Tohru KANEKO Koji HIROSE Akira MATSUZAWA
A current mirror circuit is often used in Gm-cells and current amplifiers in order to obtain high linearity and high accurate current gain. However, it is expected that a threshold voltage mismatch between transistors pair in the current mirror affects these performances in recent scaled technologies. In this paper, negative effects caused by the mismatch in the current mirror are considered and a new calibration technique for the mismatch issues is proposed. In the current mirror without the mismatch, the high-linearity operation is provided by distortion canceling under the condition that the transistors have the same operating points. The threshold voltage mismatch disturbs the cancellation, therefore the distortion is appeared. In order to address the issue, a new calibration technique using a backgating effect is considered. This calibration can reduce the threshold voltage mismatch directly by controlling the body bias voltage with DACs. According to simulation results with Monte Carlo sampling in 65nm CMOS process, owing to the proposed calibration, the worst HD2 and HD3 are improved by 18.4dB and 11.6dB, respectively. In addition, the standard deviation of the current gain is reduced from 399mdB to 34mdB.
Shen-Li CHEN Yu-Ting HUANG Shawn CHANG
In this study, the reference pure metal-oxide semiconductor field-effect transistors (MOSFETs) and low-voltage (LV) and high-voltage (HV) MOSFETs with a super-junction (SJ) structure in the drain side were experimentally compared. The results show that the drain-side engineering of SJs exerts negative effects on the electrostatic discharge (ESD) and latch-up (LU) immunities of LV n-channel MOSFETs, whereas for LV p-channel MOSFETs and HV n-channel laterally diffused MOSFETs (nLDMOSs), the effects are positive. Compared with the pure MOSFET, electrostatic discharge (ESD) robustness (It2) decreased by approximately 30.25% for the LV nMOS-SJ, whereas It2 increased by approximately 2.42% and 46.63% for the LV pMOS-SJ and HV nLDMOS-SJ, respectively; furthermore, LU immunity (Vh) decreased by approximately 5.45% for the LV nMOS-SJ, whereas Vh increased by approximately 0.44% and 35.5% for the LV pMOS-SJ and HV nLDMOS-SJ, respectively. Thus, nMOS-SJ (pMOS-SJ and nLDMOS-SJ) has lower (higher) It2 and Vh, and this drain-side SJ structure of MOSFETs is an inferior (superior) choice for improving the ESD/LU reliability of LV nMOSs (LV pMOS and HV nLDMOS).
Ami SAKAKIBARA Toshimitsu USHIO
In this paper, we study a control problem of a concurrent discrete event system, where several subsystems are partially synchronized via shared events, under local and global constraints described by linear temporal logic formulas. We propose a hierarchical control architecture consisting of local supervisors and a coordinator. While the supervisors ensure the local requirements, the coordinator decides which shared events to be disabled so as to satisfy the global specification. First, we construct Rabin games to obtain local supervisors. Next, we reduce them based on shared transitions. Finally, we construct a global Rabin game from the reduced supervisors and a deterministic Rabin automaton that accepts every run satisfying the global specification. By solving it, we obtain a coordinator that disables shared events to guarantee the global requirement. Moreover, the concurrent system controlled by the coordinator and the local supervisors is deadlock-free.
Chun-Hao LIAO Makoto SUZUKI Hiroyuki MORIKAWA
Concurrent transmission (CT) is a revolutionary multi-hop protocol that significantly improves the MAC- and network-layer efficiency by allowing synchronized packet collisions. Although its superiority has been empirically verified, there is still a lack of studies on how the receiver survives such packet collisions, particularly in the presence of the carrier frequency offsets (CFO) between the transmitters. This work rectifies this omission by providing a comprehensive evaluation of the physical-layer receiver performance under CT, and a theoretical analysis on the fading duration of the beating effect resulting from the CFO. The main findings from our evaluations are the following points. (1) Beating significantly affects the receiver performance, and an error correcting mechanism is needed to combat the beating. (2) In IEEE 802.15.4 systems, the direct sequence spread spectrum (DSSS) plays such a role in combatting the beating. (3) However, due to the limited length of DSSS, the receiver still suffers from the beating if the fading duration is too long. (4) On the other hand, the basic M-ary FSK mode of IEEE 802.15.4g is vulnerable to CT due to the lack of error correcting mechanism. In view of the importance of the fading duration, we further theoretically derive the closed form of the average fading duration (AFD) of the beating under CT in terms of the transmitter number and the standard deviation of the CFO. Moreover, we prove that the receiver performance can be improved by having higher CFO deviations between the transmitters due to the shorter AFD. Finally, we estimate the AFD in the real system by actually measuring the CFO of a large number of sensor nodes.
