• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.1-15
• /
• 2022

The MobileNetV3 is specially designed for mobile devices with limited memory and computing power. To reduce the network parameters and improve the network inference speed, a new lightweight network is proposed based on MobileNetV3. Firstly, to reduce the computation of residual blocks, a partial residual structure is designed by dividing the input feature maps into two parts. The designed partial residual structure is used to replace the residual block in MobileNetV3. Secondly, a dual-path feature extraction structure is designed to further reduce the computation of MobileNetV3. Different convolution kernel sizes are used in the two paths to extract feature maps with different sizes. Besides, a transition layer is also designed for fusing features to reduce the influence of the new structure on accuracy. The CIFAR-100 dataset and Image Net dataset are used to test the performance of the proposed partial residual structure. The ResNet based on the proposed partial residual structure has smaller parameters and FLOPs than the original ResNet. The performance of improved MobileNetV3 is tested on CIFAR-10, CIFAR-100 and ImageNet image classification task dataset. Comparing MobileNetV3, GhostNet and MobileNetV2, the improved MobileNetV3 has smaller parameters and FLOPs. Besides, the improved MobileNetV3 is also tested on CPU and Raspberry Pi. It is faster than other networks

• Proceedings of the Korean Vacuum Society Conference
• /
• 2000.02a
• /
• pp.173-173
• /
• 2000

Multilayered films (MLF) consisting of transition metals and semiconductors have drawn a great deal of interest because of their unique properties and potential technological applications. Fe/Si MLF are a particular topic of research due to their interesting antiferromagnetic coupling behavior. although a number of experimental works have been done to understand the mechanism of the interlayer coupling in this system, the results are controversial and it is not yet well understood how the formation of an iron silicide in the spacer layers affects the coupling. The interpretation of the coupling data had been hampered by the lack of knowledge about the intermixed iron silicide layer which has been variously hypothesized to be a metallic compound in the B2 structure or a semiconductor in the more complex B20 structure. It is well known that both magneto-optical (MO0 and optical properties of a metal depend strongly on their electronic structure that is also correlated with the atomic and chemical ordering. In order to understand the structure and physical properties of the interfacial regions, Fe/Si multilayers with very thin sublayers were investigated by the MO and optical spectroscopies. The Fe/si MLF were prepared by rf-sputtering onto glass substrates at room temperature with a totall thickness of about 100nm. The thicknesses of Fe and Si sublayers were varied from 0.3 to 0.8 nm. In order to understand the fully intermixed state, the MLF were also annealed at various temperatures. The structure and magnetic properties of Fe/Si MLF were investigated by x-ray diffraction and vibrating sample magnertometer, respectively. The MO and optical properties were measured at toom temperature in the 1.0-4.7 eV energy range. The results were analyzed in connection with the MO and optical properties of bulk and thin-film silicides with various structures and stoichiometries.

• Korean Journal of Materials Research
• /
• v.32 no.12
• /
• pp.522-527
• /
• 2022

Among efforts to improve techniques for the chemical vapor deposition of large-area and high-quality graphene films on transition metal substrates, being able to reliably transfer these atomistic membranes onto the desired substrate is a critical step for various practical uses, such as graphene-based electronic and photonic devices. However, the most used approach, the wet etching transfer process based on the complete etching of metal substrates, remains a great challenge. This is mainly due to the inevitable damage to the graphene, unintentional contamination of the graphene layer, and increased production cost and time. Here, we report the systematic study of an H₂ bubbling-assisted transfer technique for graphene films grown on Cu foils, which is nondestructive not only to the graphene film but also to the Cu substrate. Also, we demonstrate the origin of the graphene film tearing phenomenon induced by this H₂ bubbling-assisted transfer process. This study reveals that inherent features are produced by rolling Cu foil, which cause a saw-like corrugation in the poly(methyl methacrylate) (PMMA)/graphene stack when it is transferred onto the target substrate after the Cu foil is dissolved. During the PMMA removal stage, the graphene tearing mainly appears at the apexes of the corrugated PMMA/graphene stack, due to weak adhesion to the target substrate. To address this, we have developed a modified heat-press-assisted transfer technique that has much better control of both tearing and the formation of residues in the transferred graphene films.

