Search

The unique samples fabricated using seed crystals with artificial grain boundaries (GBs) allowed a systematic investigation of the effect of the macroscopic GB structure (orientation, asymmetric angle, Σ values) on the GB growth direction.

Silicon quantum dot multilayer (Si-QDML) is a promising material for a light-absorber of all silicon tandem solar cells due to tunable bandgap energy in a wide range depending on the silicon quantum dot (Si-QD) size, which is possible to overcome the Shockley-Queisser limit. Since solar cell performance is degenerated by carrier recombination through dangling bonds (DBs) in Si-QDML, hydrogen termination of DBs is crucial. Hydrogen plasma treatment (HPT) is one of the methods to introduce hydrogen into Si-QDML. However, HPT has a large number of process parameters. In this study, we employed Bayesian optimization (BO) for the efficient survey of HPT process parameters. Photosensitivity (PS) was adopted as the indicator to be maximized in BO. PS (σp/σd) was calculated as the ratio of photo conductivity (σp) and dark conductivity (σd) of Si-QDML, which allowed the evaluation of important electrical characteristics in solar cells easily without fabricating process-intensive devices. 40-period layers for Si-QDML were prepared by plasma-enhanced chemical vapor deposition method and post-annealing onto quartz substrates. Ten samples were prepared by HPT under random conditions as initial data for BO. By repeating calculations and experiments, the PS was successfully improved from 22.7 to 347.2 with a small number of experiments. In addition, Si-QD solar cells were fabricated with optimized HPT process parameters; open-circuit voltage (VOC) and fill factor (FF) values of 689 mV and 0.67, respectively, were achieved. These values are the highest for this type of device, which were achieved through an unprecedented attempt to combine HPT and BO. These results prove that BO is effective in accelerating the optimization of practical process parameters in a multidimensional parameter space, even for novel indicators such as PS.

We report on our recent attempt to pioneer “multicrystalline informatics” through collaboration of experiments, theory, computation, and machine learning to establish universal guideline how we can obtain high-performance multicrystalline materials. We employ silicon as a model material, and develop various useful machine learning models. One example is a neural network to predict distribution of crystal orientations in a large-area sample from multiple optical images. Transfer learning of pre-trained image classifier could predict spatial distribution of probability of dislocations generation from photoluminescence images. Extracted regions with high probability of dislocations generation could be characterized by multiscale experiments as well as computation using artificial-neural-network interatomic potential to disclose physics behind. The obtained knowledge could be useful for process development of high-performance multicrystalline materials.

Three-dimensional distribution of arsenic (As) dopants at Σ3{111}, Σ9{221}, Σ9{114}, and Σ9{111}/{115} grain boundaries (GBs) in silicon (Si) is examined by correlative analytical methods using ato...

To design a time-dependent control recipe which can ensure consistently suitable growth conditions in an unsteady growth system with dynamic environmental changes, an adaptive control method based on high-speed machine learning prediction models was proposed and applied to the solution growth of SiC crystals. This approach comprised three parts, namely, a quasi-unsteady computational fluid dynamics (CFD) model for thermal and flow field simulation, machine learning models for approximating the simulation results and giving instant prediction, and an optimization algorithm for searching the optimal growth conditions. First, the evolution of the flow, temperature and carbon concentration fields over 50 h unsteady growth following an original recipe with fixed control parameters was analyzed by CFD simulation. Then, adaptive control was applied to design a time-dependent growth process with a 100-timestep sequence. The hybrid of machine learning models and CFD simulation accelerated the entire design and optimization process by 300 times, compared with CFD simulations alone. The adaptive control facilitated superior performance compared with the fixed recipe, where the single SiC crystal thickness increased by ∼30% and the growth interface was more uniform. Further, crucible dissolution and polycrystal precipitation were suppressed by ∼50%, enabling longer growth time and more stable growth. It is the first time that the importance of adaptive control during long-term SiC solution growth is discussed, and the method proposed in this study demonstrated the potential for real-time optimization in the future.

