Search

We present a CMOS imager for low-latency face detection empowered by parallel imaging and computing of machine-learning (ML) classifiers. The energy-efficient parallel operation and multi-scale detection eliminate image capture delay and significantly alleviate backend computational loads. The proposed pixel architecture, composed of dynamic samplers in a global shutter (GS) pixel array, allows for energy-efficient in-memory charge-domain computing of feature extraction and classification. The illumination-invariant detection was realized by using log-Haar features. A prototype 240×240 imager achieved an on-chip face detection latency of 5.1ms with a 97.9% true positive rate and 2% false positive rate at 120fps. Moreover, a dynamic nature of in-memory computing allows an energy efficiency of 419pJ/pixel for feature extraction and classification, leading to the smallest latency-energy product of 3.66ms∙nJ/pixel with digital backend processing.

While the performance of convolutional neural networks (CNNs)-based single image super-resolution (SISR) has been greatly improved, the enormous parameter sizes and computational complexity of the underlying CNNs make hardware implementation difficult. Recently, several lightweight SISR methods have been developed, but they still do not consider various structural problems that may occur in hardware implementation. To solve this problem, we propose a slice-based SR using light-weight network (LWN) and a slice-based SR using LWN with relation loss (LWNRL). First, LWN(RL) adopts a slice-based architecture to facilitate system-on-chip (SoC) implementation. Second, LWN(RL) avoids global connection modules that are not suitable for SoC implementation, with minimal performance penalty. Finally, we propose a new loss to improve the performance of LWN without additional cost. Experimental results show that LWNRL achieves significant efficiency of SR model. Especially, the larger the resolution or scale factor, the better the performance of LWNRL than the conventional methods.

We present a neural interface IC for high-precision optical stimulation and electrical recording from implanted optoelectrodes with the highest system integration to date. The system combines 12 high-resolution constant-current LED driver channels, 32 recording channels, and an efficient PMU for operation from a low voltage Li battery on a wireless platform. A novel LED driver architecture features a high-efficiency wide-current-range fully-integrated voltage regulator and a high-resolution pulse shaper to minimize recording artifacts from sharp current pulse edges. A prototype, fabricated in 0.18µm CMOS, occupies 7.35mm2. The LED driver has a peak efficiency of 50% and consumes 31µW quiescent power. We demonstrate in-vivo that the pulse-shaping scheme prevents stimulation artifacts. Key Words: neural interface, LED driver, optogenetics.

We report a neural recording system with embedded lossless compression using spatiotemporal correlation and sparsity of neural signals to reduce dynamic power (P d ) dissipation for data transmission in high-density neural recording systems. We could successfully compress the data rate of neural signals by a factor of 5.35 (local field potential, LFP) and 10.54 (action potential, AP), respectively. Consequently we reduced Pd consumption by 89% while achieving the state-of-the-art recording performance of 3.37 μW/Ch, 5.18 μV rms input-referred noise, and 3.41NEF2Vdd.

We report a CMOS imager embedded with energy-efficient object detection and bio-inspired 2D optic flow generation cores for navigation of nano-air-vehicles (NAVs). The proposed vision-based navigation system employs spatial difference imaging and gradient orientation using mixed-signal circuits to achieve both energy-efficient and area-efficient implementation. The system achieved 272.49 pJ/pixel with 75% reduction in memory size for integrated operation of object detection and 2D optic flow generation.

We report an energy- and area-efficient modular analog front-end (AFE) architecture incorporating Δ-modulated ΔΣ (Δ-ΔΣ) signal acquisition for 1,024-channel brain activity monitoring platforms. The AFE employs spectrum-equalizing and continuous-time (CT)-ΔΣ quantization to make use of the inherent spectral characteristics of brain signals. The dynamic range (DR) of the neural signals has been compressed by 27dB (spectrum equalization). The energy-area product is the most critical figure of merit for massively-parallel recordings and the AFE achieves 4.84fJ/C-s·mm2, the smallest ever reported. The fabricated circuits consume 0.05mm2 and 3.05μW/channel, exhibiting 63.8dB SNDR, 3.02 NEF, and 4.56NEF2V DD .

