This study discusses a structural relaxation model for stressed thermal oxide films grown on Si (100) substrates at a temperature that is too low for viscous flow to relax an Si thermal oxide film. This relaxation model had been developed on the basis of activation energy analysis through ab initio calculations on transformations between different crystal types. The analysis revealed a systematic crystal transformation among three types of cristobalite crystals with their domain structures on the substrate. The findings of this study are expected to provide a correct understanding of the relaxation mechanism of oxidized films on a Si (100) surface using a dry-O-2 gas.
Missing-pixel-based image restoration is a promising approach for lightweight image transmission, where a subset of pixels is intentionally removed and reconstructed by a neural network at the decoder. Conventional restoration models typically rely on both upper- and lower-side pixels of the target block, but this requirement introduces latency because the restoration must wait for future lines to be decoded. When the input is restricted to upper-side pixels only, the restoration accuracy significantly degrades due to limited contextual information. This study introduces a lightweight sequential inference mechanism that reuses previously restored pixel values as auxiliary inputs for subsequent block restoration. This sequential inference effectively enlarges the available context without increasing the physical input window or relying on heavy recurrent architectures. A generation-based training scheme (gen-1 to gen-5) was adopted to simulate realistic prediction noise and stabilize sequential inference. Experiments were conducted on 2x2 block restoration using two structured missing-pixel patterns. The results show that sequential inference consistently improves performance over the conventional upper-input model. In the smallest 4x3 input window, the proposed method outperformed the Upper model by more than 1 dB and even surpassed the Normal model that uses both upper- and lower-side pixels. In contrast, for larger windows such as 6x4 and 10x6, the improvement was minimal because sufficient context was already available. These findings indicate that sequential inference provides a practical enhancement for lightweight, low-latency codecs, particularly under limited-context conditions where conventional models struggle.
Realistically simulating natural walking is essential to create an immersive virtual reality experience. However, differences between the virtual reality environment (VRE) and real environment (RE) can alter gait characteristics. Therefore, this study aimed to investigate differences in lower limb spatiotemporal parameters, joint kinematics, and muscle activity between overground walking in a VRE and RE. A total of 13 participants walked at 3 cadences (60 steps per minute [SPM], 80 SPM, and 100 SPM) in the VRE and RE. Motion capture and electromyography were employed to collect the spatiotemporal gait parameters, muscle activities of the right tibialis anterior and medial gastrocnemius muscles, and right lower limb joint angles of the participants. Our results showed that overground walking in the VRE altered several spatiotemporal gait parameters. In addition, the mean electromyography activity of the tibialis anterior and medial gastrocnemius muscles was reduced in the VRE during the initial double- and single-support phases. Most measured biomechanical parameters showed no significant interaction between the walking cadence and environment. This study clarifies biomechanical differences underlying the more cautious gait observed in a VRE, offering practical implications to enhance usability and optimize virtual reality locomotion design.
There is a growing amount of research on the association of physical activity level (PAL) and physical activity-related energy expenditure (PAEE) with physical activity (PA) intensity or body composition variables; however, the age of the target population, especially young adults, is limited. In this study, we aimed to investigate the association between PA indices such as each PA intensity, including sedentary behaviour (SB), and free-living energy expenditure using the doubly labelled water (DLW) method. Using the DLW method under free-living conditions, total energy expenditure was evaluated in 32 healthy women without regular exercise habits in their 20 s. PA, including the percentage of time spent engaging in light, moderate, and vigorous physical intensities, was measured using a validated accelerometer. The %SB was also calculated. Partial correlation coefficients of the relationship between the PAL and PA variables, including sleeping time, age, and body composition indices, were calculated while controlling for the effect of multiple variables. PAEE normalised for body weight (PAEE/kg) and average PAL were 10.2±3.0 kcal/kg/d (42.7±12.7 kJ/kg/d) and 1.60±0.17, respectively. PAL and PAEE/kg had significant negative partial correlations with %SB (r=-0.428, p=0.033 and r=-0.484, p=0.014, respectively). The results indicate that reducing SB attributes to higher values of PAL and PAEE/kg. This is the first study to clarify the contribution of SB and other PA indices to the intensity of PAL and free-living energy expenditure using DLW in young adult women without regular exercise habits.
Small-signal C–V characterization is widely used to extract parameters in simulation models of gallium nitride (GaN) high-electron-mobility transistors (HEMTs). However, it fails to reproduce the gate charge behavior under dynamically varying bias conditions during the switching transient, resulting in significant inaccuracies in circuit simulations. This work presents a large-signal characterization method that captures the dynamic gate-charging behavior of Schottky-gate p-GaN HEMTs. Switching tests are conducted under two distinct bias conditions to extract the gate-source charge profile, which is subsequently subtracted from the total gate charge to obtain the gate-drain charge profile. The extracted C–V characteristics are compared with the industry-standard SPICE model and also with the conventional small-signal measurements, clearly demonstrating the necessity of large-signal characterization for accurate modeling of GaN HEMT gate capacitances.