Seongkyu MUN Suwon SHON Wooil KIM David K. HAN Hanseok KO
Various types of classifiers and feature extraction methods for acoustic scene classification have been recently proposed in the IEEE Detection and Classification of Acoustic Scenes and Events (DCASE) 2016 Challenge Task 1. The results of the final evaluation, however, have shown that even top 10 ranked teams, showed extremely low accuracy performance in particular class pairs with similar sounds. Due to such sound classes being difficult to distinguish even by human ears, the conventional deep learning based feature extraction methods, as used by most DCASE participating teams, are considered facing performance limitations. To address the low performance problem in similar class pair cases, this letter proposes to employ a recurrent neural network (RNN) based source separation for each class prior to the classification step. Based on the fact that the system can effectively extract trained sound components using the RNN structure, the mid-layer of the RNN can be considered to capture discriminative information of the trained class. Therefore, this letter proposes to use this mid-layer information as novel discriminative features. The proposed feature shows an average classification rate improvement of 2.3% compared to the conventional method, which uses additional classifiers for the similar class pair issue.
Maki ARAI Masashi IWABUCHI Kei SAKAGUCHI Kiyomichi ARAKI
This paper proposes a new methodology to design optimal antennas for MIMO (Multi-Input Multi-Output) communication systems by using spherical mode expansion. Given spatial channel properties of a MIMO channel, such as the angular profile at both sides, the optimal MIMO antennas should provide the largest channel capacity with a constraint of the limited implementation space (volume). In designing a conventional MIMO antenna, first the antenna structure (current distribution) is determined, second antenna directivity is calculated based on the current distribution, and thirdly MIMO channel capacity is calculated by using given angular profiles and obtained antenna directivity. This process is repeated by adjusting the antenna structure until the performance satisfies a predefined threshold. To the contrary, this paper solves the optimization problem analytically and finally gives near optimal antenna structure (current distribution) without any greedy search. In the proposed process, first the optimal directivity of MIMO antennas is derived by applying spherical mode expansion to the angular profiles, and second a far-near field conversion is applied on the derived optimal directivity to achieve near optimal current distributions on a limited surface. The effectiveness of the proposed design methodology is validated via numerical calculation of MIMO channel capacity as in the conventional design method while giving near optimal current distribution with constraint of an antenna structure derived from proposed methodology.
Chunpeng MA Akihiro TAMURA Lemao LIU Tiejun ZHAO Eiichiro SUMITA
Conventional feature-rich parsers based on manually tuned features have achieved state-of-the-art performance. However, these parsers are not good at handling long-term dependencies using only the clues captured by a prepared feature template. On the other hand, recurrent neural network (RNN)-based parsers can encode unbounded history information effectively, but they perform not well for small tree structures, especially when low-frequency words are involved, and they cannot use prior linguistic knowledge. In this paper, we propose a simple but effective framework to combine the merits of feature-rich transition-based parsers and RNNs. Specifically, the proposed framework incorporates RNN-based scores into the feature template used by a feature-rich parser. On English WSJ treebank and SPMRL 2014 German treebank, our framework achieves state-of-the-art performance (91.56 F-score for English and 83.06 F-score for German), without requiring any additional unlabeled data.
Guo-Ming SUNG Leenendra Chowdary GUNNAM Wen-Sheng LIN Ying-Tzu LAI
This work develops a third-order multibit switched-current (SI) delta-sigma modulator (DSM) with a four-bit switched-capacitor (SC) flash analog-to-digital converter (ADC) and an incremental data weighted averaging circuit (IDWA), which is fabricated using 0.18µm 1P6M CMOS technology. In the proposed DSM, a 4-bit SC flash ADC is used to improve its resolution, and an IDWA is used to reduce the nonlinearity of digital-to-analog converter (DAC) by moving the quantization noise out of the signal band by first-order noise shaping. Additionally, the proposed differential sample-and-hold circuit (SH) exhibits low input impedance with feedback and width-length adjustment in the SI feedback memory cell (FMC) to increase the conversion rate. A coupled differential replicate (CDR) common-mode feedforward circuit (CMFF) is used to compensate for the mirror error that is caused by the current mirror. Measurements indicate that the signal-to-noise ratio (SNR), dynamic range (DR), effective number of bits (ENOB), power consumption, and chip area are 64.1 dB, 64.4 dB, 10.36 bits, 18.82 mW, and 0.45 × 0.67 mm2 (without I/O pad), respectively, with a bandwidth of 20 kHz, an oversampling ratio (OSR) of 256, a sampling frequency of 10.24 MHz, and a supply voltage of 1.8 V.
