Particle and pattern discriminant freeze-cleaning method

Recently the reduction of the devise sizes causing the semiconductor processes more complicated and becoming more and more sensitive to the particle contamination [1]. Numerous studies have been carried out to improve the device yield with high particle removal efficiency (PRE) and Wet cleaning process along with megasonic is one of the well-established techniques used for particle removal in the semiconductor industry [1, 2]. However as the pattern size is reduced bellow 60 nm this method is not effective to improve the PRE. Recently, two-flow jet spray cleaning process became popular in semiconductor manufacturing processes due to its many advantages [2]. In this process, micron size water droplets are jetted onto a wafer surface at high velocity, which can remove the particles which are on the wafer surface. It consists of two separate nozzles for liquids and carrier gases which are concurrently delivered on the wafer surface during cleaning process. However, there is still lot of scope to improve the PRE during jet spray wet cleaning process by optimizing the process parametrs. Hence in the present study we focused on the jet spray wet cleaning parameters such as nozzle distance from the wafer and the flow rate of the carrier gas to improve the PRE with minimum pattern damage.

1. IntroductionIPA (iso-Propyl alcohol) has been widely used in semiconductor manufacturing for drying silicon wafer to mitigate water mark and keep the surface clean after wet cleaning process. Then, IPA process has been extended by combining with the other engineering techniques such as surface modification treatment to dry fragile nano-sized Si patterns [1]. As the scaling of advanced semiconductor devices continues, however, the pattern collapse is one of the most challenging issues for IPA drying process today. Capillary force model explains the pattern collapse in the earlier studies, where Laplace pressure originated from meniscus between patterns is a critical factor [2]. And the model has been well referred to and extensively studied in semiconductor industry [3]. Although, the model does not take dynamics of fluid into account, our experimental result suggests that drying rate is also critical, i.e., pattern collapse rate decreases as drying rate increases. This result is not explained by the reported models since the drying rate is not considered as a variable. In this report, a new model is suggested and verified by unique experiments. Our proposed model assumed that osmosis of metastable liquid to nano-size structures occurs from the triple line which is the air-liquid-solid interface, and the region of osmosis is less favorable for pattern collapse. The width of osmosis region would be determined by the drying rate. Also, based on this study, extendibility of IPA drying process to smaller device feature size for advanced node is discussed.2. ExperimentalWafer processing was conducted in spin process chamber on HVM equipment for drying experiment. IPA was dispensed on rotating wafer then dried at 0 or 1500 rpm. Nano-structured wafer used was with 34 nm-diameter of Si pillars, and aspect ratio (AR) of the pillar was 12:1. Pattern collapse was observed by top-view SEM, and collapse rate was calculated by automated image analysis with obtained SEM pictures. For osmosis observation at the triple line, FPM was prepared by mixing 49% HF, 30% H2O2 and H2O at volume ratio of 1:2:100. A droplet of FPM was dropped on the surface of Si pillar pattern wafer and observed by optical microscopy while the drop was drying out. After drying, pattern collapse of the pillar pattern was observed by X-SEM. To study effect of drying rate and to maximize it, IPA was dispensed on 10×12 mm2 coupon, then vaporized at 0 rpm with flash lamp annealing. AR of the pillar was 16:1. Observation of pattern collapse by top-view SEM was also conducted.3. Results and DiscussionIPA drying at 0 and 1500 rpm resulted in collapse rate of 91.1% and 46.9%, respectively, on the pillar AR = 12:1 (Fig. 1). IPA was dried out faster when wafer rotation speed was 1500 rpm. Thus, collapse rate decreases as drying speed increases. On osmosis observation, FPM droplet was sucked into between the pillars, causing osmosis at around droplet, and generated colored interference according to liquid film thickness. X-SEM inspection after the droplet drying indicated the most pattern collapse seemed to occur at the osmosis region where liquid thickness nearly equals to the pillar height. We assumed by increasing drying rate the osmosis region would be minimized and the pattern collapse would be suppressed. To verify this model, the following evaluation was conducted. With extremely high drying rate of IPA by flash lamp annealing, we obtained 0% of collapse rate even in the pillar AR = 16:1 (Fig. 2). This result indicates that there is some more process margin for the pattern collapse on IPA drying process.4. ConclusionIn the IPA drying process, collapse rate decreases as drying rate increases. Our proposed model assumed that osmosis region of metastable liquid to nano-size structures occurs from the triple line, and the pattern collapse happens there. That explains reasoning for lower collapse rate at higher drying speed is that osmosis is minimized when drying rate is maximized. Test result from maximized IPA drying rate by flash lamp annealing indicates that there is some more process margin for the pattern collapse. In our view, by this method IPA drying process might be extended to two more device nodes.

