Development of Plasma Information Based Virtual Metrology (PI-VM) for Plasma-assisted Processes

Abstract

moreover, they are utilized as plasma information based on the VM (PI-VM) model for plasma-assisted processes. PI-VM has shown a noticeably enhanced performance (R2 > 90%) for both C4F8-oxide etching and a-Si thin-film deposition processes compared to the statistical VM results (R2 ~50%). It therefore satisfies the required accuracy for industrial application. Cumulated variance analysis of the PCs extracted from the PI parameter-applied PCA validated that the performance of PI-VM was enhanced by the dominant inclusion of the PI parameter. In addition, it was implied that the deviation of the process results was due to the variation of the process plasma properties correlated with the condition of the process device. Therefore, PI-VM has the possibility of a process fault or deviation in the cause analysis because this methodology is based on the reaction mechanisms of the process plasma. This implies that, for the development of fault detection and classification (FDC) or advanced process control (APC) algorithms, information about the reaction mechanisms in the process plasma should be included in the model, and that the adoption of PI-VM can be an efficient method.

accordingly, the developed PI parameters are applied to the VM model according to the reaction mechanisms in the process plasma volume, sheath, and surface. The PI parameters are combined as one overall reaction rate representing the parameter defined as the r-factor

therefore, the performance of a statistical VM, without consideration of the process plasma information, cannot satisfy industrial requirements of prediction accuracy for ultra-fine processes. In the plasma-assisted process reactor, three kinds of reactions with different reaction mechanisms occur in the plasma volume, sheath, and substrate surface. In the plasma volume, generations of the reactive radical and ion species by inelastic processes of electron impact collision are dominant. This reaction rate is governed by the thermal equilibrium state of the plasma represented by the electron energy distribution function (EEDF) in the low-density low-temperature process plasmas. At the sheath region, transfer of generated reactive species with their own energy occurs. Ion species are accelerated by the electric field of the sheath and bombarded onto the reacting surface of the substrate. The electric field strength of the sheath is a sensitive function of the electronegativity and the EEDF. On the substrate surface, particles that arrived through the sheath experience sticking or diffusion to find the most stable site, and react with the solid material. This surface reaction is a function of the activation energy of the materials and surface temperature. These three types of reactions in process plasma determine the total process rates

thus, information about these reactions in the plasma should be efficiently included in the applied variables to develop a high-performance virtual metrology for plasma-assisted processes. The core variables include good information for the VM

therefore, plasma information (PI) parameters representing the reaction properties in the plasma volume, sheath, and surface are introduced in this dissertation. Shape factor b of the EEDF is introduced to monitor the variation of the volume reaction rate

ion bombarding energy monitoring parameter q is introduced to include information about the sheath property in the electronegative plasmas. Information about the passivation reaction on the substrate surface is additionally monitored as the optical signals of the passivating species. These PI parameters are monitored by using the optical emission spectroscopy (OES) sensor. OES is non-invasive to the process plasma and is an already widely used sensor for end point detection (EPD)

thus, this sensor has practicality to the application. The good information included variables

Plasma-assisted processes applied to semiconducting device- and display-manufacturing processes must be monitored by virtual metrology (VM) to maintain the process results and increase the throughput of the process. VM models that are widely used to predict the process results, such as etch rate and deposition rate, are based on statistical methods that analyse the correlation between sensing variables and performance variables. Principal component analysis (PCA) is a frequently used statistical method for VM modelling. The method identifies sensing variables to give highly correlated information for the prediction of performance variables and compounds them to the principal components (PCs), which are orthogonal to each other. However, in identified sensing variables obtained from the engineering equipment system (EES), and other sensors, such as for I-V signal, impedance data, and optical raw signals, the information about the reacting plasmas in the process reactor is not efficiently included. The inclusion of a good parameter, which efficiently contains information about the state of the process, is important for ensuring the accuracy of the VM