AVS 64th International Symposium & Exhibition | |
Plasma Science and Technology Division | Wednesday Sessions |
Session PS+SS+TF-WeA |
Session: | Plasma Deposition |
Presenter: | Hyun-Joon Roh, Seoul National University, Republic of Korea |
Authors: | H.-J. Roh, Seoul National University, Republic of Korea S. Ryu, Seoul National University, Republic of Korea Y. Jang, Seoul National University, Republic of Korea N.-K. Kim, Seoul National University, Republic of Korea Y. Jin, Seoul National University, Republic of Korea G.-H. Kim, Seoul National University, Republic of Korea |
Correspondent: | Click to Email |
Advanced process control (APC) is required to assure the quality and throughput of plasma-assisted process. For this purpose, the process result of all wafers should be measured. However, direct metrology can measure only 1~3 wafers within a lot due to slow time response. To improve the speed of metrology, virtual metrology (VM) is alternatively adopted to support APC. VM can predict the process results close to real-time, since it predicts the process results by using statistical methods based on equipment engineering systems (EES) and sensor variables. However, previously developed VMs face the degradation of prediction accuracy as the chamber wall condition drifts in long-term process. This robustness issue is originated from that the used input variables of VM cannot recognize the drift of chamber wall condition. To enhance the robustness even in a process with severe drift of chamber wall condition, we propose PI-VM that uses plasma information (PI) as input variables of statistical methods. Experimental application of PI-VM is performed to predict the nitride film thickness in multi-layer plasma-enhanced chemical vapor deposition (PECVD) for 3D NAND fabrication which has a severe drift of chamber wall condition. PI variables are composed of the chamber wall condition (PIWall) and property of bulk plasma (PIPlasma) considering plasma-surface interaction. Each PI variable is decomposed from N2 emissions in optical emission spectroscopy (OES) by analyzing them based on optics and plasma physics. Then, PI-VM is constructed by implementing PI and EES variables to partial least squares regression (PLSR). Compared to conventional VM, PI-VM improves the robustness more than twice in long-term variation by implementing PIWall on PLSR. Also, evaluation of the ranking of variables on PI-VM shows that the robustness is improved by decomposing PIWall and PIPlasma from OES based on optics and plasma physics. This result showed that an effective VM model for plasma-assisted process can be constructed by making phenomenological-based, statistical-tuned VM model that recognizes the drift of chamber wall condition and property of plasma separately, based on optics and plasma physics.