Implantable biomedicai systems usually operate in energy-limited environments and exhibit large variation of power consumption ranging from constant low-power (bio-signal sensing) to sporadic high-power (stimulation and/or burst data transmission) [1]. Thus, energy-efficient switching power converters that can support a wide range of load variation are necessary to meet this requirement. Pulse frequency modulation (PFM) power converters can support this large load variation with a high power-conversion efficiency (PCE) [2]. However, their unpredictable switching noise degrades the power supply integrity and hence deteriorates the signal quality in implantable biomedical systems. Pulse-width modulation (PWM) power converters can provide predictable output spectrum, but their PCE in light loads suffers. Although various techniques have been explored over the past years [3-5], the PCE of the PWM converters is still not adequate for low current levels (

The Spring Loaded Inverted Pendulum (SLIP) describes the dynamic walking of humans and animals in a simplified manner. However, realizing such movements by means of a combination of typical articulated legs and linear actuators has some limitations. This paper proposes a leg mechanism that accurately reflects the SLIP based on its mechanical constitution. The SLIP design based leg is able to do decoupled swing motion and extension motion. Also this study has focused on improving the straightness of extension motion.

Obstructive sleep apnea (OSA) is a sufficiently common disease that may negatively affect high-quality sleeping. Reduction in muscle strain of body and laxity in the respiratory tract during sleeping can cause apnea in sleep period. OSA decreases blood oxygen saturation and increases effort to breathe to cause arousal via reaction of neurologic mechanisms, indicating that it induces autonomic imbalance by enhancing activities in the sympathetic nervous system. In this context, the purpose of this paper is to identify freeing airways via changing positions during sleeping in order to alleviate causes of airway blockage by muscular laxity while sleeping. The authors caused changes in sleeping positions and airway blockage while sleeping, analyzing the time and the frequency domains of the Heart rate variability (HRV) signals. Among the parameters used in this study, there were statistically significant differences in OSA section and the transformation of those parameters in the standard deviation of NN interval (SDNN) of the time domain and the NLF, the NHF, and the low frequency (LF)/high frequency (HF) of the frequency domain. In conclusion, sleep apnea symptoms may be alleviated via changing positions during sleep apnea.

JINPOONG, a walking robot using hydraulic pressure, requires a power source to operate the power-pack. Therefore, power was supplied to JINPOONG by using an engine that has high power per unit engine weight. Hydraulic pump requires a constant rotating speed in order to supply stable hydraulic pressure because the pressure and flow rate changes according to rotating speed. This study proposes a way to maintain relatively stable rotating speed of engine with irregular and fluctuating load.

Hydraulic actuation has the advantage of substantial power and wide range of speed with a small size and low weight. Because of this advantage, hydraulic actuation is used often for a robot dealing with heavy-weight material or walking fast. However, in a walking robot, the power of hydraulic actuator can act as the ground reaction forces between the feet and the ground surface, breaking the balance of the body. Therefore, compact design of a body including the feet and module design for absorption of the ground reaction forces is critical. This paper explains leg design for compact design of hydraulic actuated quadruped robot. Also, this paper describes about a shock absorber design included in the feet module by walking-in-place experiments.

JINPOONG, a walking robot using hydraulic pressure, requires a power source to operate the power-pack. Therefore, power was supplied to JINPOONG by using an engine that has high power per unit engine weight. Hydraulic pump requires a constant rotating speed in order to supply stable hydraulic pressure because the pressure and flow rate changes according to rotating speed. This study proposes a way to maintain relatively stable rotating speed of engine with irregular and fluctuating load.

The Nuclear power plant (NPP) industries in Korea have been making efforts to reduce the human errors which largely contributed to 120 nuclear reactor trips from the year of 2001 to 2006. This study aims to develop a Crew Resource Management (CRM) training program that helps to improve plant performance by reducing the number of the reactor trips caused by the operators' errors.

Electronic ballast for the ceramic MH lamps is proposed. The ballast consists of active PFC, inverter, power control circuit and igniter. The lamp driving stage is a dual buck inverter which is used to drive lamps with a low-frequency square-wave voltage and current. The dual buck inverter switches are controlled by power control circuit to maintain the rated lamp power. It can not only dim the lamp but eliminate the acoustic resonance. The prototype of electronic ballast driving 250W ceramic MH lamps was verified by experimental results.

