Search

The paper proposes a model-based nonlinear model predictive control (NMPC) method for online control of crystal mean size and standard deviation in cooling crystallization process. Image analysis method using deep learning neural network and mathematical statistical method are performed to obtain the mean size and standard deviation of crystal population. The nonlinear prediction model for the NMPC is derived from the input and output data. The effectiveness of the proposed NMPC method is evaluated by the alum cooling crystallization experiments. Experimental results demonstrate the benefits of the proposed combination of image analysis and feedback control of the crystal mean size and standard deviation. The control performance of NMPC is superior to model-free path following control (PFC) method due to the prediction and optimization capabilities of NMPC. • Model-based nonlinear model predictive control for crystal mean size and standard deviation. • Model-free path following control for crystal mean size and standard deviation. • Image analysis method using deep learning to obtain pixel-level instance segmentation. • The effectiveness of the proposed control methods is evaluated by the alum cooling crystallization experiments. [ABSTRACT FROM AUTHOR]

As flash technologies face scaling issues at 32 nm and beyond, phase change memory (PCM) emerges as the best performing candidate for scaled non-volatile memories of next generation. To fabricate PCM device with high volume and low power consumption, the structure of PCM has evolved from planar structure to confined cells, which pose challenges for chemical mechanical polishing (CMP) of the phase change material Ge 2 Sb 2 Te 5 (GST).In this study, we discuss the properties, polishing mechanisms and slurry formulations for amorphous GST. Static etching characteristics and zeta potentials of GST powders are first explored to evaluate the chemical properties of GST. Then the polishing mechanisms of GST at pH 2 and pH 11 are discussed by means of static etching experiments, polishing performance, GST film hardness measurement, GST solubility and open circuit potential test. Finally, our efforts for GST slurry development are described. By a combination effect of inhibitors, in which one can form adsorption layer on GST by carbonyl O and the other one can further protect GST surface by forming hydrogen bond through multiple OH groups, we have overcome the corrosion issue of GST. Meanwhile, by the adsorption of the inhibitor layers on GST surface and extremely low removal rate of oxide film (<0.05 nm/min), we have achieved a self-stop GST CMP process with over-polish dishing <8 nm and oxide loss <6 nm for 60 nm GST vias. We have also resolved the issue of particle residue by using our self-formulated post-cleaning solution. [ABSTRACT FROM AUTHOR]

Abstract: We report the exploration of Ge2Sb2Te5 (GST) chemical mechanical polishing (CMP) mechanism. Static etching experiments of GST film were first conducted in two typical silica-based slurries (pH 2 and pH 11). To investigate the chemical nature of GST in different chemical environments, solubility of GST in slurries and also the zeta potentials of GST powders vs pH were measured. We further compared the polishing performance to the removal rate (RR), surface roughness, polishing selectivity of GST over oxide, and influence on the phase change property by using the two typical slurries. Then measurements were done for the hardness of GST films before and after polishing, particle size of the slurry during polishing, and also the open circuit potential (OCP) of GST in the two slurries. On the basis of the aforementioned results, we proposed a possible mechanism for GST CMP including reaction formulas, removal of GST as molecules instead of as a lump, surface hydration in alkaline region and electrochemical process, which was partially supported by residue analysis for 8″ patterned wafers using energy dispersive spectroscopy (EDS) under transmission electron microscope (TEM). [Copyright &y& Elsevier]

Highlights: [•] CMP process of 12in. GST blank wafer and pattern wafer with 60nm width damascene trenches in the acidic slurry. [•] Successfully EPD technique of GST CMP process using optical methods. [•] It has a little difference of EPD curve between blank wafer and pattern wafer. The EPD curve of pattern curve can divide into four stages. [•] Average dishing of GST are less than 3.5nm and the wafer-to-wafer variation is less than 8.0%. [•] A reproducible result is obtained using EPD technique although large variation of RR. [Copyright &y& Elsevier]