Search

Chemical mechanical planarization (CMP) is a complex and high-accuracy polishing process that creates a smooth and planar material surface. One of the key challenges of CMP is to predict the material removal rate (MRR) accurately. With the development of artificial intelligence techniques, numerous data-driven models have been developed to predict the MRR in the CMP process. However, these methods are not capable of considering surface topography in MRR predictions because it is difficult to observe and measure the surface topography. To address this issue, we propose a graphical model and a conditional variational autoencoder to extract the features of surface topography in the CMP process. Moreover, process variables and the extracted features of surface topography are fed into an ensemble learning-based predictive model to predict the MRR. Experimental results have shown that the proposed method can predict the MRR accurately with a root mean squared error of 6.12 nm/min, and it outperforms physics-based machine learning and data-driven methods. [ABSTRACT FROM AUTHOR]

With the reduction of technology nodes to less than 10 nm, ruthenium (Ru) has emerged as a viable replacement for copper (Cu) for back-end-of-the-line (BEOL) interconnects. Good surface uniformity while maintaining an appropriate thickness of Ru is required during the fabrication process, which can be achieved by employing the chemical mechanical planarization (CMP) process. However, a crucial step in the CMP process is to attain reasonable removal rates of high-mechanical-strength and chemically inert metals such as Ru. Hence, this study investigates a competent CMP slurry comprising potassium iodate (KIO3) as an oxidizer and fumed silica as an abrasive for polishing Ru. The polishing results show that adding KIO3 to fumed silica modifies the silica particles, thereby enhancing the removal rates of the metal. thermogravimetric analysis (TGA), Fourier transform infrared spectroscopy (FTIR) and ultraviolet-visible spectroscopy (UV-Vis) characterization confirm the modification of the silica particles by KIO3 as an oxidizer. The studies reveal that the abrasive concentration usage was reduced to 60% due to silica modification. The surface morphology of the treated metal surface was evaluated using field emission scanning electron microscopy (FESEM). No adsorbed contamination or pitting is observed on the metal surface. The inhibition effect of 1,2,3-benzotriazole (BTA) was investigated in detail by performing potentiodynamic polarization experiments. The nature of the metal dissolution in KIO3 was investigated by examining the effect of solution temperature on the etch rates. The thermodynamic process of the system was found to be endothermic in nature, and the dissolution followed an associative mechanism. The results reported in this paper reveal that the proposed slurry can be used to significantly improve Ru polishing efficiency and provide desired selectivity without compromising the surface quality of the metal. [ABSTRACT FROM AUTHOR]

With the advent of the 5G information age, the production and processing of computer chips have attracted people's attention. In order to adapt to the rapid development in the field of chip manufacturing, higher requirements are placed on chemical mechanical planarization (CMP). Due to the porous structure of the polishing pad surface, the polishing pad is prone to clogging and its performance gradually decreases during the polishing process. Therefore, the use of diamond disc conditioner to dress the polishing pad in the CMP process is an indispensable part. In this paper, a new method of measuring the conditioner disc is proposed, the wax is melted and covered on the disc, and after a period of time, it solidifies on the surface of the dressing disc to obtain the protrusion height of the grains; this method has more advantages than measuring with an optical microscope because of the error caused by the diffuse reflection of the microscope. For diamond conditioners used in the experiments, the motion trajectory of abrasive grains is simulated, which is the basis on choosing the experimental method, and the material removal rate model of the polishing pad is established. In addition, after polishing the silicon wafer with a polishing pad for 2 h, the diamond conditioners with orderly distributed abrasive grains are used to dress the pad, the discs can remove the blocking material on the surface of the polishing pad after dressing 30 min, and the dressing effect and efficiency of the diamond disc with grain size of 100 μm are more obvious. The surface roughness and the liquid contact angle of the polishing pad before and after dressing are measured; after a period of conditioning, they can be restored to a state similar to that of the new polishing pad. General scheme of the research [ABSTRACT FROM AUTHOR]