Takashi MAEHATA Suguru KAMEDA Noriharu SUEMATSU
We propose an architecture for a 1-bit band-pass delta-sigma modulator (BP-DSM) that outputs concurrent multiband RF signals. The proposed BP-DSM consists of parallel bandpass filters (BPFs) in the feedback loop to suppress the quantization noise at each target frequency band while maintaining the stability. Each BPF is based on second-order parallel infinite impulse response (IIR) filters. This architecture can unify and reconfigure the split BPFs according to the number of bands. The architecture complexity is proportional to the bandwidth of each RF signal and is independent of the carrier spacing between the bands. The conventional architecture of a concurrent multiband digital modulator, reported previously, has multiple input ports to the dedicated BPF at each band and so it cannot be efficiently integrated. Measurements show that the proposed architecture is feasible for transmitting a concurrent dual-band and a triple-band by changing the 1-bit digital data stream while keeping a data transmission rate of 10Gb/s. We demonstrate that the proposed architecture outputs the signal with LTE intra-band and inter-band carrier aggregation on 0.8GHz, 2.1GHz and 3.5GHz, each with 40MHz bandwidth in 120MHz aggregated bandwidth, whose bandwidth surpasses the bandwidth with carrier aggregation of LTE-A up to 100MHz. Adjacent channel leakage ratios of -49dBc and -46dBc are achieved at 3.5GHz in the concurrent dual-band and triple-band, respectively.
Tohru KANEKO Yuya KIMURA Masaya MIYAHARA Akira MATSUZAWA
60GHz wireless communication requires analog baseband circuits having a bandwidth of about 1GHz. This paper presents a wide bandwidth current-mode low pass filter technique which involves current amplifiers, resistors and capacitors. The proposed current-mode filter is obtained by replacing an integrator employing an op-amp with another integrator employing a current amplifier. With the low input impedance current amplifier having little variation of the input impedance, the proposed filter is expected to improve linearity and power efficiency. The proposed current amplifier which employs super source follower topology with complementary input is suitable for the filter because of its class AB operation. Although simulation results shows the conventional current amplifier which employs super source follower topology without the complementary input has 12Ω variation and 30Ω input impedance, the proposed current amplifier has 1Ω variation and 21Ω input impedance. A fourth order 1GHz bandwidth filter which involves the proposed current amplifiers is designed in a 65nm CMOS technology. The filter can achieve IIP3 of 1.3dBV and noise of 0.6mVrms with power consumption of 13mW under supply voltage of 1.2V according to simulation results with layout parasitic extraction models. Active area of the filter is 380μm×170μm.
Kei IKEDA Atsuki KOBAYASHI Kazuo NAKAZATO Kiichi NIITSU
High-resolution bio-imaging is a key component for the advancement of life science. CMOS electronics is one of the promising candidates for emerging high-resolution devices because it offers nano-scale transistors. However, the resolution of the existing CMOS bio-imaging devices is several micrometers, which is insufficient for analyzing small objects such as bacteria and viruses. This paper presents the results of an analysis of the scalability of a current-mode analog-to-time converter (CMATC) to develop a high-resolution CMOS biosensor array. Simulations were performed using 0.6-µm, 0.25-µm, and 0.065-µm CMOS technology nodes. The Simulation results for the power consumption and resolution (cell size) showed that the CMATC has high-scalability and is a promising candidate to enable high-resolution CMOS bio-imaging.
