on CAN bus leading to a further improve the security and safety of autonomous vehicles.

In the second part of this dissertation, we consider Plug-and-Secure, an industrial practice on key management for automotive CAN networks. It has been proven to be information theoretically secure. However, we discover side-channel attacks based on the physical properties of the CAN bus that can leak almost the entire secret key bits. We analyze the fundamental characteristics that lead to such attacks and propose techniques to minimize information leakage at the hardware level.

Next, we extend our study from in-vehicle secure CAN communication to the communication between vehicle and outside world. We take the example of the popular GPS spoofing attack and show how we can use the rich information from CAN bus to build a cross-validation system to detect such attacks. Our approach is based on the belief that the local driving data from the in-vehicle network can be authenticated and thus trusted by secure CAN networks mechanisms. Such data can be used to cross-validate the GPS signals from the satellite which are vulnerable to spoofing attacks. We conduct driving tests on real roads to show that our proposed approach can defend both GPS spoofing attacks and location-based attacks on the VANETs.

Finally, we propose a blockchain based Anonymous Reputation System (BARS) to establish a privacy-preserving trust model for VANETs. The certificate and revocation transparency is implemented efficiently with the proofs of presence and absence based on the extended blockchain technology. To prevent the broadcast of forged messages, a reputation evaluation algorithm is presented relying on both direct historical interactions of that vehicle and indirect opinions from the other vehicles.

This dissertation features solutions to vehicle security problems based on hardware or physical characteristics, instead of cryptographic algorithms. We believe that given the critical timing requirement on vehicular systems and their very limited resource (such as the bandwidth on CAN bus), this will be a very promising direction to secure vehicles and vehicular network.