Number Theoretic Transform (NTT) and Inverse NTT (INTT) are the key algorithms used to speed up polynomial multiplications in post-quantum cryptography (PQC). This brief proposes a cross-bank-write-back (CBWB) memory mapping scheme for memory-based NTT/INTT architecture with conflict-free access patterns. The proposed scheme enables a light-weight memory access patterns for high-radix butterflies. Moreover, it makes the datapath between the memory banks and the butterfly unit straightforward and hardware-friendly. Based on the CBWB scheme, this work also proposes a solution to adopt a single decimation strategy for both NTT and INTT in PQC accelerators to reduce hardware resources. As a case study, an efficient NTT/INTT architecture has been implemented on the FPGA using the parameters of the PQC algorithm Falcon. Experimental results show that our design achieves better Area-Time Product (ATP) compared to the state-of-the-art designs.
As a classical artificial intelligence algorithm, the convolutional neural network (CNN) algorithm plays an important role in image recognition and classification and is gradually being applied in the Internet of Things (IoT) system. A compact CNN accelerator for the IoT endpoint System-on-Chip (SoC) is proposed in this paper to meet the needs of CNN computations. Based on analysis of the CNN structure, basic functional modules of CNN such as convolution circuit and pooling circuit with a low data bandwidth and a smaller area are designed, and an accelerator is constructed in the form of four acceleration chains. After the acceleration unit design is completed, the Cortex-M3 is used to construct a verification SoC and the designed verification platform is implemented on the FPGA to evaluate the resource consumption and performance analysis of the CNN accelerator. The CNN accelerator achieved a throughput of 6.54 GOPS (giga operations per second) by consuming 4901 LUTs without using any hardware multipliers. The comparison shows that the compact accelerator proposed in this paper makes the CNN computational power of the SoC based on the Cortex-M3 kernel two times higher than the quad-core Cortex-A7 SoC and 67% of the computational power of eight-core Cortex-A53 SoC.
11 April 2018 in IET Computers & Digital Techniques
Fast Fourier transform (FFT) plays an important role in digital signal processing systems. In this study, the authors explore the very large‐scale integration (VLSI) design of high‐precision fixed‐point reconfigurable FFT processor. To achieve high accuracy under the limited wordlength, this study analyses the quantisation noise in FFT computation and proposes the mixed use of multiple scaling approaches to compensate the noise. In addition, a statistics‐based optimisation scheme is proposed to configure the scaling operations of the cascaded arithmetic blocks at each stage for yielding the most optimised accuracy for a given FFT length. On the basis of this approach, they further present a VLSI implementation of area‐efficient and high‐precision FFT processor, which can perform power‐of‐two FFT from 32 to 8192 points. By using the SMIC process, the area of the proposed FFT processor is with a maximum operating frequency of 400 MHz. When the FFT processor is configured to perform 8192‐point FFT at 40 MHz, the signal‐to‐quantisation‐noise ratio is up to 53.28 dB and the power consumption measured by post‐layout simulation is 35.7 mW.
15 January 2025 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
This brief proposes a scalable multistage and multipath architecture for variable number-theoretic transform (NTT). The proposed architecture adopts multiple parallel paths, each of which uses cascaded radix-2 butterfly units (BFUs). The radix-2 scheme simplifies the control logic and the cascaded BFU structure reduces the amount of RAM banks and the frequency of memory accesses. Moreover, a conflict-free and hardware-friendly in-place memory mapping scheme is proposed to ease the adaption to multiple paths, letting it be scalable for various throughputs. Compared with state-of-the-art works, the proposed architecture uses fewer resources and has better area-time product performance without penalty in throughput.
18 December 2024 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Sparse convolutional neural networks (SCNNs) which can prune trivial parameters in the network while maintaining the model accuracy has been proved to be an attractive approach to alleviate the heavy computation of convolutional neural networks (CNNs). However, the invalid data resulting from sparse patterns leads to unnecessary and irregular computation workload, which challenges the efficiency of the underlying hardware accelerators. Therefore, this article proposes an SCNN inference accelerator, which can deal with the imbalanced workload both intra- and interprocessing element (PE). A valid weight encoding (VWE) scheme is proposed to compress sparse weights into dense ones to alleviate the load imbalance intra-PE. Leveraging the VWE scheme, a randomized load rearrangement (RLR) method is proposed to dynamically schedule convolution kernels with similar sparsity into the same computation batch to alleviate the load imbalance inter-PEs. In addition, to reduce off-chip memory accesses, a recurrent weight stationary (RWS) dataflow is proposed, which adopts a small-batch and multichannel strategy to stack data from multiple channels within one off-chip access and let them compute simultaneously thereby enabling efficient reuse of on-chip data. Based on the proposed scheme, an efficient SCNN inference accelerator has been designed and verified on the field-programmable gate array (FPGA). Compared with state-of-the-art works, our design achieves $1.16\times $ to $2.77\times $ higher digital signal processors (DSPs) efficiency and $1.75\times $ to $15\times $ higher logic efficiency.
1 October 2020 in IEICE Transactions on Information and Systems
Optical navigation (OPNAV) is the use of the on-board imaging data to provide a direct measurement of the image coordinates of the target as navigation information. Among the optical observables in deep-space, the edge of the celestial body is an important feature that can be utilized for locating the planet centroid. However, traditional edge detection algorithms like Canny algorithm cannot be applied directly for OPNAV due to the noise edges caused by surface markings. Moreover, due to the constrained computation and energy capacity on-board, light-weight image-processing algorithms with less computational complexity are desirable for real-time processing. Thus, to fast and accurately extract the edge of the celestial body from high-resolution satellite imageries, this paper presents an algorithm-hardware co-design of real-time edge detection for OPNAV. First, a light-weight edge detection algorithm is proposed to efficiently detect the edge of the celestial body while suppressing the noise edges caused by surface markings. Then, we further present an FPGA implementation of the proposed algorithm with an optimized real-time performance and resource efficiency. Experimental results show that, compared with the traditional edge detection algorithms, our proposed one enables more accurate celestial body edge detection, while simplifying the hardware implementation.