Search

Most of the recent state-of-the-art results for speaker verification are achieved by X-vector and its subsequent variants. In this paper, we propose a new network architecture which aggregates the channel and context interdependence features from multi aspect based on Time Delay Neural Network (TDNN). Firstly, we use the SE-Res2Blocks as in ECAPA-TDNN to explicitly model the channel interdependence to realize adaptive calibration of channel features, and process local context features in a multi-scale way at a more granular level compared with conventional TDNN-based methods. Secondly, we explore to use the encoder structure of Transformer to model the global context interdependence features at an utterance level which can capture better long term temporal characteristics. Before the pooling layer, we aggregate the outputs of SE-Res2Blocks and Transformer encoder to leverage the complementary channel and context interdependence features learned by themself respectively. Finally, instead of performing a single attentive statistics pooling, we also find it beneficial to extend the pooling method in a multi-head way which can discriminate features from multiple aspect. The proposed MACCIF-TDNN architecture can outperform most of the state-of-the-art TDNN-based systems on VoxCeleb1 test sets., 6 pages. arXiv admin note: text overlap with arXiv:2005.07143 by other authors

A novel Pico cell cite selection scheme in heterogeneous networks (HetNet) is proposed in this paper. Unlike traditional deployment of Pico cells, we consider the scenario with Pico cells at the edge instead of under the umbrella of macro cells in HetNet. In the novel scheme, one Pico cell eNodeB (PeNB) with directional antennas is deployed on the border of neighbor macro cells which is equivalent to a combiner of PeNBs to associate MeNBs around. Compared with the traditional scheme, we have proved that the novel Pico cell cite selection performs better in both fairness and efficiency, in terms of coverage area, offload rate, and field strength for both edge users and overall users in Pico cells with fixed system parameters. Meanwhile, performance gains provided by novel Pico cell site selection is also derived with range expansion consideration. Our results show that, compared with the traditional scheme, more than 0.6944% macro cell users are offloaded to Pico cell in the novel scheme and the offload rate is enhanced by 98.4%. Furthermore, the field strength improved by the novel scheme is twice as much as the traditional scheme.

In this paper few case studies of Auger elemental analysis in failure analysis of wafer fabrication, using a state-of-the-art scanning Auger nanoprobe, will be presented. Material identification in particle defects is quite challenging especially in advanced microelectronic technologies, 90nm and beyond, where, due to decreasing device size, the tight pitch and high aspect ratio of features introduces constraints of analysis. Thus, the case studies discussed in the paper are all pertaining to the front-end related defects, at poly+contact region of the die, where these contraints are expected to be severe. It will be shown that the scanning Auger nanoprobe can be used for the elemental analysis of features as small as 10nm.

Chemical mechanical polishing (CMP) process has been used for dielectric planarization, and contact and via formation in aluminum (Al)-tungsten (W) metallization. However, more challenges are encountered when it is used in copper (Cu) damascene process because copper is very easily corroded in a wet environment and its oxidation is not self-limiting like Al. Some new and unique defects would be introduced during copper CMP process. These would contribute to the defect density for the present or even the next few layers and ultimately influence the final yield of the wafer. In this paper, a very unusual defect, copper dendrite, which is due to the CMP process, would be discussed.