The material removal rate (MRR) plays a critical role in the chemical mechanical planarization (CMP) process in the semiconductor industry. Many physics-based and data-driven approaches have been proposed to-date to predict the MRR. Nevertheless, most of them neglect the underlying equipment structure containing essential interaction mechanisms among different components. To fill the gap, this paper proposes a novel hypergraph convolution network (HGCN) based approach for predicting MRR in the CMP process. The main contributions include: (1) a generic hypergraph model to represent the interrelationships of complex equipment; and (2) a temporal-based prediction approach to learn the complex data correlation and high-order representation based on the hypergraph. To validate the effectiveness of the proposed approach, a case study is conducted by comparing with other cutting-edge models, of which it outperforms in several metrics. It is envisioned that this research can also bring insightful knowledge to similar scenarios in the manufacturing process. [ABSTRACT FROM AUTHOR]

Achieving high processing quality for chemical mechanical planarization (CMP) in semiconductor manufacturing is difficult due to the distinct process variations associated with this method, such as drift and shift. Run-to-run control aims to maintain the targeted process quality by reducing the effect of process variations. The goal of controller learning is to infer an underlying output–input reverse mapping based on input–output samples considering the process variations. Existing controllers learn reverse mapping by minimizing the total mapping error for sample data. However, this approach often fails to generate inputs for unseen target outputs because conditional input distributions on target outputs are not captured in the learning. In this study, we propose a controller based on a least squares generative adversarial network (LSGAN) that can capture the input distributions. GANs are deep-learning architectures composed of two neural nets: a generator and a discriminator. In the proposed model, the generator attempts to produce fake input distributions that are similar to the real input distributions considering the process variation features extracted using convolutional layers, while the discriminator attempts to detect the fake distributions. Competition in this game drives both networks to improve their performance until the generated input distributions are indistinguishable from the real distributions. An experiment using the data obtained from a work-site CMP tool verified that the proposed model outperformed the comparison models in terms of control accuracy and computation time. [ABSTRACT FROM AUTHOR]

The prediction of average Material Removal Rate (MRR) in Chemical Mechanical Planarization (CMP) process is regarded as a crucial research objective of Virtual Metrology (VM) for semiconductor manufacturing. In this paper, a novel VM model is proposed to predict MRR in CMP process based on the integration of Gaussian Process Regression (GPR) with a reference-based strategy. The proposed method estimates the similarity of the changing trends of the sensor traces using Maximum Mean Discrepancy (MMD) as a metric to extract reliable references for further prediction. The temporal information is also combined in the strategy by filtering the data samples by timestamps. Afterwards, GPR is used to fuse the reference data samples to predict the metrology with a confidence interval. Compared with the benchmarks in the recent literature, the proposed method, referred as MMD-GPR model, gives better prediction performance than ensemble learning methods and equivalent accuracy compared with Deep Neural Networks (DNN) but with a more efficient framework. [ABSTRACT FROM AUTHOR]

The kinetics of dissolution of Co in alkaline glycine solution under anodic conditions was investigated. Potentiodynamic polarization and electrochemical impedance spectroscopic (EIS) data of Co dissolving in 0.1 M glycine solution at pH 10 were acquired. At low overpotentials, in the active region, the polarization current increases with potential, whereas at high overpotentials, in the passive region, it decreases with potential. Co surface exposed to the alkaline glycine solution was characterized using scanning electron microscopy and x-ray photoelectron spectroscopy, and based on the binding energies of the deconvoluted peaks, oxides, hydroxides, and Co-glycine complexes were identified as the likely species present on the surface. Impedance data was subjected to electrical equivalent circuit analysis, and a circuit with three Maxwell elements was found to model the faradaic impedance. Polarization and EIS data were also analyzed in the framework of mechanistic analysis, and a mechanism with four adsorbed intermediates is proposed. The proposed model captures the salient features of the polarization and EIS results. [ABSTRACT FROM AUTHOR]

