In this paper, a new true random number generator (TRNG), based entirely on digital components is proposed. The design has been implemented using a fast random number generation method, which is dependent on a new type of ring oscillator with the ability to be set in metastable mode. Earlier methods of random number generation involved employment of jitter, whereas the proposed method leverages the metastability phenomenon in digital circuits and applies it to a ring oscillator. The new entropy employment method allows an increase in the TRNG throughput by significantly reducing the required entropy accumulating time. Samples obtained from simulation of TRNG design have been evaluated using AIS.31 and FIPS 140-1/2 statistical tests. The results of these tests have proven the high quality of generated data. Corners analysis of the TRNG design was also performed to estimate the robustness to technology process and environment variations. Investigated in FPGA technology, phase distribution highlighted the advantages of the proposed method over traditional architectures.
Abstract. The elliptic curve cryptosystem(ECC) is increasingly being used in practice due to its shorter key sizes and efficient realizations. However, ECC is also known to be vulnerable to various side channel attacks, including power attacks and fault injection attacks. This paper proposes new countermeasures for ECC scalar multiplications against differential power attacks and fault attacks. The basic idea of proposed countermeasures lies in extending the definition field of an elliptic curve to its random extension ring and performing the required elliptic curve operations over the ring. Moreover, new methods perform a point validation check in a small subring of the extension ring to give an efficient fault attack countermeasure.
There has been a rapid advance of custom hardware (HW) for accelerating the inference speed of deep neural networks (DNNs). Previously, the softmax layer was not a main concern of DNN accelerating HW, because its portion is relatively small in multi-layer perceptron or convolutional neural networks. However, as the attention mechanisms are widely used in various modern DNNs, a cost-efficient implementation of softmax layer is becoming very important. In this paper, we propose two methods to approximate softmax computation, which are based on the usage of LookUp Tables (LUTs). The required size of LUT is quite small (about 700 Bytes) because ranges of numerators and denominators of softmax are stable if normalization is applied to the input. We have validated the proposed technique over different AI tasks (object detection, machine translation, sentiment analysis, and semantic equivalence) and DNN models (DETR, Transformer, BERT) by a variety of benchmarks (COCO17, WMT14, WMT17, GLUE). We showed that 8-bit approximation allows to obtain acceptable accuracy loss below 1.0%.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.