An advanced alarm system (AAS) of a NPP (nuclear power plant) is primarily a digital system employing advanced alarm process logics and a VDU (visual display unit) based control and display for the alarms. The role of the AAS is to provide the information necessary to safely shutdown the reactor under all plant conditions, to monitor the plant parameters approaching or exceeding the operating limits and to minimize the number of alarms, and to group and prioritize the alarms. The secondary function of the alarm system is to serve as an alarm and event log. This paper proposes a new approach to an active alarm processing in a NPP by using active database and data mining techniques.

Using the FROG CRAB method, complete temporal reconstruction of attosecond harmonic pulses was achieved. The compensation of intrinsic attosecond chirp was also demonstrated in the harmonic generation medium itself, thereby achieving transform-limited attosecond pulses.

In this paper we propose a practical propagation channel robust algorithm of the block error rate estimation in the orthogonal frequency division multiplexing with the time division multiple access and the orthogonal frequency division multiple access broadband wireless access networks. We estimate the block error rate as a function of two variables, that is, the average signal-to-noise ratio (SNR) and the normalized root mean square value of the SNR. We calculate the average and the root mean square values of the SNR within the set of the SNR per subcarrier values within the coding block our block error rate estimation algorithm adds more flexibility to the link adaptation algorithms. Moreover, it increases robustness of the link adaptation in the unknown propagation channels.

Instead of the CVD method to form a gate oxide layer, a wet-etching method was used to form an insulating layer and isolate cathode layer from gate electrode. The wet-etching process has many advantages such as simple manufacture process, mass production and low cost. A gate triode structure was fabricated because of its simplicity in fabrication process and beam broadening. A new type of triode structure using wet-etching process was made and their field emission characteristics, uniformity and efficiency for light source for backlight units(BLU) were investigated

In this report, we present field emission properties of the carbon nanotube (CNT) paste on cathode with a curved surface for MPA. The CNT paste was prepared using a mixture of multiwalled CNTs powders synthesized by chemical vapor deposition method, organic vehicles, and inorganic binders. We made use of both spin on glass (SOG) and a sensitizer as an inorganic binder and additive to the CNT paste. Then the paste was rubbed over a round cathode (diameter, D-4.3 mm) with a curved surface (a radius of curvature, R/spl sim/9.7 mm) designed specifically for our X-band (8-12 GHz) TWT-MPA.

In this paper, we analyze the IEEE 802.16 OFDMA system with adaptive allocation of the frequency subchannels to the users. We perform adaptive coding and modulation, allocation of the time-frequency resource, the transmit power, and the frequency subchannels to the downlink and uplink service flows, which are scheduled for transmission in the current frame. Moreover, we consider adaptation of the frame boundary between the downlink and uplink subframes. We minimize the sum time-frequency resource, allocated in both downlink and uplink subframes, under the constraints on the maximum transmit power and the time-frequency resource available in the frame, while satisfying the QoS requirements for all downlink and uplink service flows scheduled for transmission. We propose the cross-layer optimization algorithms, which take into account the MAC and PHY layers processing during the optimization. We also present the results of the proposed algorithms evaluation by system level simulation.

Highly efficient high-harmonic generation was achieved in helium using an orthogonally polarized two-color laser field that consisted of the fundamental and second harmonic of a femtosecond Ti:sapphire laser. By applying a higher intensity second harmonic, the harmonics in the two-color field were stronger than those obtained in the fundamental field by more than two orders of magnitude. The conversion efficiency was as high as 5/spl times/10/sup -5/ for the 38/sup th/ harmonic at 21.6 nm.

A novel isolated LDMOS structure, using a simple yet effective concept of an asymmetric hetero-doped source/drain, is proposed. The asymmetric hetero-doped source/drain reduces the on-state resistance of the transistor due to the high n-type doping used for device drain drift, provides excellent ruggedness for parasitic NPN turned-on due to a minimized n+ source spacer, and also raises the device breakdown voltage due to charge compensation in the composite drain drift region. Therefore, the asymmetric hetero-doped source/drain structure allows the isolated LDMOS to have a high current handling capability with a small device size. Measured results show that a 24 V breakdown voltage new device with a low cost two-layer metal (Al) back-end achieves very low R (sp, on) of 16.6 mOmega.mm2. Furthermore, the new device with 65 V high-side capability achieves good isolation performance even when switching source/drain to -20 V and also gets a cut-off frequency of 13 GHz at a gate voltage of 5.5 V

