This review is concerned with a physical cleaning method based on ultrasound-superposed water jets that collide with cleaning targets and spread over their surface; this method is called ultrasonic water flow cleaning and can be used in semiconductor manufacturing processes. Visualization of the acoustic wave and cavitation bubbles in ultrasonic water flow is presented to examine the role of bubbles oscillating under ultrasound irradiation in the process of removing small silica particles from glass surfaces. The visualization results suggest that hydrodynamic force from single-phase water flow does not suffice for particle removal and that acoustic cavitation bubbles play a major role in particle removal as in the case of ultrasonic bath immersion cleaning.

PVA roller brushes are widely used for cleaning after the CMP process, one of the semiconductor manufacturing processes. The PVA roller brush features protrusions called nodules and rotates on the semiconductor wafer in cleaning. This article introduces a cleaning model that uses that PVA roller brush to remove nanoscale impurities. The evanescent fields on a prism enable us to observe the contact behavior of the brush nodules and clarify that there is little brush volume near the surface during sliding. We classified brush deformation into three types depending on the relative velocity of the wafer and nodule. A stamped contact occurs at a negative relative velocity, i.e., when the wafer overtakes the nodule, and this contact is related to cross-contamination from the brush. Finally, we present a model where water absorption and desorption associated with nodule volume deformation plays an important role in nanoscale impurities removal.

In the semiconductor device manufacturing process, wet cleaning is an important process that determines product yield. In this paper, spray cleaning in the wet process of semiconductor device manufacturing is described from the perspective of macroscopic fluid dynamics. When micrometer-order particles adhere to the substrate, the van der Waals force, as discussed in DLVO theory, is dominant. When these particles are removed by spraying, the fluid drag force on the particles is a major factor. In addition, in semiconductor device cleaning, it is not enough to simply increase the fluid drag; as a trade-off, increasing the fluid drag also increases the probability of pattern collapse and electrostatic damage. Thus, as semiconductor device miniaturization progresses, cleaning methods with even higher selectivity are needed.

A convolutional neural network (CNN)-based image analysis method was developed to identify flow regimes of upward gas-liquid two-phase flow. First, two-phase flow images were taken at 377 different flow conditions, ranging from bubbly to annular flow. Also, based on the time-series void fraction data measured by the instrumentations, selected flow conditions were identified which allow clear identification of flow regimes, and the information were associated with the images acquired for each flow condition as correct labels. Thereafter, one simple convolutional neural network defined with a simple structure and seven types of high-precision neural networks provided by the machine learning framework were prepared. They were trained with labeled two-phase flow images, and eight different deep learning models were built. Using those models to identify all images, the results showed that all CNN-based models identified the flow regime with sufficiently high accuracy in conditions where the correct label was given. For the images in the flow regime transition region, we focused on the output values of the model corresponding to the predicted probability of each flow regime. The time-averaged value of the predicted probability was shown to change gradually with changes in superficial gas velocity. By fitting this change with a continuous function, a quantitative definition of the transition region was attempted. The flow pattern transition region, defined by changes in predicted probability, was compared to the Mishima-Ishii flow regime map. Finally, the feature maps in the hidden layer of the model were extracted to visualize what gas-liquid distribution shapes the model focuses on to identify flow regimes.

The spreading and solidification of molten metal drops are essential physical processes that determine the mechanical properties of the products obtained with additive manufacturing, such as metal 3D printing and sintering. To obtain higher mechanical strength for the products, one needs to prevent the molten metal solid interface from forming gas pores and layers. We experimentally investigated the formation mechanism of layered solidification patterns by directly observing the impact dynamics with the molten tin drops on a sapphire substrate under reduced pressures of argon gas environment. We also quantitatively evaluate the relevant lengths of the air and the solidified layers using a laser microscope. As a result, we revealed that ambient pressure hardly affected the air layer depth, which was instead determined by heat conduction. We also found that the width of the solidified layer correlated to the rim diameter increasing monotonically as the drop spreads.

Predicting the dynamics of impacting drops is crucial in various industrial applications such as spray and inkjet technologies. Especially in painting technology, the entrapment of air bubbles greatly reduces the product quality. The formation of a thin air film in front of the moving contact line causes the entrapment of air bubbles in the spreading phase. To clarify the main forces acting at the gas-liquid interface that induce air bubbles entrapment, we investigate the contact line velocity of impacting drops by taking bottom view images using the total internal reflection (TIR) method. We used droplets of glycerol or glycerol-ethanol solutions and solid surfaces covered with either glycerol or silicone oil. Our results show the contact line velocity decreased rapidly at T_c when a thin air film was formed and settled down at a constant value V_const after the formation of the air film. Furthermore, we revealed that drop viscosity affects T_c and V_const. Moreover, we found that after T_c, the lubrication pressure of the air had the same order of magnitude as the viscous shear stress of the drop, which implied the importance of the air lubrication pressure.

Hybrid cavitation model is developed using both interface tracking model and homogeneous fluid model. The target of the analysis is cavitating turbulent flow around Clark-Y11.7% hydrofoil at angle of attack of 2 and 8 degrees, and the one-equation dynamic model is used as the subgrid-scale model. The end of the tracked sheet cavity interface is closed without interfering with the progress of the re-entrant jet. The separated flow and its low-pressure region behind the sheet cavity are simulated as the streamlines go around the sheet cavity, increasing lift coefficient. On the other hand, separated flow promotes the progression of the re-entrant jet, leading to underestimation of the area that is always covered by the sheet cavity. These results suggest that the reproducibility of cavity near the front of the re-entrant jet should be considered in addition to the streamlines to improve the accuracy of quantitative reproduction of lift characteristics.

Expansion valves are among key components of air conditioners. A one-dimensional model has been developed with the maximum mass flow. Temperature difference across the valve might cause heat transfer and affect inlet subcooling. Heat conduction coefficient was modified based on experimental data to obtain reliable discharge rate predictions.