Backdoor attacks compromise deep neural networks by injecting them with covert, malicious behaviors during training. A backdoor can then be activated at test-time using a trigger pattern. As backdoors become more sophisticated, defenses struggle to catch up. This paper introduces a simple yet effective input-specific Backdoor Attack using Inpainted Triggers, dubbed BAIT. Its trigger relies on a randomly-drawn polygonal patch, filled via inpainting with an off-the-shelf generative adversarial network. We show with BAIT that several defenses, including common test-time input purification methods, can be bypassed by patch-based backdoors. To counter this new attack, we propose four defense recommendations.

In this work, we study the micro architectural vulnerabilities of the open-source RISC-V CPU named CVA6. We build a realistic scenario for extracting information and propose an analysis on how to reduce the impact of noise on the attack, while staying as close as possible to hardware level through baremetal simulations.

Security in true random number generation in cryptography is based on entropy per bit at the generator output. The entropy is evaluated using stochastic models. Several recent works propose stochastic models based on assumptions related to selected physical analog phenomena such as noise or jittery signal and on the knowledge of the principle of randomness extraction from the obtained analog signal. However, these assumptions simplify often considerably the underlying analog processes, which include several noise sources. In this paper, we present a new comprehensive multilevel approach, which enables to build the stochastic model based on detailed analysis of noise sources starting at transistor level and on conversion of the noise to the clock jitter exploited at the generator level. Using this approach, we can estimate proportion of the jitter coming only from the thermal noise, which is included in the total clock jitter.

Side-channel attacks pose a critical threat to the deployment of secure embedded systems. Differential-power analysis is a technique relying on measuring the power consumption of device while it computes a cryptographic primitive, and extracting the secret information from it exploiting the knowledge of the operations involving the key. There is no open literature describing how to properly employ Digital Signal Processing (DSP) techniques in order to improve the effectiveness of the attacks. This paper presents a pre-processing technique based on DSP, reducing the number of traces needed to perform an attack by an order of magnitude with respect to the results obtained with raw datasets, and puts it into practical use attacking a commercial 32-bit software implementation of AES running on a Cortex-M3 CPU. The main contribution of this paper is proposing a leakage model for software implemented cryptographic primitives and an effective framework to extract it.