The chemical mechanical planarization (CMP) performance of the Si-face and C-face 6H-SiC substrates were compared using potassium permanganate (KMnO4) as an oxidizer and the alumina (Al2O3) nanoparticles as the abrasive particles over a pH range from 2 to 10. The results indicate that there was a significant difference in the CMP performance between Si-face and C-face 6H-SiC substrates, indicating that the CMP process was sensitive to the surface polarity of the 6H-SiC substrates. A higher material removal rate (MRR) was obtained during the CMP of C-face 6H-SiC substrates, as compared to that of Si-face 6H-SiC substrates. The maximum MRR of the C-face 6H-SiC substrates was reached to 6412 nm/h with an average surface roughness Ra of 0.54 nm when using the slurries containing 2 wt% alumina nanoparticles, 0.05 M KMnO4 at pH 2. While the Si-face 6H-SiC substrates was 1554 nm/h with an average surface roughness Ra of 0.53 nm. Furthermore, the polishing mechanism was also discussed based on the MRR, atomic force microscopy (AFM) and X-ray photoelectron spectroscopy (XPS) analysis of the dipped and polished 6H-SiC surfaces. [ABSTRACT FROM AUTHOR]

Technologies in semiconductor industry have been developed into a three-dimensional multilayer wiring for high integration of devices. Chemical mechanical planarization (CMP) process is one of the key technologies for achieving multilayer wiring, which enables global planarization. In addition, highly integrated devices can be realized by increasing the depth of focus in the photolithography process. However, in the inter-layer dielectric (ILD) CMP of the transistor, the uppermost oxide layer has the step due to the arrangement of the devices. The ideal material removal mechanism is to gradually remove materials from the top of the step height which allows for global planarization. However, in the CMP of the patterned wafers, simultaneous polishing of the upper and lower layers occurs when the step height reaches a certain height. This means that the polishing is strongly dependent on the structural characteristics of the pattern. Especially, the difference in the material removal rate depending on the pattern density acts as a constraint in terms of device layout. Therefore, it is essential to develop an accurate prediction model of material removal rate as a function of pattern density, size and arrangement. This study aims to define the mathematical planarization model according to contact mode between a polishing pad and patterned wafer. Considering that the real contact area between the actual polishing pad and the wafer is about 1 %, the mathematical model is derived based on the microscopic deformation of the pad asperities, not the macroscopic deformation of the bulk pad. Finally, we describe the verification between the theoretical material removal rate model and step height reduction and the actual CMP results. The root mean square error of the upper layer material rate, the lower material removal rate, and the step height reduction were 24.59 nm/min, 22.03 nm/min and 22.6 nm, respectively. Compared with the previous studies, the new model of this study improved the error by up to 50.9 %. [ABSTRACT FROM AUTHOR]

In semiconductor manufacturing, the chemical mechanical planarization (CMP) process produces higher thickness variability in the edge area of the wafer than that in the center area due to the characteristics of the polishing operation. To address this problem, advanced CMP equipment includes a function that controls the removal rate of each area. However, to take full advantage of this capability, effective advanced process control systems must be implemented with a virtual metrology (VM) model. However, the prediction performance of the VM model often deteriorates due to process drift. Here, we present a deep learning-based VM model that demonstrates high performance in the presence of nonlinear process drift. The proposed model combines a recurrent neural network and a convolutional neural network to extract time-dependent and time-independent features. A two-stage model training method is proposed that alternately updates the weights of each network to improve prediction performance. In the experiments using on-site CMP process data, the performance of the deep learning models exceeded that of standard machine learning models. The proposed model showed an 8.48% decrease in process variability relative to the best-performing machine learning model, which was elastic nets. [ABSTRACT FROM AUTHOR]

Soda Lime glass continues to be a common variety of glass used in majority of applications. Thus for various applications to be involved in a smooth surface is the demand in the industry for commercial usage. The polishing of glass using Chemical Mechanical Polishing continues to be a major challenge in polishing of glass. The Soda lime glass used was polished considering Slurry Flow Rate, Speed and Down Pressure and Process time as the major influencing parameters and was examined for Surface Roughness and Material Removal Rate. Topographical studies were done on the polished surfaces using Atomic Force Microscopy (AFM) to compare the surface peaks. The Taguchi's Design of Experiments was implemented to obtain the best combination of parameters. Analysis of Variance was also done to verify the significant of the combination of the parameters. [ABSTRACT FROM AUTHOR]