• Nuclear Engineering and Technology
• /
• v.56 no.1
• /
• pp.247-256
• /
• 2024

The glass series modified by tungsten oxide was created using the compounds (75-x) B₂O₃- 10SrCO₃- 8TeO₂- 7ZnO - xWO₃, where x = 0, 1, 5, 10, 22, 27, 34, and 40% mole percentage. A UV-visible spectrophotometer and thermogravimetric-differential thermal analysis (TG-DTA) methods were employed to characterize the specimen's optical and phase transition attributes, respectively. The mass-attenuation coefficient (AC) of all created glasses from BSTZW0 to BSTZ7 was estimated using Geant4 code from 0.05 to 3 MeV and compared to the XCOM software results, with a relative difference of less than 2% between the two results. The increase of WO₃ percentage lead to an increase in the Linear-AC at each studied energy, and this is mainly due to the fact that the higher the percentage of WO₃ in the glass increases its density which causes an increase in the Linear-AC, so an energy of 0.06 MeV, as an example, the values of the Linear-AC was 4.009, 4.509, 5.442, 6812, 8.564, 9.856, 10.999 and 11.628 cm^-1 form BSTZW0 too BSTZW7, respectively. The Half-VL (value layer), Mean-FP (free path), Tenth-VL, and Radiation attenuation performance (RAP) were also calculated for the current BSTZW-glass samples and revealed that BSTZW7 had the best gamma ray attenuation performance at all discussed energies when compared to other studied glass samples.

• Journal of the Mineralogical Society of Korea
• /
• v.16 no.3
• /
• pp.223-232
• /
• 2003

The co-precipitation method is applied to synthesize the cathode material Li[L $i_{x}$M $n_{1-x-y}$C $r_{y}$ ] $O_2$ for lithium rechargeable batteries at $650^{\circ}C$ (CR650) and 8$50^{\circ}C$ (CR850), respectively. Rietveld indices indicate that $R_{wp}$ with respect to $R_{exp}$ ( $R_{wp}$/ $R_{exp}$) are 9.2%/10.1% for CR650 and 15.9%/9.76% for CR850, respectively. $R_{B}$ and S (GofF) shows 10.9%, 8.54% and 1.9, 1.6, respectively. Rietveld structure refinement reveals that layer structure of LiMn $O_2$ (R3m) coexists with lower symmetry of Li[L $i_{1}$3/M $n_{2}$3/] $O_2$ (C2/c) due to superlattice ordering of Li and Mn in metal-transition containing layers. Unit-cell parameters are calculated as a=2.8520(2)$\AA$, c=14.248(2)$\AA$, V=100.40(l)$\AA^3$ for CR650, and a=2.8504(1)$\AA$, c=14.2371(7)$\AA$, V=100.179(8)$\AA^3$ for CR850. Final chemistry is obtained as Li[L $i_{0.35}$M $n_{0.56}$C $r_{0.09}$] $O_2$ (CR650) and Li[L $i_{0.27}$M $n_{0.61}$C $r_{0.13}$] $O_2$ (CR850), respectively.y...y..vely.y...y..