In the design of a crystal growth system, the ability to efficiently regulate intertwined geometrical parameters is crucial for its successful development and commercialization. However, the traditional experimental and computational methods consume tremendous amounts of time and resources. To address this problem, a machine learning approach was developed in this study to accelerate the geometry optimization process. It was found that the combination of machine learning with a genetic algorithm could generate various possible solutions through a global search at a relatively high speed, which lie outside the solution range of the experimental optimization methods that are currently used. By applying this technique, an optimal geometrical design was obtained for a 150 mm top-seed solution growth system, indicating that the proposed method represents an innovative and attractive strategy for the development of crystal growth systems with superior characteristics.

Multicrystalline Mg2Si crystal with a diameter of 15 mm was grown via vertical Bridgman method. To clarify the growth mechanism of the multicrystalline structure, the grain growth behavior of the crystal was analyzed. This was carried out through segmenting grains by mean shift clustering using the light intensity profile obtained from multiple optical reflection images of the wafers and stacking the segmented images through the growth direction. Further crystal orientation measurement revealed that a grain with a higher surface energy competitively expanded to the lateral direction during crystal growth. We speculated that the growth behavior occurred because the supercooling was high enough to show difference in each grain’s growth rate. This idea was supported by crystal growth simulation to show a tendency for the crystallization rate to increase toward the latter half growth stage, which is consistent with the assumption for crystal growth with high supercooling.

Rubrene is a promising and archetypal organic semiconductor owing to its high reported hole mobility. However, this high mobility only exists in its crystalline state, with orders of magnitude reduction in disordered films. Thus, as it pertains to thin film polycrystalline rubrene, it is important to understand structure and the presence of disordered regions at grain boundaries. Here, we use scanning transmission X-ray microscopy (STXM) to investigate polycrystalline rubrene thin films with either platelet or spherulite morphology. The STXM images allow us to distinguish and quantify the arrangement of the local structure in the crystal. The analysis suggests that the platelet film has more oriented molecules in the crystal than in the spherulite phase. Also, at spherulite grain boundaries, we reveal a high number of misaligned molecules compared to the smooth boundary in the platelet case, with grain boundary sinuosity of 0.045 ± 0.002 and 0.139 ± 0.002 μm for the platelet and spherulite cases, respectively, which help to explain the higher mobility in the former case.

We report on our attempt to perform the three-dimensional (3D) visualization of dislocation clusters in multicrystalline silicon (mc-Si) ingot by processing photoluminescence (PL) images and analysis of dislocation clusters in mc-Si. As-sliced wafers prepared using a high-performance (HP) mc-Si ingot were sequentially measured by PL imaging with intentional superposition of reflection. Then, various image processing techniques were applied to all the PL images to extract dark regions, which most likely correspond to dislocation clusters, as well as microstructures. By 3D reconstruction using a large quantity of 2D images, we could successfully visualize the generation, propagation and annihilation of dislocation clusters in HP mc-Si ingot. In addition, relationship between source region of dislocation clusters and crystal orientation were investigated by combining data scientific and experimental approaches. As a result, it was suggested that small angle grain boundaries with angular deviation of less than 10 degrees cause the generation of dislocation clusters.

We report on the effects of grain boundary (GB) structures on the carrier recombination velocity at GB ( vGB) in multicrystalline Si (mc-Si). The fabricated artificial GBs and an originally developed machine learning model allowed an investigation of the effect of three macroscopic parameters, misorientation angle α for Σ values, asymmetric angle β, and deviation angle θ from the ingot growth direction. Totally, 13 GBs were formed by directional solidification using multi-seeds with controlled crystal orientations. vGB was evaluated directly from photoluminescence intensity profiles across GBs using a pre-trained machine learning model, which allowed a quantitative and continuous evaluation along GBs. The evaluation results indicated that the impact of θ on vGB would be relatively large among the three macroscopic parameters. In addition, the results for the Σ5 and Σ13 GBs suggested that the minimum vGB would be related to the GB energy. These results were discussed in terms of the complexity of the local reconstruction of GB structures. The deviation would make a more complex reconstructed GB structure with local distortion, resulting in an increase in the electrical activity of GBs. The obtained knowledge will contribute to improving various polycrystalline materials through a comprehensive understanding of the relationship between GB structures and their properties.