Shen-Li CHEN Yu-Ting HUANG Yi-Cih WU
Improving robustness in electrostatic discharge (ESD) protection by inserting drain-side isolated silicon-controlled rectifiers (SCRs) in a high-voltage (HV) p-channel lateral-diffused MOSFET (pLDMOS) device was investigated in this paper. Additionally, the effects of anti-ESD reliability in the HV pLDMOS transistors provided by this technique were evaluated. From the experimental data, it was determined that the holding voltage (Vh) values of the pLDMOS with an embedded npn-arranged SCR and discrete thin-oxide (OD) layout on the cathode side increased as the parasitic SCR OD row number decreased. Moreover, the trigger voltage (Vt1) and the Vh values of the pLDMOS with a parasitic pnp-arranged SCR and discrete OD layout on the drain side fluctuated slightly as the SCR OD-row number decreased. Furthermore, the secondary breakdown current (It2) values (i.e., the equivalent ESD-reliability robustness) of all pLDMOS-SCR npn-arranged types increased (>408.4%) to a higher degree than those of the pure pLDMOS, except for npn-DIS_3 and npn-DIS_2, which had low areas of SCRs. All pLDMOS-SCR pnp-arranged types exhibited an increase of up to 2.2A-2.4A, except for the pnp_DIS_3 and pnp_DIS_2 samples; the pnp_DIS_91 increased by approximately 2000.9% (249.1%), exhibiting a higher increase than that of the reference pLDMOS (i.e., the corresponding pnp-stripe type). The ESD robustness of the pLDMOS-SCR pnp-arranged type and npn-arranged type with a discrete OD layout on the SCR cathode side was greater than that of the corresponding pLDMOS-SCR stripe type and a pure pLDMOS, particularly in the pLDMOS-SCR pnp-arranged type.
Recurrent neural networks (RNNs) are a powerful model for sequential data. RNNs that use long short-term memory (LSTM) cells have proven effective in handwriting recognition, language modeling, speech recognition, and language comprehension tasks. In this study, we propose LSTM conditional random fields (LSTM-CRF); it is an LSTM-based RNN model that uses output-label dependencies with transition features and a CRF-like sequence-level objective function. We also propose variations to the LSTM-CRF model using a gate recurrent unit (GRU) and structurally constrained recurrent network (SCRN). Empirical results reveal that our proposed models attain state-of-the-art performance for named entity recognition.
Nobutaka KITO Kazushi AKIMOTO Naofumi TAKAGI
A floating-point multiplier with concurrent error detection capability by partial duplication is proposed. It uses a truncated multiplier for checking of the significand (mantissa) multiplication instead of full duplication. The proposed multiplier can detect any erroneous output with error larger than one unit in the last place (1 ulp) of the significand, which may be overlooked by residue checking. Its circuit area is smaller than that of a fully duplicated one. Area overhead of a single-precision multiplier is about 78% and that of a double-precision one is about 65%.
Mizuki IKEYA Takashi NOGUCHI Takafumi KOJIMA Takeshi SAKAI
In this paper, we describe the fabrication of low leakage Superconductor/Insulator/Superconductor (SIS) junctions with a Nb/Al/AlOx/Al/Nb structure. In other words, an extra Al layer was added onto the top of the insulator in a conventional Nb/Al/AlOx/Nb junction. We measured the current and voltage (IV) characteristics of both the Nb/Al/AlOx/Al/Nb and Nb/Al/AlOx/Nb junctions at the temperature of liquid helium, and found that the sub-gap leakage current in the Nb/Al/AlOx/Al/Nb junctions was much lower than that of the Nb/Al/AlOx/Nb junctions. Our analysis of the IV characteristics indicates that the quality of the AlOx insulator used in the Nb/Al/AlOx/Al/Nb junction was close to ideal, while the insulator used in the Nb/Al/AlOx/Nb junction had possible defects. According to the scanning transmission electron microscope (STEM) images and energy-dispersive X-ray spectroscopy (EDX) analyses, it was evident that the Nb atoms diffused into the bottom electrode of the Nb/Al/AlOx/Nb junction, while a smaller number diffused into the bottom electrode of the Nb/Al/AlOx/Al/Nb junction. Therefore, we conclude that the extra Al layer effectively acted as a buffer layer that prevented the Nb atoms from diffusing into the insulator and bottom electrode. The presence of the top Al layer is expected to favorably improve the quality of junctions with a very high current density, and support the extension of the RF and IF bandwidths of SIS mixers.