• Proceedings of the Korean Vacuum Society Conference
• /
• 2016.02a
• /
• pp.292-292
• /
• 2016

Transition metal dichalcogenides (TMDs) with a two-dimensional layered structure have been considered highly promising materials for next-generation flexible, wearable, stretchable and transparent devices due to their unique physical, electrical and optical properties. Recent studies on TMD devices have focused on developing a suitable doping technique because precise control of the threshold voltage ($V_{TH}$) and the number of tightly-bound trions are required to achieve high performance electronic and optoelectronic devices, respectively. In particular, it is critical to develop an ultra-low level doping technique for the proper design and optimization of TMD-based devices because high level doping (about $10^{12}cm^{-2}$) causes TMD to act as a near-metallic layer. However, it is difficult to apply an ion implantation technique to TMD materials due to crystal damage that occurs during the implantation process. Although safe doping techniques have recently been developed, most of the previous TMD doping techniques presented very high doping levels of ${\sim}10^{12}cm^{-2}$. Recently, low-level n- and p-doping of TMD materials was achieved using cesium carbonate ($Cs_2CO_3$), octadecyltrichlorosilane (OTS), and M-DNA, but further studies are needed to reduce the doping level down to an intrinsic level. Here, we propose a novel DNA-based doping method on $MoS_2$ and $WSe_2$ films, which enables ultra-low n- and p-doping control and allows for proper adjustments in device performance. This is achieved by selecting and/or combining different types of divalent metal and trivalent lanthanide (Ln) ions on DNA nanostructures. The available n-doping range (${\Delta}n$) on the $MoS_2$ by Ln-DNA (DNA functionalized by trivalent Ln ions) is between $6{\times}10^9cm^{-2}$ and $2.6{\times}10^{10}cm^{-2}$, which is even lower than that provided by pristine DNA (${\sim}6.4{\times}10^{10}cm^{-2}$). The p-doping change (${\Delta}p$) on $WSe_2$ by Ln-DNA is adjusted between $-1.0{\times}10^{10}cm^{-2}$ and $-2.4{\times}10^{10}cm^{-2}$. In the case of Co-DNA (DNA functionalized by both divalent metal and trivalent Ln ions) doping where $Eu^{3+}$ or $Gd^{3+}$ ions were incorporated, a light p-doping phenomenon is observed on $MoS_2$ and $WSe_2$ (respectively, negative ${\Delta}n$ below $-9{\times}10^9cm^{-2}$ and positive ${\Delta}p$ above $1.4{\times}10^{10}cm^{-2}$) because the added $Cu^{2+}$ ions probably reduce the strength of negative charges in Ln-DNA. However, a light n-doping phenomenon (positive ${\Delta}n$ above $10^{10}cm^{-2}$ and negative ${\Delta}p$ below $-1.1{\times}10^{10}cm^{-2}$) occurs in the TMD devices doped by Co-DNA with $Tb^{3+}$ or $Er^{3+}$ ions. A significant (factor of ~5) increase in field-effect mobility is also observed on the $MoS_2$ and $WSe_2$ devices, which are, respectively, doped by $Tb^{3+}$-based Co-DNA (n-doping) and $Gd^{3+}$-based Co-DNA (p-doping), due to the reduction of effective electron and hole barrier heights after the doping. In terms of optoelectronic device performance (photoresponsivity and detectivity), the $Tb^{3+}$ or $Er^{3+}$-Co-DNA (n-doping) and the $Eu^{3+}$ or $Gd^{3+}$-Co-DNA (p-doping) improve the $MoS_2$ and $WSe_2$ photodetectors, respectively.

• Journal of the Korean Vacuum Society
• /
• v.18 no.2
• /
• pp.127-132
• /
• 2009

We have investigated the intrinsic defects remaining in epitaxial GaSb layers grown on SI-GaAs substrates compared to those in bulk GaSb crystal substrate, which is a basic material of Sb-based strained-layer superlattice infrared photodetectors. From the functional dependence of the band-to-band transition energy of the photomuminescence (PL) spectra observing up to near room-temperature (250 K), the temperature parameters of [$E_o$, $\alpha$, $\beta$] of undoped GaSb crystal are determined by using the Varshni empirical equation describing the temperature variation of the bandgap energy. Additionally to the antisite-Ga ([$Ga_{Sb}$]) with an ionization energy of 29 meV that is well known to a major intrinsic defect in GaSb, epitaxial GaSb layers show a pair of deep states at the emission energy of 732/711 meV that may be related with a complex of two antisite-Ga and antisite-Sb ([$Ga_{Sb}-Sb_{Ga}$]). Based on the analysis of the temperature and the excitation-power dependences of PL, it suggests that excess-Sb substitutes Ga-site by self-diffusion and two anti sites of [$Ga_{Sb}$] and [$Sb_{Ga}$] could form as a complex of [$Ga_{Sb}-Sb_{Ga}$] in GaSb epilayers grown under Sb-rich condition.

