Search

Leakage power reduction techniques are crucial for energy-efficient circuits. This paper investigates the leakage suppression capability, performance, and reliability of dynamic leakage suppression logic (DLSL) and feedforward leakage self-suppression logic (FLSL) techniques, crossing different technology nodes from TSMC 180 nm bulk CMOS to 7 nm FinFET Plus process. Compared with CMOS benchmarks, experimental results show that DLSL-based benchmarks demonstrate a leakage power reduction for four orders of magnitude in 180 nm and 130 nm technologies, while only two orders of magnitude in other technologies. Moreover, FLSL offers a 4-28× performance improvement over DLSL at a cost of 2× leakage power.

Leakage power consumption has become a critical issue for low power Static Random-Access Memory (SRAM) design in the near-threshold regime. In this paper, an ultra-low leakage fourteen-transistor SRAM bitcell structure with the feedforward self-suppression scheme is presented. To reduce the leakage power significantly as well as maintain the data stability in hold state, a cross-coupled dynamic leakage- suppression inverter-based structure is adopted. Furthermore, the bypass scheme is employed to enable the speed modulation for bitcell read and write operations. As compared with state- of-the-art designs, a 65 nm 8kb SRAM array with the proposed bitcell structure achieves 75x leakage power, 45% write power as well as 65% read power reduction at 0.4V. Further comparisons with different processes verify up to 38k times leakage power reduction in 130nm planar process and 139x in 7nm plus FinFET process.

This work presents an operational amplifier (op amp) which output current is 300mA through 0.18 μm CMOS standard process. This circuit is composed of four parts: fully differential folding cascode as input stage, class-AB output stage and three-branch reference current source. Through simulation under the conditions of 5V power supply voltage, 100kΩ resistance and 10pF capacitive load, the gain, phase margin, bandwidth and the peak-to-peak output current value of the operational amplifier are 105dB, 72°, 3.6MHz and 370mA, respectively. This paper implements a high current output op amp with die area of 1.0013 mm2, the actual measurement results show that the circuit can achieve more than 300mA current output. Meanwhile rail-to-rail output voltage swing could be got under the condition of 5V power supply voltage. It can work normally over the temperature range from -40°C~85°C.

Network slicing (NS) is a promising technology of 5G that provides customized end-to-end network service to multi-tenant. How to improve resource utilization efficiency with guarantee of performance isolation in a shared infrastructure is one of the main challenges in resource allocation problem of NS. To address this challenge, we characterize the degree of performance isolation based on the relationship between the requested resource amount, the allocated resource amount, and the time-varying network loads. We propose a dynamic resource allocation problem with probabilistic isolation guarantee (DRAIG), which is formulated as a chance constrained program. As the true probability distribution of network loads is usually unknown, we use the Conditional Valuate-at-Risk (CVaR) measure to provide a distributionally robust formulation that approximate the basic chance constraints of DRA-IG in a data-driven manner. We estimate the second-order moment of network loads by the periodic history information. Then, we further reformulate the distributionally robust optimization problem as a tractable semidefinite programming (SDP). Finally, numerical evaluation verifies the effectiveness of the proposed method.

The goal of Dialogue State Tracking (DST) is to estimate the current dialogue state given all the preceding conversation. Due to the increased number of state candidates, data sparsity problem is still a major hurdle for multi-domain DST. Existing methods generally choose to predict a value for each possible slot over all domains with quite low efficiency. In this paper, we propose a hierarchical dialogue state tracker which consists of three sequential modules: domain classification, slot detection and value extraction. It predicts domains, slots and values dynamically by given the dialogue history and outputs of the preceding module, which can dramatically improve the model efficiency. Experimental results on MultiWOZ2.1 also show that our approach achieves state-of-the-art joint goal accuracy, and confirm that the hierarchical structure can enhance existing DST models significantly.