The role of non-ionic surfactant concentration in the planarization performance of copper wafer is investigated. The basic components of the slurry contain H2O2, a mixture of benzotriazole (BTA) and ammonium dodecyl sulfate (ADS), FA/O II (macromolecular organic complexing agent), and silica. The static performance of the alkaline slurry including the surface tension, viscosity, particle size distribution and zeta potentials is analyzed. The effect of AEO on material removal rate (MRR) profiles, within-wafer non-uniformity (WIWNU) and surface roughness is also discussed. The AEO in combination with a small amount of BTA and ADS has a large effect on the stability of abrasives, MRR, WIWNU and surface roughness. A new MRR model is proposed to elucidate the polishing mechanism, revealing that the inhibition properties of AEO can be characterized through the corrosion inhibition efficiency and adsorption behavior. An optimal choice of the AEO concentration can achieve a good stability of the slurry, high MRR, low WIWNU and planar surface. The surfactant concentration ranging from 0.1 wt% to 0.5 wt% would contribute to acquiring a good surface planarization performance. Therefore, this work provides a comprehensive understanding of the effect of AEO concentration on the surface planarity of copper CMP in alkaline slurries. [ABSTRACT FROM AUTHOR]

During the Chemical mechanical planarization (CMP), the pad conditioning process can affect the pad surface characteristics. Among many CMP process parameters, the improper applied load on the conditioner arm may have adverse effects on the polyurethane pad. In this work, we evaluated the pad surface properties under the various conditioner arm applied during pad conditioning process. The conditioning pads were evaluated for surface topography, surface roughness parameters such as Rt and Rvk and Material removal rate (MRR) and within-wafer non-uniformity after wafer polishing. We observed that, the pad asperities were collapsed in the direction of conditioner rotation and blocks the pad pores applied conditioner load. The Rvk value and MRR were founded to be in relation with 4 > 1 > 7 kgF conditioner load. Hence, this study shows that, 4 kgF applied load by conditioner is most suitable for the pad conditioning during CMP. [ABSTRACT FROM AUTHOR]

Polystyrene (PS)-core silica-shell composite particles with various core sizes were synthesized and applied in chemical mechanical planarization (CMP). The structures of the obtained PS/SiO core/shell particles (CSPs) were characterized by field-emission scanning electron microscopy (FESEM), transmission electron microscopy (TEM), and dynamic light scattering (DLS). The effects of the PS core size on oxide CMP behaviors were investigated by atomic force microscopy (AFM) based on the surface profilogram, roughness, and material removal rate (MRR). The roughness of the substrate surface slightly increased with increasing core size, while the MRR obviously increased. The lowest root-mean-square roughness value (0.198 nm) was obtained for the PS/SiO CSPs with the smallest cores (216 nm), while the highest MRR (148.9 nm/min) was observed for the CSPs with the largest cores (552 nm). In addition, these polishing results were rationalized according to the indentation-based mechanism, incorporating the total number of active abrasives and the particle indentation depth into the wafer surface. [ABSTRACT FROM AUTHOR]

A diamond disk is indispensable to dressing chemical mechanical planarization (CMP) polishing pads. Current CMP diamond disks contain numerous diamond grits (approximately 10,000-25,000) on the flat surface of a metal substrate. Although a diamond disk has a regular distribution of grits on its surface, their tips are not of uniform height. Hence, only a small proportion of the grits can penetrate the flexible pad surface, and a large number of grits remain unused on the disk. In this work, we developed a novel technique to find and remove tall diamond grits on a diamond disk to improve the leveling of the diamond tips and increase the number of working grits. We examined an original diamond disk and disks in which three, five, and seven tall diamond grits were removed. We then evaluated the surface characteristics and pad dressing rates (PDRs) as well as the removal rate for a dielectric oxide film dressed with these diamond disks after the removal of the tall grits. We compared the results with those obtained for the original diamond disk. The removal of the tall diamond grits facilitated diamond leveling and helped increase the number of working grits. The PDR decreased as the number of tall grits removed increased. The experimental results revealed that, after the removal of the tall grits, the diamond disk exhibited better non-uniformity and a lower pad consumption rate without damaging the local circuitry. In addition, the lifetime of the pad and disk was prolonged, which should lower the cost of the CMP process. [ABSTRACT FROM AUTHOR]