Pattern collapse phenomenon was first time observed in BEOL application with the integration of ultra low-k film scheme. With the dimension and pitch shrinkage, the pattern collapse defects are getting worse during wet clean process. In this study, the line collapse defects can be significantly reduced by adding surfactant solution to the rinse liquid. Moreover, higher aspect ratio (>4) will also deteriorate the collapse window. In addition, the kink or bowing trench profile will induce localized stress at the interface. Accordingly, optimization of both wet clean and dry etch process are the successful keys to solve line collapse issue toward future generation and beyond.

We propose freeze cleaning as a method of photomask cleaning in which particles small enough to be embedded in the region less than 100 nm from the substrate surface, where there is virtually no fluid flow, are selectively removed without causing pattern collapse. In freeze cleaning, a high particle removal efficiency is achieved by repeating the sequence of liquid (deionized water) being poured onto the substrate, freezing, and thawing (rinsing) multiple times. Based on the mechanism of particle removal, the timings at which the water freezes, ice growth, and freezing of the entire surface are important parameters that govern freeze cleaning performance. In contrast, when these timings were monitored during repeated processing, a maximum variation of about 16% was observed. The most significant cause of these fluctuations is attributed to the process performed in a system that is open to the atmosphere at room temperature, despite the use of cryogenic N2 at -120°C. Even with these timing fluctuations, by developing and applying an algorithm that monitors individual changes and automatically determines step switching using this monitor information, it is possible to construct a stable and highly efficient processing system without any tool modification.

Pattern collapse of nano-structures during the wet cleaning process is one of the main problems which leads to poor device yield. In general, aspect ratio (AR) is often used as the indicator for determining the likelihood of pattern collapse occurrence, because high aspect ratio structures tend to collapse more easily. However, pattern collapse is also influenced by flexural rigidity of the structures, material and shape. Therefore, AR lacks versatility in comparing different structures and material. We propose “γPC ” as the substitute parameter for aspect ratio. In this paper, we demonstrate with experimental data that γPC is more accurate than aspect ratio, and can be used to quantitatively determine the pattern collapse prevention performance of dry technologies.

Pattern collapse of nano-structures during the wet cleaning process is one of the main problems which leads to poor device yield. In general, aspect ratio (AR) is often used as the indicator for determining the likelihood of pattern collapse occurrence, because high aspect ratio structures tend to collapse more easily. However, pattern collapse is also influenced by flexural rigidity of the structures, material, and shape. Therefore, AR lacks versatility in comparing different structures and material. We propose “$\gamma_{PC}$” as the substitute parameter for aspect ratio. In this paper, we indicate with experimental data that $\gamma_{PC}$ is more accurate than aspect ratio, and can be used to quantitatively determine the pattern collapse prevention performance of dry technologies.}