Toshiki YAMADA Yoshihiro HARUYAMA Katsuyuki KASAI Takahiro KAJI Yukihiro TOMINARI Shukichi TANAKA Akira OTOMO
We prepared a bR thin film by the wire-bar coating technique, and investigated the transient photo-current characteristics of the bR photocell. The transient photo-current signal of bR photocells prepared by the wire-bar coating technique and the dip coating technique was compared. An almost identical transient photo-current signal intensity was obtained both for the wire-bar coating technique and dip coating technique, while the thickness of bR thin film prepared by the wire-bar coating technique is slightly thinner than that prepared by the dip-coating technique. Transparent conductive oxide dependence of the transient photo-current signal is almost the same dependence for the bR photocells with a bR thin film prepared by both techniques. Application of the wire-bar coating technique is significant from the viewpoints of the bR's sample consumption as well as simplicity of sample preparation.
Wei HAN Xiongwei ZHANG Meng SUN Li LI Wenhua SHI
In this letter, we propose a novel speech separation method based on perceptual weighted deep recurrent neural network (DRNN) which incorporate the masking properties of the human auditory system. In supervised training stage, we firstly utilize the clean label speech of two different speakers to calculate two perceptual weighting matrices. Then, the obtained different perceptual weighting matrices are utilized to adjust the mean squared error between the network outputs and the reference features of both the two clean speech so that the two different speech can mask each other. Experimental results on TSP speech corpus demonstrate that the proposed speech separation approach can achieve significant improvements over the state-of-the-art methods when tested with different mixing cases.
All the existing sender-based message logging (SBML) protocols share a well-known limitation that they cannot tolerate concurrent failures. In this paper, we analyze the cause for this limitation in a unicast network environment, and present an enhanced SBML protocol to overcome this shortcoming while preserving the strengths of SBML. When the processes on different nodes execute a distributed application together in a broadcast network, this new protocol replicates the log information of each message to volatile storages of other processes within the same broadcast network. It may reduce the communication overhead for the log replication by taking advantage of the broadcast nature of the network. Simulation results show our protocol performs better than the traditional one modified to tolerate concurrent failures in terms of failure-free execution time regardless of distributed application communication pattern.
Xin WANG Shinji TAKAKI Junichi YAMAGISHI
Building high-quality text-to-speech (TTS) systems without expert knowledge of the target language and/or time-consuming manual annotation of speech and text data is an important yet challenging research topic. In this kind of TTS system, it is vital to find representation of the input text that is both effective and easy to acquire. Recently, the continuous representation of raw word inputs, called “word embedding”, has been successfully used in various natural language processing tasks. It has also been used as the additional or alternative linguistic input features to a neural-network-based acoustic model for TTS systems. In this paper, we further investigate the use of this embedding technique to represent phonemes, syllables and phrases for the acoustic model based on the recurrent and feed-forward neural network. Results of the experiments show that most of these continuous representations cannot significantly improve the system's performance when they are fed into the acoustic model either as additional component or as a replacement of the conventional prosodic context. However, subjective evaluation shows that the continuous representation of phrases can achieve significant improvement when it is combined with the prosodic context as input to the acoustic model based on the feed-forward neural network.
The 60 GHz band compact-range communication is very promising for short-time, short distance communication. Unfortunately, due to the short wavelengths in this frequency band the shadowing effects caused by human bodies, furniture, etc are severe and need to be modeled properly. The numerical methods like the finite-difference time-domain method (FDTD), the finite-element method (FEM), the method of moments (MoM) are unable to compute the field scattered by large objects due to their excessive time and memory requirements. Ray-based approaches like the geometrical theory of diffraction (GTD), uniform geometrical theory of diffraction (UTD), uniform asymptotic theory of diffraction (UAT) are effective and popular solutions but suffer from computation of corner-diffracted field, field at the caustics. Fresnel zone number (FZN) adopted modified edge representation (MER) equivalent edge current (EEC) is an accurate and fast high frequency diffraction technique which expresses the fields in terms of line integration. It adopts distances, rather than the angles used in GTD, UTD or UAT but still provides uniform and highly accurate fields everywhere including geometrical boundaries. Previous work verified this method for planar scatterers. In this work, FZN MER EEC is used to compute field distribution in the millimeter-wave compact range communication in the presence of three dimensional scatterers, where shadowing effects rather than multi-path dominate the radio environments. First, circular cylinder is disintegrated into rectangular plate and circular disks and then FZN MER is applied along with geodesic path loss. The dipole wave scattering from perfectly conducting circular cylinder is discussed as numerical examples.