Key information:
| Student | Callum Aitchison |
| Academic Supervisors | |
| Cohort | 2 |
| Pure Link | Active Project |
As digital devices continue to become increasingly pervasive in everyday life the ability to trust that they have not been tampered with will continue to be of great concern. These threats to the security and trustworthiness of sometimes mission-critical hardware are constantly evolving and, as such, strong security must be built in from the ground up.
Currently, when a device is manufactured, it is typically identified with a unique code such as a serial number. However, in many cases it would be trivial for an attacker to copy such a code to masquerade as the device, meaning such an identifier cannot be trusted. This copy could be tampered with such that it causes errors in the attacker’s favour. For example, a machine-learning based safety system may be tampered to give erroneous results, disabling security in specific scenarios.
My research aims to help protect against these types of attacks, providing a security system that enables devices’ identity and trustworthiness to be ensured all the way from fabrication, through the supply chain, to final implementation. This will be achieved using hardware based on emerging memristor-based memory devices. These devices encode data using an adjustable resistance state in an array of memristors, rather than simple binary as in common CMOS-based memories today. By amplifying minute, uncontrollable, differences between these memristors, introduced when they are first fabricated or used, a unique fingerprint may be generated to uniquely identify the device they comprise. This is known as a “physical unclonable function”
Some attacks on CMOS-based unclonable functions have been demonstrated using machine learning based approaches. In such attacks an attacker is able to extract enough of a response from testing a device for a short period to enable them to model the entire device and generate any future keys with high accuracy. This must be hardened against for a truly secure system. Conversely, machine leaning may be applied to help extract the features of the devices which offer the highest degree of unpredictability.