International Association
for Cryptologic Research

<p>In recent years, there has been a growing interest in low-latency ciphers. Since the first low-latency block cipher PRINCE was proposed at ASIACRYPT 2012, many low-latency primitives sprung up, such as Midori, MANTIS, QARMA and SPEEDY. Some ciphers, like SPEEDY and Orthros, introduce bit permutations to achieve reduced delay. However, this approach poses a challenge in evaluating the resistance against some cryptanalysis, especially differential and linear attacks. SPEEDY-7-192, was fully broken by Boura et.al. using differential attack, for example. In this paper, we manage to propose a novel low-latency block cipher, which guarantees security against differential and linear attacks. Revisiting the permutation technique used in Orthros, we investigate the selection of nibble permutations and propose a method for selecting them systematically rather than relying on random search. Our new nibble permutation method ensures the existence of impossible differential and differential trails for up to 8 rounds, while the nibble permutations for both branches of Orthros may lead to a 9-round impossible differential trail. Furthermore, we introduce a new approach for constructing low-latency coordinate functions for 4-bit S-boxes, which involves a more precise delay computation compared to traditional methods based solely on circuit depth. The new low-latency primitive uLBC we propose, is a family of 128-bit block ciphers, with three different versions of key length, respectively 128-bit and 256-bit key, as well as a 384-bit tweakey version with variable-length key. According to the key length, named uLBC-128, uLBC-256 and uLBC-384t. Our analysis shows that uLBC-128 exhibits lower latency and area requirements compared to ciphers such as QARMA9-128 and Midori128. On performance, uLBC-128 has excellent AT performance, the best performance except SPEEDY-6, and even the best performance in UMC 55nm in our experiments. </p>

The machine learning problem of extracting neural network parameters has been proposed for nearly three decades. Functionally equivalent extraction is a crucial goal for research on this problem. When the adversary has access to the raw output of neural networks, various attacks, including those presented at CRYPTO 2020 and EUROCRYPT 2024, have successfully achieved this goal. However, this goal is not achieved when neural networks operate under a hard-label setting where the raw output is inaccessible. In this paper, we propose the first attack that theoretically achieves functionally equivalent extraction under the hard-label setting, which applies to ReLU neural networks. The effectiveness of our attack is validated through practical experiments on a wide range of ReLU neural networks, including neural networks trained on two real benchmarking datasets (MNIST, CIFAR10) widely used in computer vision. For a neural network consisting of $10^5$ parameters, our attack only requires several hours on a single core.