Hydride vapor-phase epitaxy (HVPE) is a high-growth-rate, cost-effective means to grow epitaxial semiconductor material. Thick HVPE-based metamorphic buffer layers (MBLs) can serve as 'pseudosubstrates' with controllable lattice parameter. In our structures, the indium content in InGaAs is gradually increased from zero to the final composition corresponding to the desired lattice constant, and then a thick (∼10 μm) constant-composition capping layer is grown. This thick capping layer promotes maximum strain relaxation while permitting use of polishing procedures to achieve surface planarity. Lattice-mismatched growth of MBLs invariably results in rough, cross-hatched surface morphology exhibiting up to 200 nm peak-to-valley roughness. This roughness can be eliminated by chemical mechanical planarization, thus creating a suitable surface for subsequent regrowth. Polishing of InGaAs is complicated by the sensitivity of the surface layer to the polishing parameters, particularly the applied pressure. Polishing at high applied pressure (12 psi) results in the formation of circular asperities hundreds of nanometers high and tens of microns in diameter. When lower applied pressure (4 psi) was used, the cross-hatching height of MBLs was lowered from 200 nm to <10 nm over a 350 μm lateral scale. The successfully planarized InGaAs MBLs were used as a substrate for a superlattice (SL) structure such as that used in quantum cascade lasers. Use of planarization before regrowth of the SL resulted in a reduction of the high-resolution x-ray diffraction peak full-width at half-maximum from 389″ to 159″. [ABSTRACT FROM AUTHOR]

The effect of wafer size on the removal rate and profile during chemical mechanical polishing was investigated with two representative thin films: SiO 2 and Cu, which were used as the dielectric and interconnection, respectively. Experiments were conducted using SiO 2 and Cu blanket wafers with 100, 200, and 300 mm diameters, while the results for 450 mm diameter wafers were estimated using geometric calculation. The experimental results showed that the heat generated by an increase in the wafer size affects the removal rate and the Within-Wafer Non-uniformity (WIWNU). In particular, the polishing temperature is one of the most important factors affecting the removal rate and profile in the Cu CMP process. An optimum slurry flow rate may exist for each wafer size, which should be carefully considered, particularly when using 450 mm diameter wafers in the semiconductor industry. [ABSTRACT FROM AUTHOR]

In this paper, the effect of slurry injection position on removal rate and nonuniformity was experimentally investigated on a 300-mm polisher developed by the authors, and the effect of slurry injection position on slurry fluid pressure distribution during polishing was also measured by a novel in situ fluid pressure measurement system integrated in the polisher. In situ pad conditioning was employed in all experiments. The results showed that removal rate, nonuniformity, and fluid pressure distribution were influenced dramatically by the slurry injection position. The slurry injection position determines the slurry distribution, and therefore, the removal rate profile and fluid pressure distribution were affected. [ABSTRACT FROM AUTHOR]

In this work, a new feature-scale model is proposed for investigating the interaction between the wafer pattern and individual pad asperities in the process of chemical mechanical planarization (CMP). Based on the contact mechanics equation and the modified Greenwood-Williamson (GW) model which captures the evolution of feature curvature and the modification of the pad asperity height distribution, the discrete convolution and fast Fourier transform (DC-FFT) technique is adopted and combined with the Picard iteration method to calculate the direct contact pressure distribution between the wafer surface and the polishing pad. The computed pressure is then used to determine the local removal rate of the underlying patterns and predict the evolution of the wafer surface profile. Furthermore, the method is extended to capture the metal dishing as the feature size changes. It is shown that the present model can avoid the false simulated results produced by directly applying the original GW model for CMP when the feature size approaches zero. Otherwise, the calculated surface profile and dishing values of pattern geometries are in good agreement with the experimental data. Therefore, this model can not only be used to simulate the evolution of the wafer surface for global planarization at lower technology nodes, but can also be applied to provide some basic design rules for improving the process parameters and reducing the time and cost for developing new architectures. [ABSTRACT FROM AUTHOR]