We have developed machine learning methods for crystal growths of bulk and film Si. For the Czochralski growth of bulk Si, a neural network model was trained to predict interstitial oxygen concentration using log data of experimental conditions and monitored parameters. By utilizing this model, a real-time prediction system was developed wherein the crystal growth data are input into the model, and the Oi concentration at the current growth interface is calculated immediately. For the epitaxial growth of Si through the chemical vapor deposition method, we optimized process parameters using Bayesian optimization with adaptive constraints. The crystal growth rate was increased to be approximately twice as high as that under standard conditions, while satisfying the five quality parameter conditions.

Dislocation clusters that would degrade the electric property can be generated from a grain boundary (GB) neighboring a triple junction of GBs. The atomic plane of the GB is bent via the movement of the triple junction, supposedly due to Σ3{111} micro-twins intersecting the GB, and a number of dislocations would be generated nearby the bending corner. Bundles of dislocation arrays expanding nearly parallel to the growth direction and honeycombed dislocation networks lying on a {111} plane nearly normal to the growth direction can coexist, suggesting that multiple slip systems would be operated when the dislocations are tangled.

Analysis of nucleation sites in multicrystalline silicon (mc-Si) grown by directional solidification is required for further grain refinement to reduce dislocation density. In this study, Voronoi diagrams were utilized to analyze nucleation sites of mc-Si grown by single-layer Si beads (SLSB)-seeding method. The grain distribution at the bottom of the ingot was almost reproduced by the weighted Voronoi diagram with a relaxation method to optimize the positions of generating points of the diagram, which correspond to the nucleation sites, and the difference of nucleation timing of each crystal grain. Comparison of the generating points with the optical image indicated that the nucleation started at the remarkably deep portions of the nucleation layer. Further grain refinement by SLSB-seeding method could be achieved by suppressing the formation of the remarkably deep portions of the nucleation layer.

The mechanical properties and elastic moduli of wurtzite-structured nitrides (BN, AlN, GaN and InN) and AlGaN nitride alloy are evaluated by micro- and nano-indentation studies at room temperature (RT), and the physical processes are discussed. The hardness of BN, AlN, GaN and InN varies depending on the a-axis lattice constant a as an (n ∼ −6), whereas the Young’s, shear and bulk moduli of the nitrides vary as an (n ∼ −5). The properties are governed by the atomic bonding. A homology of indentation hardness scaled using the shear modulus and the magnitude of the Burgers vector for the nitrides is observed. The yield strength and stress-intensity factor of the nitrides at RT are presumed. The alloy-hardening effect in AlGaN alloy is weak compared with that in InGaN alloy.

We report on the application of Bayesian optimization (BO), which could accelerate the time-intensive process optimization of many parameters, to fabrication of the high-performance titanium oxide/silicon oxide/crystalline silicon passivating contact. The process contains pre-deposition treatment to form SiOy interlayer, atomic layer deposition (ALD) of TiOx, and hydrogen plasma treatment (HPT) as post-process. We attempted to optimize seven parameters for ALD and HPT by dealing with samples treated by three kinds of chemical solutions in the same batch. This permits to perform BO for each structure at the same time and determine the superior pre-deposition treatment. Consequently, carrier selectivity S10 estimated by independent measurements of the saturation current density and contact resistance was significantly improved by BO of only 12 cycles and 10 initial random experiments. These results certify that BO could efficiently provide experimental conditions in multidimensional parameter space although we need to consider the impact of the metallization process on the passivation performance.