We prepared carbon nanotube (CNT) pastes with different inorganic binders such as glass frit and spin on glass (SOG). MWNT powders grown by CVD were used for electron emissive source. The three-roll mill process was carried out for mixing and dispersion of CNT powders in organic vehicle as polymer matrix. CNT paste was printed onto various substrates such as an indium thin oxide (ITO) coated soda lime glass and a nickel plate. Then CNT paste was sintered under different ambient and temperature. For the analysis of the surface morphology of the cathode layer, field-emission scanning electron microscopy (FESEM) was used. The FE characteristics of CNT paste were measured in a high vacuum chamber with a parallel diode type configuration at 5/spl times/10/sup -6/ Torr. We obtained stable and high emission current from CNT paste with SOG. Our experiments have shown that CNT paste with SOG can use as efficient electron emitter in vacuum nanoelectronics such as radio frequency amplifier, field emission display and X-ray tube.

Magnetic and electronic properties of Fe- and Co-doped TiO/sub 2-/spl delta// thin films are investigated by vibrating sample magnetometer (VSM), Mossbauer spectroscopy, and X-ray photoelectron spectroscopy (XPS). When the precursor films are annealed in vacuum, the resulting oxygen-deficient TiO/sub 2-/spl delta//:Fe and TiO/sub 2-/spl delta//:Co films are found to become semiconducting with p-type carriers in the 10/sup 18/ cm/sup -3/ range. The oxygen-deficient films go through conductivity transitions from n-type to p-type by Fe and Co doping. Results of VSM measurements show that ferromagnetism is exhibited at room temperature. XPS measurements on the thin films reveal that Fe ions have both Fe/sup 2+/ and Fe/sup 3+/ ionic valences while Co ions have Co/sup 2+/ mostly. Also, the density of Fe/sup 2+/ ions is found to decrease after annealing.

The spinel ferrite Fe/sub x/Co/sub 3-x/O/sub 4/ thin film was prepared on silicon substrate by sol-gel method. The evolution of structural, magnetic, and optical properties of Fe composition in the film was investigated by vibrating sample magnetometer (VSM), Mossbauer spectroscopy, X-ray photoelectron spectroscopy (XPS), and spectroscopic ellipsometry.

We have directly grown CNTs on the metal substrates by direct-current plasma-enhanced chemical-vapor deposition (DC-PECVD) using a gas mixture of C/sub 2/H/sub 2/ and NH/sub 3/ at relatively low temperature (550 /spl deg/C). In this study, we report field emission properties of CNTs depending on pattern sizes and spacings. Field-emission characteristic of CNTs was evaluated in a vacuum of 10/sup -6/ Torr in a parallel diode configuration. It is found that there is a correlation between pattern morphologies (such as window area, spacing, and dimension of window edge, etc.) and the field-emission properties. Furthermore, we found that the contact resistance between CNTs and substrate is a crucial factor for the saturation of field emission from CNTs.

The anticipated increase in mobility and popular use of mobile services will require more technologies for authenticating mobile hosts and protecting their privacy. However, issues such as privacy and confidentiality in mobile computing environments and the resource restriction of mobile hosts have not been given enough consideration. In this paper, we present an authentication protocol that provides mobile hosts with the functionalities of authentication and session key distribution for user privacy and secure communication. Considering mobile computing environments, our protocol has design principles such as reducing computation and communication overhead of mobile hosts, and uses random numbers, time-stamps, and authentication codes in order to protect systems from attacks by eavesdroppers. While our protocol supports authentication, secure communication, anonymity, and untraceability for mobile hosts, it can be an efficient authentication protocol for mobile computing environments.

We investigated effects of post treatment of carbon nanotubes (CNTs) on emission characteristics of CNTs. Annealing and plasma treatment of CNTs were carried in different ambient. Emission current from CNTs was increased by annealing in Ar atmosphere at 250/spl deg/C. Uniformity of emission from CNTs was improved by Ar plasma treatment at 250/spl deg/C but, there was degradation in emission properties after O/sub 2/ plasma treatment.