Chemical mechanical planarization (CMP) is the major manufacturing step used to planarize semiconductor wafers and obtain mirror surface finish. In CMP, diamond disk pad conditioning is traditionally employed to restore pad planarity and surface roughness. The conditioning tool typically consists of a metal disk with one side embedded with protruding diamond grits (abrasives). The conditioner design has a significant effect on the pad conditioning process and hence the wafer planarization process. This paper proposes the application of engineering optimization methods such genetic algorithm to the conditioner design problem for the first time. A new metric to evaluate conditioning performance based on the conditioning density generated by a specific conditioner design is developed. The metric is applied in a genetic algorithm to optimize conditioner design parameters (including geometric arrangement of diamonds, grit density and disk size). The model searches for the design parameters that produce a desired CMP pad surface texture. Results show that the model can effectively serve as a platform to evaluate and tune conditioner design for different applications in CMP. [ABSTRACT FROM AUTHOR]

The surface planarization in the chemical mechanical planarization is mainly realized by the mechanical effect which depends on the micro-tribology behavior of nanoparticles. The material removal rate is decreased and micro-scratch is generated in the wafer surface due to the agglomeration of particles. The slurry particles hold definite energy level and microstructure which makes the classical materials processing theory unable to give reasonable explanation about this process. Large-scale classical molecular dynamic simulation of interaction among nanoparticles and solid surface has been carried out to investigate the evolvement mechanism of particle cluster and surface planarization. It is showed that the particle aggregation behavior is a size-dependent process because of the high ratio of surface atoms to bulk atoms which leads to the active reactivity and catalytic activity of nanoparticles. The substrate materials (nanoparticle) become transforming from low-energy stable system to high-energy unstable system after the two-body collision which is favorable for aggregation process. There exists obvious geometry boundary between the particle elements in the large aggregates and only the dendritic aggregates formed. The larger particles have definite microstructure and additional internal degree of freedom which can be used to take up, store, and transfer energy, thus lower the activity of chemical reaction which weaken the particle aggregation process. [ABSTRACT FROM AUTHOR]

Here we use two-dimensional models of fluid film lubrication and contact mechanics to calculate the contact stress and fluid (i.e., slurry) pressure distributions on the wafer-pad interface in chemical mechanical planarization (CMP). In particular, the effective rigidity of the wafer (determined by the wafer carrier structure), the retaining-ring width and its back pressure are taken to be the design parameters. The purpose is to study the synergetic effects of such parameters on the contact stress uniformity, which directly affects the spatial uniformity of the material removal rate on the wafer surface. Our numerical results indicate that, for a given wafer rigidity, one may choose the retaining-ring width and back pressure to minimize the contact stress non-uniformity (NU). Also, the resulting minimum NU decreases with the effective wafer rigidity, suggesting that it is beneficial to use a soft (e.g., floating-type) wafer carrier. Moreover, for a soft wafer carrier, it is demonstrated that using a multi-zone wafer-back pressure profile is even more effective in reducing NU. [ABSTRACT FROM AUTHOR]

The objective of this work is to tackle the challenges of monitoring and detecting subtle process changes in chemical mechanical planarization (CMP), an ultraprecision manufacturing process. Monitoring ultraprecision processes is usually of difficulty due to their innate complexity and low signal-to-noise ratio in sensor signals. Especially for subtle signal variations during small process changes, the conventional statistical process control charts could fail to detect such process anomalies from the sensor signals in the time domain. In this paper, frequency spectra representation of the microelectromechanical systems (MEMS) vibration sensor signals during subtle process changes is investigated, and the signal patterns uncovered by frequency spectra are utilized to formulate a machine learning–driven in situ process monitoring approach to detect process anomalies in CMP. The proposed approach overcomes the obstacles of differentiating subtle signal changes by transforming them into the frequency domain with Fourier transform and Hilbert-Huang transform and classifying the resulted frequency spectra with random forest. Based on frequency analysis, it can unveil the differences in the signals obscured in the time domain and suppress the high-frequency noise. Consequently, the presented machine learning–driven in situ process monitoring approach detects process anomalies by differentiating the deviated frequency spectra with machine learning. It is validated on our experimental CMP testbed for anomaly detection, and outperforms three benchmark statistical process control charts. For instance, it detects a slurry shutoff anomaly in CMP about ten times faster than the benchmark methods. [ABSTRACT FROM AUTHOR]