Since the pattern collapse was reported for the first time it has been getting more critical issue in wet cleaning process along with semiconductor scaling. The model of the pattern collapse is that structures such as fins deform due to capillary force during drying step, then they contact and stick each other due to surface tension. One technique to prevent pattern collapse involves lowering the surface tension by modifying the surface’s silanol groups with functional groups[1]. The expected mechanism is that the adhesive force becomes smaller than the elastic force of the pattern due to the reduced surface tension. However, the details, including the dynamics, are still unclear. In particular, how functional groups affect fin contact from a molecular perspective remains unknown. Clarifying these issues is crucial because it will lead to a better understanding of the pattern collapse, subsequently enhancing semiconductor fabrication efficiency. Molecular dynamics (MD) simulation is a useful tool to elucidate these phenomena, as it can obtain physical properties and forces of the system from statistical thermodynamics and visualize nanoscale dynamics[2].In this study, we prepared models of fins modified with trimethyl silanol (TMS) and calculated the adhesive forces between them by using MD techniques. We utilized the MD solver Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) and the visualizer Open Visualization Tool (OVITO) [3, 4].The fin model utilized in these calculations is depicted in Figure 1. The size of the simulation box was 15.360×2.304×42.000 nm. Periodic boundary conditions were applied in the x and y directions, while specular boundary condition was applied in the z direction. This model primarily consisted of Si atoms. The height of fins was 30 nm, with the top 10 nm terminated by OH groups (912 in total) and the remaining terminated by H atoms. Here we introduced TMS substitutions for randomly selected OH groups. The numbers of substitutions were 0, 91, 228, and 365, resulting in 0%, 10%, 25%, and 40% TMS models, respectively. To reduce bias arising from the random placement of TMS groups, five TMS models were created for each level of TMS substitution. The first step in the calculation process involved placing water between the fins to simulate a wet condition and induce pattern collapse, followed by conducting a 2.0 ns MD calculation. As a result, in every model, the fins came into contact each other (Figure 1b). Subsequently, all water molecules were removed to simulate a dried condition, followed by a 2.0 ns MD calculation. In all models, under the dry condition, the fin contact established during the wet condition was kept (Figure 1c, d). The average adhesive forces between 1.0 ns and 2.0 ns of the MD trajectory after water removal were analyzed, revealing that the forces in the 10%, 25%, and 40% TMS models were less than 1/3 of those in the 0% TMS models. We interpret that, despite the persistence of pattern collapse in the dry condition, restoring the initially separated fins appears more feasible in TMS-modified surface models. The details will be reported in an oral presentation.[1] T. Koide et al., "Effect of Surface Energy Reduction for Nano-Structure Stiction," ECS Transactions, vol. 69, no. 8, p. 131, 2015, doi: 10.1149/06908.0131ecst.[2] R. Seki et al., "Insights into FinFET Structure Collapse: A Reactive Force Field-Based Molecular Dynamics Investigation," Solid State Phenomena, vol. 346, p. 123, 2023, doi: 10.4028/p-mUO0Oa.[3] A. P. Thompson et al., "LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales," Computer Physics Communications, vol. 271, p. 108171, 2022, doi: 10.1016/j.cpc.2021.108171.[4] A. Stukowski, "Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool," Modelling and Simulation in Materials Science and Engineering, vol. 18, no. 1, p. 015012, 2010, doi: 10.1088/0965-0393/18/1/015012.Figure 1: (a) The wet Fin model with inter-fin water molecules for MD simulations. The top 10 nm of the fins were terminated with OH groups, while the remaining were terminated with H atoms. Through the introduction of TMS substitutions for randomly selected OH groups, we prepared 10%, 25%, and 40% TMS models. (b) The snapshot of MD simulation after 2.0 ns from the initial configuration of (a). (c) The dry model prepared by removing water molecules from the configuration of (b). (d) The snapshot of MD simulation after 2.0 ns from the configuration of (c). Figure 1

Crystal structures in small Al-rich Fe particles were studied by powder X-ray diffraction and transmission electron microscopy. The specimens of the small particles were made by evaporating an Al alloy material containing 5.0-33.3 at.% Fe in Xe gas at a pressure of 3.3 × 103 Pa. The small particles were 70-150 nm in diameter. In the particles, f.c.c. Al-Fe solid solution, Al13Fe4, Al2Fe and AlFe structures were observed. The Al13Fe4 structure, which is abundant in all specimens, was found to have many lattice defects and twins, specifically planar defects lying on (100), (201) and (001) planes and twins on (100), (201) and (001) planes. Three kinds of planar defect and the (100) twin were directly observed by high-resolution electron microscopy. The high-resolution image of the (100) twin supports Black's model. The observed facts and the atomic position proposed by Black suggest a similarity between the atomic arrangement around the (100) plane and that around the (201) plane. Also, two types of...

