Convolutional neural network (CNN) is the state-of-the-art deep learning approach employed in various applications. Real-time CNN implementations in resource limited embedded systems are becoming highly desired recently. To ensure the programmable flexibility and shorten the development period, field programmable gate array is appropriate to implement the CNN models. However, the limited bandwidth and on-chip memory storage are the bottlenecks of the CNN acceleration. In this paper, we propose efficient hardware architectures to accelerate deep CNN models. The theoretical derivation of parallel fast finite impulse response algorithm (FFA) is introduced. Based on FFAs, the corresponding fast convolution units (FCUs) are developed for the computation of convolutions in the CNN models. Novel data storage and reuse schemes are proposed, where all intermediate pixels are stored on-chip and the bandwidth requirement is reduced. We choose one of the largest and most accurate networks, VGG16, and implement it on Xilinx Zynq ZC706 and Virtex VC707 boards, respectively. We achieve the top-5 accuracy of 86.25% using an equal distance non-uniform quantization method. It is estimated that the average performances are 316.23 GOP/s under 172-MHz working frequency on Xilinx ZC706 and 1250.21 GOP/s under 170-MHz working frequency on VC707, respectively. In brief, the proposed design outperforms the existing works significantly, in particular, surpassing related designs by more than two times in terms of resource efficiency.

Binary weight convolutional neural networks (BCNNs) can achieve near state-of-the-art classification accuracy and have far less computation complexity compared with traditional CNNs using high-precision weights. Due to their binary weights, BCNNs are well suited for vision-based Internet-of-Things systems being sensitive to power consumption. BCNNs make it possible to achieve very high throughput with moderate power dissipation. In this paper, an energy-efficient architecture for BCNNs is proposed. It fully exploits the binary weights and other hardware-friendly characteristics of BCNNs. A judicious processing schedule is proposed so that off-chip I/O access is minimized and activations are maximally reused. To significantly reduce the critical path delay, we introduce optimized compressor trees and approximate binary multipliers with two novel compensation schemes. The latter is able to save significant hardware resource, and almost no computation accuracy is compromised. Taking advantage of error resiliency of BCNNs, an innovative approximate adder is developed, which significantly reduces the silicon area and data path delay. Thorough error analysis and extensive experimental results on several data sets show that the approximate adders in the data path cause negligible accuracy loss. Moreover, algorithmic transformations for certain layers of BCNNs and a memory-efficient quantization scheme are incorporated to further reduce the energy cost and on-chip storage requirement. Finally, the proposed BCNN hardware architecture is implemented with the SMIC 130-nm technology. The postlayout results demonstrate that our design can achieve an energy efficiency over 2.0TOp/s/W when scaled to 65 nm, which is more than two times better than the prior art.

The training of Deep Neural Networks (DNNs) brings enormous memory requirements and computational complexity, which makes it a challenge to train DNN models on resource-constrained devices. Training DNNs with reduced-precision data representation is crucial to mitigate this problem. In this work, we conduct a thorough investigation on training DNNs with low-bit posit numbers, a Type-III universal number (Unum). Through a comprehensive analysis of quantization with various data formats, it is demonstrated that the posit format shows great potential to be employed in the training of DNNs. Moreover, a DNN training framework using 8-bit posit is proposed with a novel tensor-wise scaling scheme. The experiments show the same performance as the state-of-the-art (SOTA) across multiple datasets (MNIST, CIFAR-10, ImageNet, and Penn Treebank) and model architectures (LeNet-5, AlexNet, ResNet, MobileNet-V2, and LSTM). We further design an energy-efficient hardware prototype for our framework. Compared to the standard floating-point counterpart, our design achieves a reduction of 68%, 51%, and 75% in terms of area, power, and memory capacity, respectively.

Quantization is a promising technique to compress the size of Convolutional Neural Network (CNN) models. Recently, various precision-scalable designs have been presented to reduce the computational complexity in CNNs. However, most of them adopt straightforward calculation scheme to implement the CNN, which causes high bandwidth requirement and low hardware utilization efficiency. This paper proposes a new precision-scalable architecture which can fully reduce the computational complexity in CNN inference and meanwhile has a finely simplified calculation scheme. Based on the proposed scheme, a well-optimized multiplier called Compositional Processing Element (C-PE) is devised. Compared with the previous multipliers, the new C-PE requires less area and power. Furthermore, two levels of optimization are introduced to the design to relieve the bandwidth problem and increase the hardware utilization efficiency. Implemented under the TSMC 90nm CMOS technology, the whole design achieves 6-68.1 fps in various precisions on VGG16 benchmark and a 49.8TOPS/W energy efficiency at 500MHz when scaled to 28nm, which is much better than previous precision-scalable ones.

While long polar codes can achieve the capacity of arbitrary binary-input discrete memoryless channels when decoded by a low complexity successive-cancellation (SC) algorithm, the error performance of the SC algorithm is inferior for polar codes with finite block lengths. The cyclic redundancy check (CRC)-aided SC list (SCL) decoding algorithm has better error performance than the SC algorithm. However, current CRC-aided SCL decoders still suffer from long decoding latency and limited throughput. In this paper, a reduced latency list decoding (RLLD) algorithm for polar codes is proposed. Our RLLD algorithm performs the list decoding on a binary tree, whose leaves correspond to the bits of a polar code. In existing SCL decoding algorithms, all the nodes in the tree are traversed, and all possibilities of the information bits are considered. Instead, our RLLD algorithm visits much fewer nodes in the tree and considers fewer possibilities of the information bits. When configured properly, our RLLD algorithm significantly reduces the decoding latency and, hence, improves throughput, while introducing little performance degradation. Based on our RLLD algorithm, we also propose a high throughput list decoder architecture, which is suitable for larger block lengths due to its scalable partial sum computation unit. Our decoder architecture has been implemented for different block lengths and list sizes using the TSMC 90-nm CMOS technology. The implementation results demonstrate that our decoders achieve significant latency reduction and area efficiency improvement compared with the other list polar decoders in the literature.

Low-complexity and high-performance low-density parity-check (LDPC) decoders are highly demanded in various modern communication and storage systems. In this paper, a novel hard-decision decoding algorithm, called tabu-list random-penalty gradient descent bit-flipping (TRGDBF) algorithm, is proposed. Compared to the state-of-the-art hard-decision algorithms, the TRGDBF algorithm has much better error-correction performance due to several algorithmic improvements. First, a random-penalty term is introduced to the inversion function to help the decoder escape from trapping sets, which are the main causes of the error-floor phenomenon. Second, a tabu-list is employed to improve the decoding efficiency. Numerical results show that the TRGDBF algorithm can achieve up to two orders of magnitude better error-correction performance than the probabilistic gradient descent bit-flipping (PGDBF) algorithm and reduce the average iteration count by about 15%. In addition, a well-optimized hardware architecture is developed to implement the TRGDBF algorithm. Algorithmic transformation and architecture optimization are well explored to reduce the hardware complexity and latency. Synthesis results show that the TRGDBF decoder can work at a higher frequency and offer a larger throughput than the PGDBF decoder.