Attacking Hardware Random Number Generators in a Multi-Tenant Scenario

Abstract

True random number generators are important building blocks for cryptographic systems and can be the target of adversaries that want to break cryptographic protocols by reducing the unpredictability of the used random numbers. This paper examines the viability of three different types of potential attacks on these generators when they are implemented on field programmable gate arrays, namely the voltage manipulation attack, the ring-oscillator locking attack and the replica observation attack. The proposed attacks only make use of the available programmable logic of the device and as such do not require physical access to it. They can technically be mounted remotely in a multi-tenant scenario by adversaries that only have bitstream write access to a part of the programmable logic. The attacks try to exploit interactions that can exist between an attack circuit and the targeted circuit because they reside on the same chip. The paper presents two case studies: an elementary ring oscillator design and a transition effect ring oscillator design. For the first case study, all three scenarios were tested and for the second case study, only the voltage manipulation attack scenario is examined. Our results show that this voltage manipulation attack is the most effective of the three proposed attacks.

Cryptographically Secure PseudoRandom Bit Generator for Wearable Technology

“…This would constitute a forbidden situation, which NIST explicitly warns against in its recommendations [9]. This problem is well known and increasingly reported in research on secure initialization of random generators for cryptographic systems [10][11][12][13][14][15][16]. Having a well-designed PRBG is not sufficient, because in the case of the described entropy problems, random generators implemented serially on several different devices can generate almost identical random sequences.…”

“…Using the same value repeatedly in the process of initializing pseudo-random algorithms raises risks for attacks that affect DRBG (Deterministic Random Bit Generator) initialization [9]. Despite the knowledge of these standards, it is still challenging to provide multiple efficient and attack-resistant entropy sources in the design process of random generators [10][11][12][13][14][15][16][17][18].…”

TROT: A Three-Edge Ring Oscillator Based True Random Number Generator With Time-to-Digital Conversion

“…Since the minimal estimate is above the model-predicted 0.770, we can conclude that this type of attack on the TROT seems insufficiently effective. We performed the frequency injection attacks equivalent to those proposed in [45] and [46], which were proven successful when applied on the single-edge RO-based TRNGs. For our attacks, we inject oscillating signals in the delay line next to the three-edge RO to induce strong interaction between the RO signal and the injected signal.…”