Predictive control of the plasma processes in the OLED display mass production referring to the discontinuity qualifying PI-VM

Abstract

Metal target dry etching process applied for the organic light emitting diode display manufacturing is hard to control without the generation of the defect particles. A large amount of the metal-halide by-prodcucts with the non-volatile physical nature are produced in the large area plasma-assisted process chamber. To achieve high-density plasma-based throughput, the inductively coupled plasma type dry etchers were adopted for large-area display manufacturing processes. However, this type of plasma source causes the ion flux-driven damages on the chamber inner walls near the RF power supplied antenna. Sputtered Al atoms from the ceramic parts or etching targets were redeposited onto the chamber inner walls after they form the metal-halide compounds. Redeposited by-prodcucts have very high binding energies to decompose. Undecomposed layers were stuck on the chamber inner wall and flaked off later to form the defect particles. To control this undesired phenomenon, decomposition reaction activated—and plasma locality controlled—two types of ISDs (In Situ Dry cleanings) were designed. A more appropriate type of ISD had selected referring to the developed PI-VM (Plasma Information based Virtual Metrology) model, which qualifies the start of mass production after the discontinuities of the process. The big data set of equipment engineering system and optical emission spectroscopy, accumulated during the mass production, were parameterized to the PI parameters and were applied to the PI-VM modeling. Management of the mass production with the designed ISD and PI-VM model could reduce the 25% of defect particle driven yield loss.

Review: Machine learning for advancing low-temperature plasma modeling and simulation

“…The reader is referred to corresponding comprehensive surveys by Brunton and Vinuesa et al 191,192 Concerning data-driven process control and MPC, it should be noted that the plasma information based VM methodology for LTP etching processes previously introduced may be adapted based on physical or data-driven model data. 19,20,193 Despite a possible discrepancy between models and reality, in particular systematic deviations may be straightforwardly taken into account in a data-driven approach by means of discrepancy learning. A learnable translation layer may suffice to capture and combine the relevant information by means of data-fusion.…”

2022 Review of Data-Driven Plasma Science

“…They have been used to predict various process performances such as etch rate, deposition rate, defect particles, etching profile, deposited thin film quality, and spatial uniformity of the processed results. They have been applied to the control and management of the OLED mass production lines last six years [577][578][579][580][581][582][583].…”

“…Developed PI-VM algorithms were applied to the mass production line of the OLED display manufacturing to solve four kinds of problems that occurred in the real field: The defect particle caused process fault prediction [578], root cause analysis of the high-aspect-ratio contact (HARC) etching process faults [579], the management of the mass production discontinuities with a proper application of the in-situ dry cleaning (ISD) [580], and the micro-uniformity problems in the process results [581]. These PI-VM models optimized for each issue have shown enough prediction accuracy to apply for the long-periodic mass production running.…”

Virtual metrology as an approach for product quality estimation in Industry 4.0: a systematic review and integrative conceptual framework

“…One means of supplementing an updating algorithm -if one has prior knowledge about the physical system -is to model concept drifts separately to decompose the problem (Iskandar and Moyne 2016) with, for instance, NNs (Chang et al 2006) or wavelet transform (Chang 2010). For example, maintenance -a common cause of concept drift -has been modelled multiple times (Lin, Hsu, and Yu 2014;Lynn, Ringwood, and MacGearailt 2012;Park et al 2020). This method is rarely sufficient as all of the causes of drift are generally unknown; it should be implemented as a complement to MW or JIT approaches.…”

“…One might think that with the emergence of ZDM, real-time consideration will soon become a hot topic. Regarding plasma-etching, it is noteworthy that the use of plasma information (PI) as an input has enabled online control and opened new opportunities for VM; PI is still very commonly used (Park et al 2020). Speaking of features, there is still a great deal of emphasis on dimensionality reduction with a preference for feature extractions.…”