No abstract

Convolutional Neural Networks (CNNs) are widely used in computer vision, natural language processing, and so on, which generally require low power and high efficiency in real applications. Thus, energy efficiency has become a critical indicator of CNN accelerators. Considering that asynchronous circuits have the advantages of low power consumption, high speed, and no clock distribution problems, we design and implement an energy-efficient asynchronous CNN accelerator with a 65 nm Complementary Metal Oxide Semiconductor (CMOS) process. Given the absence of a commercial design tool flow for asynchronous circuits, we develop a novel design flow to implement Click-based asynchronous bundled data circuits efficiently to mask layout with conventional Electronic Design Automation (EDA) tools. We also introduce an adaptive delay matching method and perform accurate static timing analysis for the circuits to ensure correct timing. The accelerator for handwriting recognition network (LeNet-5 model) is implemented. Silicon test results show that the asynchronous accelerator has 30% less power in computing array than the synchronous one and that the energy efficiency of the asynchronous accelerator achieves 1.538 TOPS/W, which is 12% higher than that of the synchronous chip.

Due to the target motion, range cell migration (RCM) and Doppler frequency migration (DFM) always occur. That is harmful to the signal enhancement and detection. In order to solve the problem, a novel three-dimensional (3-D) coherent integration (TDCI) based algorithm is proposed in this paper which consists of three stages. Firstly, a 3-D space is generated by the autocorrelation function. After that, TDCI algorithm is realized and TDCI domain is obtained in which the motion parameters can be accurately estimated. Finally, compensating off the RCM and DFM by the estimates, the target signal can be accumulated and detected in range-Doppler frequency domain. Theoretical analyses and simulation experiments are given to demonstrate that the proposed algorithm is able to deal with the problems of velocity ambiguity, shadow effect, and cross-term with superior resolution. Comparisons with several representative algorithms lead us to the conclusion that the proposed algorithm can strike a good balance between computation cost and anti-noise performance. In the end, real measured data processing and result analysis are carried out, which further verify the effectiveness of the proposed algorithm.

No abstract

Visual search is a routine task used in everyday life and is an important field of research in cognitive psychology. In laboratory settings, it has been shown that search for a target defined by a unique conjunction of two colours is more efficient if one colour surrounds the other (a part-whole search) compared to when no such hierarchical structural relationship exists (a part-part search; Wolfe et al. in Perception & Psychophysics, 55, 537, 1994). A similar result has been shown to hold for size × size conjunction searches (Bilsky & Wolfe in Perception & Psychophysics, 57, 749, 1995). We show that this result also holds for topology × topology conjunction searches (where the stimuli are either hollow or filled), but not for orientation × orientation conjunction searches. We use the simultaneous-sequential paradigm to investigate a possible reason for the inefficiency of part-whole orientation search compared with the efficiency of part-whole searches of other features. We argue that two different attribute values from the same dimension can be processed independently, without interfering with each other for colour, size, and topology, but not for orientation. Because it is obviously more efficient to process a conjunction stimulus when both components of the conjunction can be processed without mutual interference, it follows that colour × colour, size × size, and topological × topology part-whole conjunction searches are likely to be more efficient than orientation × orientation partwhole conjunction searches.