Optical trapping is a technique for immobilizing and manipulating small objects in a gentle way using light, and it has been widely applied in trapping and manipulating small biological particles. Ashkin and co-workers first demonstrated optical tweezers using a single focused beam. The single beam trap can be described accurately using the perturbative gradient force formulation in the case of small Rayleigh regime particles. In the perturbative regime, the optical power required for trapping a particle scales as the inverse fourth power of the particle size. High optical powers can damage dielectric particles and cause heating. For instance, trapped latex spheres of 109 nm in diameter were destroyed by a 15 mW beam in 25 sec, which has serious implications for biological matter. A self-induced back-action (SIBA) optical trapping was proposed to trap 50 nm polystyrene spheres in the non-perturbative regime. In a non-perturbative regime, even a small particle with little permittivity contrast to the background can influence significantly the ambient electromagnetic field and induce a large optical force. As a particle enters an illuminated aperture, light transmission increases dramatically because of dielectric loading. If the particle attempts to leave the aperture, decreased transmission causes a change in momentum outwards from the hole and, by Newton's Third Law, results in a force on the particle inwards into the hole, trapping the particle. The light transmission can be monitored; hence, the trap can become a sensor. The SIBA trapping technique can be further improved by using a double-nanohole structure. The double-nanohole structure has been shown to give a strong local field enhancement. Between the two sharp tips of the double-nanohole, a small particle can cause a large change in optical transmission, thereby inducing a large optical force. As a result, smaller nanoparticles can be trapped, such as 12 nm silicate spheres and 3.4 nm hydrodynamic radius bovine serum albumin proteins. In this work, the experimental configuration used for nanoparticle trapping is outlined. First, we detail the assembly of the trapping setup which is based on a Thorlabs Optical Tweezer Kit. Next, we explain the nanofabrication procedure of the double-nanohole in a metal film, the fabrication of the microfluidic chamber and the sample preparation. Finally, we detail the data acquisition procedure and provide typical results for trapping 20 nm polystyrene nanospheres.

Optical trapping is a technique for immobilizing and manipulating small objects in a gentle way using light, and it has been widely applied in trapping and manipulating small biological particles. Ashkin and co-workers first demonstrated optical tweezers using a single focused beam1. The single beam trap can be described accurately using the perturbative gradient force formulation in the case of small Rayleigh regime particles1. In the perturbative regime, the optical power required for trapping a particle scales as the inverse fourth power of the particle size. High optical powers can damage dielectric particles and cause heating. For instance, trapped latex spheres of 109 nm in diameter were destroyed by a 15 mW beam in 25 sec1, which has serious implications for biological matter2,3. A self-induced back-action (SIBA) optical trapping was proposed to trap 50 nm polystyrene spheres in the non-perturbative regime4. In a non-perturbative regime, even a small particle with little permittivity contrast to the background can influence significantly the ambient electromagnetic field and induce a large optical force. As a particle enters an illuminated aperture, light transmission increases dramatically because of dielectric loading. If the particle attempts to leave the aperture, decreased transmission causes a change in momentum outwards from the hole and, by Newton's Third Law, results in a force on the particle inwards into the hole, trapping the particle. The light transmission can be monitored; hence, the trap can become a sensor. The SIBA trapping technique can be further improved by using a double-nanohole structure. The double-nanohole structure has been shown to give a strong local field enhancement5,6. Between the two sharp tips of the double-nanohole, a small particle can cause a large change in optical transmission, thereby inducing a large optical force. As a result, smaller nanoparticles can be trapped, such as 12 nm silicate spheres7 and 3.4 nm hydrodynamic radius bovine serum albumin proteins8. In this work, the experimental configuration used for nanoparticle trapping is outlined. First, we detail the assembly of the trapping setup which is based on a Thorlabs Optical Tweezer Kit. Next, we explain the nanofabrication procedure of the double-nanohole in a metal film, the fabrication of the microfluidic chamber and the sample preparation. Finally, we detail the data acquisition procedure and provide typical results for trapping 20 nm polystyrene nanospheres.

Nanoparticles of iron oxides were synthesized through the ablation of the bulk-targets of Fe, Fe3O4, and Fe2O3 in water by the irradiation of a Nd:YAG laser. Our samples prepared via the laser ablation method displayed small (ca. 1 nm in diameter) and large (over 5 nm in diameter) particles. The small particles were well-dispersed, whereas large particles were agglomerated. The FeO and Fe3O4 phases were formed in the synthesized powders irrespective of the kind of targets. The formation of the other phases, Fe and Fe2O3, changed depending on the the kind of targets used in the laser ablation. Poly(N-vinyl-2-pyrrolidone) (PVP) as a protective reagent was employed to disperse large particles in our samples, and accordingly, their dispersibility was improved as mole concentration of PVP increased.