We present a machine learning model to directly predict the carrier recombination velocity, vGB, at the grain boundary (GB) from the measured photoluminescence (PL) intensity profile by training it with numerical simulation results. As the training dataset, 1800 PL profiles were calculated with a combination of random values of four material properties—vGB, the GB inclination angle, and the carrier diffusion lengths in the grains on both sides of the GB. In addition, the measured noise was modeled artificially and applied to the simulated profiles. A neural network was constructed with the inputs of the PL profile and the outputs of the four properties. This served as the solver of the reverse problem of the computational simulation. The coefficient of determination and the root mean squared error of vlog, which is the common logarithm of vGB, for the test dataset were 0.97 and 0.245, respectively. This prediction error was sufficiently low for the practical estimation of vGB. Moreover, the calculation time was reduced by a factor of 198 000 compared to conventional numerical optimization of repeating the computational simulations. By utilizing this fast prediction method, continuous evaluation of vGB along a GB was demonstrated. The finding is expected to advance scientific investigation of the electrical properties of local defects.

We applied hydrogen plasma treatment (HPT) on a titanium oxide/silicon oxide/crystalline silicon heterostructure to improve the passivation performance for high-efficiency silicon heterojunction solar cells. To accelerate the time-intensive process optimization of many parameters, we applied Bayesian optimization (BO). Consequently, the optimization of six process parameters of HPT was achieved by BO of only 15 cycles and 10 initial random experiments. Furthermore, the effective carrier lifetime after HPT on the optimized experimental conditions became three times higher compared with that before HPT, which certifies that BO is useful for accelerating optimization of the practical process conditions in multidimensional parameter space.

We investigated the effect of the grain boundary (GB) character of multicrystalline Si (mc-Si) on the efficiency of external and internal gettering of impurities during phosphorus diffusion gettering (PDG). We utilized seed crystals with an artificially designed GB configuration to grow mc-Si ingots with different artificial GB characters. PDG combined with an originally developed multiple-cycle gettering technique at low temperature was introduced on intentionally Fe-contaminated mc-Si samples to enhance external and internal gettering. A significant positive PDG effect was observed after PDG combined with the multiple-cycle technique, as evidenced by the increase in lifetimes after PDG. A bright cloud-like photoluminescence signal around contaminated GBs was observed for artificial Σ5-GBs and tilt-GBs after PDG, suggesting the enhancement of the internal gettering efficiency by leaving a cleaner area around the GBs. This result suggests the importance of the control of crystal defect character as well as impurities in mc-Si ingots, which could strongly affect the PDG efficiency. Copyright © 2016 John Wiley & Sons, Ltd.

Non-coherent Σ3{111} grain boundaries (GBs) with a positive deviation in the tilt angle (θ 〈110〉 > 70.5°) exhibit a high recombination activity in high-performance multicrystalline silicon ingots. Most of the GB segments are composed of edge-type dislocations with the Burgers vector b of a/3〈111〉, unlike Lomer dislocations with b = a/2〈110〉 observed for negative deviations, arranged on coherent Σ3{111} GB segments. Stretched 〈110〉 reconstructed bonds along the tilt axis are introduced so as not to form dangling bonds, and large strains are generated around the dislocation cores. Oxygen and carbon atoms segregating due to the strains would induce the recombination activity.

We applied a Bayesian optimization (BO) to optimize the epitaxial growth process of Si thin films. The BO enables us to effectively explore the optimal film growth conditions considering several experimental parameters and their interactions. In this way we reduced the total number of experiments required in the optimization. The epitaxial growth rate was maximized while five quality parameters were maintained within an acceptable range. Additionally, we considered two practical issues: eliminating equipment errors and the time cost of the quality parameter evaluation. To overcome these issues, we adaptively conducted BO with different constraints for different situations. As a result of these optimizations, the crystal growth rate was increased to be approximately twice as high as that under standard conditions, while satisfying the five quality parameter conditions.

We developed a machine learning model to predict interstitial oxygen (Oi) concentration in a Czochralski-grown silicon crystal. A highly accurate prediction can be ensured by selecting the appropriate experimental parameters that represent the change in the furnace conditions. A neural network was trained using the dataset of 450 ingots, and its prediction error for the testing dataset was 4.2 × 1016 atoms cm−3. Finally, a real-time prediction system was developed wherein the crystal growth data are input into the model, and the Oi concentration at the current growth interface is calculated immediately.