We present a novel logic family, titled feedforward leakage self-suppression logic (FLSL), for nanowatt and sub-nanowatt always-on circuits. It addresses the prohibitively long delay of existing ultra-low leakage logic families without requiring sleep or mode control signals to exercise the leakage-suppression mode. Implemented in 180-nm CMOS, the proposed logic family achieves femto-watt per-gate leakage and a fan-out-of-4 (FO4) delay of 10.2 µs, a remarkable speed enhancement of 150X over the prior art in the same process in the same leakage level. We also investigate the impact of technology scaling by designing the FLSL logic family additionally in 65-nm and 28-nm processes. In a 28-nm process, the FLSL can achieve about 70 ns FO4 delay and 10 picowatts per-gate leakage. Finally, we prototype a finite impulse response filter core for physical sensing systems in a 180 nm. The filter can achieve the power consumption of 109 picowatts for sparse, i.e., predominantly constant or slowly changing, input signals at 1 kHz clock frequency.

The 3D coordinates have depth information, which improves the accuracy of moving target area acquisition. Therefore, a moving target tracking method based on 3D coordinates is proposed. Firstly, Zhang's calibration algorithm is used to calibrate binocular vision camera; secondly, SAD algorithm is used to extract features of the target, BM algorithm is used to stereo match feature points of the target to determine the 3D coordinates of the center of the moving target; finally, Kalman filter is used to achieve target tracking. Experiments and analysis results show that the algorithm can improve the accuracy of moving object recognition and tracking in left and right views, and has good robustness.

In recent years, action recognition has been widely explored and attains significant performance improvement. In this paper, we propose a real-time action recognition specified convolutional 3D (AR-C3D) neural network for human-computer interaction. The CNN structure is optimized to decrease the complexity. Furthermore, Winograd algorithm is adopted to accelerate computation. It achieves 89.9% accuracy in the application which refers to the robot classifies the video captured by itself and would either imitate human's action or give verbal feedback. The Artix-7 FPGA implementation result outperforms previous work in terms of resource utilization and no external storage is consumed. One video can be processed in 6.6ms, and the power consumption is only 2.7W.

Volume rendering is widely applied in 3D medical images for classifying and visualizing different organs. However, volume rendering can visualize organs for observation, it generally cannot provide 3D surface models for the organs, which are needed in quantitative analysis and 3D printing of organs. In this paper, we propose an effective and novel surface modelling method that can generate 3D surface models for the organs visualized by volume rendering, based on their incomplete classification information implicitly contained in volume rendering. Firstly, our method uses the organ classification information of volume rendering as the initial segmentation to extract topological structure of the organ; and then, a modified graph-cut algorithm is designed to segment complete region of the organ; finally, to obtain a smooth and accurate surface model, we introduce the geodesic active contour method. Our method demonstrates its effectiveness in rigid bones and elastic organs such as spleen and kidney which are of different shapes and sizes based on CT images.

In the big data age, we have to deal with tremendous amount of information, which is collected from various types of sources. For information retrieval systems, the collection of documents becomes larger and larger. For some query, an information retrieval system needs to retrieve a large number of documents as the result to the query. In reality, very often people mainly care about some top-ranked documents rather than the complete long list of documents. In such a situation, how to develop a retrieval system with desirable efficiency and effectiveness is a research problem. In this paper, we focus on the data fusion approach to information retrieval, in which each component retrieval system contributes a result and all the results are combined by a combination method. The goal of this research is to find a feasible combination method that is able to balance effectiveness and efficiency. Using 3 groups of historical runs from TREC for the experiment, we find that with the weights trained by weighted linear regression, the linear combination method can achieve good results in effectiveness and efficiency.

In this paper, we investigate the problem of results merging in distributed information retrieval when overlapping databases are used. We focus on two issues: score normalization and weights assignment for each of the component results. Empirical study with the TREC data has the following three findings: 1. The cubic regression model and logistic regression model are better than the commonly used zero-one score normalization method, 2. The weighting scheme of uneven similarity is an effective method of weights assignment. 3. Score normalization and weights assignment can be used separately or together in a results merging method to improve effectiveness. The findings obtained in this paper are very useful for effectiveness improvement when implementing a distributed information retrieval system.