In this work, a novel chemical-free technique is proposed to remove adsorbed particles of less than 100 nm from a substrate. More specifically, the small particles buried under the "stagnant layer" were removed using a process that relied on the force generated by volumetric expansion on rapid freezing of supercooled water. In the process, water penetrated into the small (narrow) spaces via the capillary force; it then flowed into the narrow interfacial gap between the substrate and particles, and lifted the particles off the substrate via the volumetric expansion force on ice formation. Because fine patterns have no such gaps, they are not damaged during this process. In other words, unlike conventional cleaning technologies, such as mega sonic cleaning and two-fluid jet cleaning, this cleaning process is able to specifically target the small particles while the fine patterns are unaffected.

Cement is a practical material for constructing the geological disposal system of radioactive wastes. However, such materials alter groundwater up to 13 in pH around the repository, changing the permeability of natural barrier. So far, the authors have examined the relation of permeability change with dissolution process by flowing a high pH solution (NaOH, 0.1 mM) into a bed packed with amorphous silica particles. Here, the particle diameters were adjusted to a size fraction of 74 to 149 μm by sieving. Its specific surface area was estimated as 350 m2/g by the BET method using nitrogen gas. The experimental results showed that the permeability did not immediately change although the soluble silicic acid continuously flowed out of the packed bed. This study proposes a new mathematical model considering the diffusion and dissolution processes in the inner pore of the particle. This model assumed that each packed particle (74 to 149μm in diameter) consists of the sphere-shaped aggregation of smaller particles (20 nm in diameter). OH− ions diffuse into the pore between such small particles, and simultaneously consumed by the reaction with small particles. The radius of the each packed particle (sphere-shaped aggregation of small particles) was defined by the length from the center of the aggregation to the region where the small particles still remains. Since the outer small particles more easily dissolve than inner small particles because of diffusion process of OH− ions, each packed particle gradually shrinks. The fundamental equations consist of a simple diffusion equation of spherical coordinates of OH− ions considering the reaction term, which is linked by the equation to describe the size change of small particles with time. Here, this model also considered a change (time and space) of the diffusion oefficient caused by the change of the porosity between small particles. Besides, the change of over-all permeability of the packed bed was evaluated by Kozeny-Carman equation and the calculated radii of packed particles. The dissolution rate constant already reported was used. The calculated result was able to well describe the experimental result, though there was no fitting parameter in the comparison with the experiment results. While the flow paths of underground cannot be simply simulated by a packed bed, this approach suggested that the dynamic behavior of permeability in a natural barrier depends also on non-uniformity of dissolution processes in inner pores (secondary pores) of minerals.

In this paper, the ultrafine particle removal using CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> gas cluster ion beam (GCIB) technology is investigated. The CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> GCIB is irradiated the sample at an angle of 0 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">°</sup> with respect to the surface normal. The higher particle removal efficiency can be achieved at the higher kinetic energy of the gas cluster, and the inside space of line and space pattern particles can be removed. We also suggested that the CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> GCIB process has the high particle removal uniformity without redeposition of the removed particles. It is possible to remove the ultrafine particle as small as 12 nm in diameter (which is required for 2014 by ITRS 2011). The pattern damage is not observed for 45 nm poly-Si pattern. Moreover, the molecular dynamics simulation is performed to investigate the mechanisms of the particle removal by GCIB irradiation.

Since the size of an atmospheric particle determines many of its effects, we conducted experiments to better understand their rate of growth. Seed particles composed of sulfuric acid and water were exposed to photolytically generated H2SO4 molecules and their change in size was monitored with a mobility particle system. H2SO4 production rates were held steady while the seed particle diameter was varied from 3 to 25 nm to explore how growth is affected by size. The growth rate of 25 nm diameter particles was about 50% less than that for 3 nm diameter particles. Gas-kinetic hard-sphere growth rates decline only 18% over this size range, but a decrease of 35-to-50% in growth over this range is expected according to theories that include the effects of a van der Waals interaction between gaseous H2SO4 and the small particles. The size-dependence of the measured growth rates, which does not require knowledge of the H2SO4 gas concentration, suggests that the attractive force between hydrated H2SO4 and small sulfuric acid particles leads to a significant enhancement of the collision rate; this force depends strongly on particle size below 10 nm in diameter. Recent calculations based on a central field approximation for the van der Waals interaction are consistent with the measurements, although empirical enhancement factors better explain the data for some conditions. Nucleation experiments were also performed with H2SO4 detection, and simulations of these nucleation experiments required similar van der Waals enhancements to secure agreement between measured H2SO4 vapor and the size of the nucleated particles.