• Korean Chemical Engineering Research
• /
• v.54 no.1
• /
• pp.120-126
• /
• 2016

Eco-friendly acrylic resin/clay nanocomposites containing pristine montmorillonite (PM) or modified clays (30B and 25A) were prepared from acrylic and styrenic monomers using non-aqueous dispersion (NAD) polymerization. Effect of nanoclays on physical properties of polymerization product and resulting nanocomposites was investigated. In view of NAD particle stability, addition of nanoclay at the beginning of polymerization is proved to be good. Results of gel fraction, acid value and viscosity of the NAD product showed that nanocomposites containing clay 25A showed better physical properties than the ones with other clays. GPC results exhibit the increase in molecular weight and decrease in polydispersity index for the 25A nanocomposite. Increase in layer distance confirmed from XRD analysis showed good dispersion of 25A in the nanocomposite. Thermal and dynamic mechanical analysis showed that highest glass transition temperature and storage modulus for 25A nanocomposites. These results indicate that 25A nanoclay gives the best properties in the process of non-aqueous dispersion polymerization of acrylic resin/nanoclay nanocomposites.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.44 no.11
• /
• pp.127-136
• /
• 2007

An abrupt scene-change frame is one that is hardly correlated with the previous frames. In that case, because an intra-coded frame has less distortion than an inter-coded one, almost all macroblocks are encoded in intra mode. This breaks up the rate control flow and increases the number of bits used. Since the reference software for H.264 takes no special action for a scene-change frame, several studies have been conducted to solve the problem using the quadratic R-D model. However, since this model is more suitable for inter frames, the existing schemes are unsuitable for computing the QP of the scene-change intra frame. In this paper, an improved rate control scheme accounting for the characteristics of intra coding is proposed for scene-change frames. The proposed scheme was validated using 16 test sequences. The results showed that the proposed scheme performed better than the existing H.264 rate control schemes. The PSNR was improved by an average of 0.4-0.6 dB and a maximum of 1.1-1.6 dB. The PSNR fluctuation was also in proved by an average of 18.6 %.

• Journal of Korea Water Resources Association
• /
• v.53 no.1
• /
• pp.1-13
• /
• 2020

The most common approach for computing engineering flow problems at high Reynolds number is still the Reynolds-averaged Navier-Stokes (RANS) computations based on turbulence models with wall functions. The recently developed generalized wall functions blending between the wall-limiting viscous and the outer logarithmic relations ensure a smooth transition of flow quantities across two regions. The performances and convergence properties of widely used turbulence models with wall functions that are applicable for turbulence kinetic energy (TKE), turbulent and specific dissipation rates, and eddy viscosity are presented through a series of near wall flow simulations. The present results show that RNG k-𝜖 model should be carefully applied with small tolerance to get the stable solution when the first grid lies in the buffer layer. The standard k-𝜖 and RNG k-𝜖 models are not sensitive to the selection of wall functions for both TKE and eddy viscosity, while the k-ω SST model should be applied together with kL-wall function for TKE and nutUB-wall functions for eddy viscosity to ensure accurate and stable boundary conditions. The applications to a backward-facing step flow at Re=155,000 reveal that the reattachment length is reasonably well predicted on appropriately refined mesh by all turbulence models, except the standard k-𝜖 model which about 13% underestimates the reattachment length regardless of the grid refinement.