Three-dimensional distribution of grain boundaries (GBs) and generation sources of dislocation clusters is examined in a cast-grown high-performance multicrystalline silicon ingot for commercial solar cells. A significant number of dislocations are generated nearby some triple junctions of random angle GBs, although it is believed that such non-coherent GBs would not induce large strain during the cast growth. This explosive generation of dislocations would take place when the triple junctions are interacted with multiple Σ3{111} GBs. A segment of the random angle GB connected with a pair of Σ3{111} GBs nearby the triple junction would act as a dislocation source.

Transmissions of dislocations across grown-in twin boundaries, i.e., Σ3 {111} boundaries, in multicrystalline silicon (mc-Si) were investigated by direct observations using the etch-pit technique and x-ray topography. Dislocations were freshly generated from a scratch by three-point bending at 800 °C. The generated screw dislocations whose Burgers vector was parallel to the twin boundaries stopped their motion at the boundaries and piled up there. This result reveals that the resistance stress of twin boundaries against screw dislocations is greater than ∼18 MPa. Alternatively, nonscrew-type dislocations were generated from the twin boundaries into the adjacent twinned grain under the stress greater than ∼15 MPa. The results are discussed in terms of cross-slip of dissociated dislocations for transmission across the twin boundaries and stress concentration by piled-up dislocations in the interactions of dislocations with grain boundaries developed in metals.

Heavily tin (Sn)-doped Si crystals in a concentration up to 4×1019 cm−3 were grown by the Czochralski method. Variation of Sn concentration in the crystals was well expressed by the Pfann equation using a segregation coefficient of k=0.016. From the occurrence of growth interface instability and the appearance of Sn precipitates in the grown crystals, the solubility limit of Sn was considered to be around 5×1019 cm−3. Interstitially dissolved oxygen Oi was presented at a concentration of 8–9×1017 cm−3 in the grown Sn-doped crystals. The FT-IR absorption peak relating to a Si–Oi–Si quasi-molecule at 1106 cm−1 showed preferential occupation of Oi at the bond-centered position of Si–Si. The Oi peak shifted to the lower wave number side with increasing Sn concentration in Si, implying expansion of the Si–Si bond.

Dislocations were introduced in wurtzite zinc oxide single crystals by Vickers indentations on { 2 1 ¯ 1 ¯ 0 } and { 10 1 ¯ 0 } surfaces at elevated temperatures, and their slip systems were determined by transmission electron microscopy combined with etch pit observations. The observed system for { 2 1 ¯ 1 ¯ 0 } indentations was { 10 1 ¯ 1 ¯ } 〈 1 ¯ 2 1 ¯ 0 〉 or (0001) 〈 1 ¯ 2 1 ¯ 0 〉 , while that for { 10 1 ¯ 0 } indentations was { 10 1 ¯ 0 } 〈 1 ¯ 2 1 ¯ 0 〉 . Activation of the slip systems was briefly discussed in terms of the Peierls stress and resolved shear stress acting on the slip systems.

Heavily indium (In)-doped Si crystals were grown by the Czochralski method under a consideration of the effects of co-doping of electrically neutral group-IV elements (C, Ge or Sn). The In concentration in In-doped Si increased with the amount of In charged into the crucible and reached 3.5×1017 cm−3. The carrier concentration was at most 6×1016 cm−3, limited by the low ionization ratio of ~20% of In. Co-doping of C and Ge effectively enhanced the In concentration while Sn did not, which was examined in terms of the atomistic size, lattice parameter change, mutual bonding energy and solubility of group-IV elements in Si. However, no sufficient increase in carrier concentrations was detected in Si by the co-doping, and formation of some clusters or complexes was suggested.

We propose a new growth method for mono-like silicon (Si): the suppression of multicrystallization using functional grain boundaries artificially formed by multiseed crystals. In our previous study, we demonstrated such suppression in an ingot 30 mm in diameter. In this paper, we grew mono-like Si ingots of 100 and 400 mm on a side. Functional grain boundaries successfully suppressed the increase in the area of multicrystalline grains nucleated on crucible side walls, which indicates a large volume of quasi-monocrystalline Si up to the top of the ingots. This enables a large increase in the yield of quasi-monocrystalline wafers in an ingot and would lead to a reduction in the cost of the solar cells.