(632d) Multivariable Run-to-Run Control for an Atomic Layer Etching Process Using a Transformer Model

Semiconductor fabrication comprises around 500 processing steps [1] that demand stringent design criteria. One of these processing steps utilizes a top-down, atomic layer etching (ALE) procedure in which monolayers of high-κ oxide films are removed from transistors to produce thin substrate coatings with nanoscale thicknesses. These films serve to minimize short-channel effects as well as heat and current losses to preserve the performance and longevity of the finished wafer [2]. ALE requires that each cycle of etching observes a self-limiting characteristic to maintain the uniformity of the film, which would result in self-aligned structures that promote transistor stacking. However, ALE processes can encounter disturbances that affect substrate quality and cause defective devices. With identified criteria for optimal performance, the control of ALE processes is necessary to maintain transistor qualities for subsequent processes [3]. Thus, offline feedback control, in the form of run-to-run (R2R) control, must be integrated into the ALE process.

A recurring obstacle plaguing the semiconductor industry is the generation of sufficient and meaningful data to characterize these ALE processes for optimizing productivity and accuracy. The use of in silico multiscale computational fluid dynamics (CFD) modeling [4], in which a mesoscopic kinetic Monte Carlo (kMC) simulation is conjoined to a macroscopic CFD model, resolves the burden of having to conduct experiments for a range of operating conditions to construct a meaningful data set. This multiscale model for a prior ALE process to fabricate Al₂O₃ films [5] will be simulated in conjunction with an R2R controller that utilizes an exponentially weighted moving average (EWMA) [6] of the correlated multiscale model in each batch run to adjust the input, precursor flow rates, by measuring the offset in etching per cycle (EPC). However, the R2R controller relies on an offline measurement of the mass loss of the substrate after each batch run using a Quartz Crystal Microbalance (QCM), which is nonproductive and not attractive for industrial practices [7]. Thus, a transformer is applied to real-time data gathered from recurring batch runs to act as a predictor of the EPC for each batch-run. The EPC measurement is then used in a self-tuning strategy used in place of autoregressive moving averages (ARMAs) [8]. Through this manner, an R2R controller is established by bounding the measured EPC within an upper and lower limit.

[1] Richard, C., 2023. Understanding Semiconductors: A Technical Guide for Non-Technical People. Apress, Berkeley, CA.

[2] Jurczak, M., Collaert, N., Veloso, A., Hoffmann, T., Biesemans, S., 2009. Review of FINFET technology. In: 2009 IEEE International SOI Conference, 1–4, Foster City, CA, USA.

[3] Moyne, J., del Castillo, E., Hurwitz, A. M., 2018. Run-to-Run Control in Semiconductor Manufacturing. CRC Press.

[4] Maroudas, D., 2000. Multiscale modeling of hard materials: Challenges and opportunities for chemical engineering. AIChE Journal, 46, 878–882.

[5] Yun, S., Tom, M., Ou, F., Orkoulas, G., Christofides, P. D., 2022. Multiscale computational fluid dynamics modeling of thermal atomic layer etching: Application to chamber configuration design. Computers & Chemical Engineering, 161, 107757.

[6] Montgomery, D. C., 2013. Introduction to Statistical Quality Control. John Wiley & Sons, Hoboken, 7th edition.

[7] Del Castillo, E. Hurwitz, A. M., 1997. Run-to-run process control: literature review and extensions. Journal of Quality Technology, 29, 184–196.

[8] Del Castillo, E., 1996. A multivariate self-tuning controller for run-to-run process control under shift and trend disturbances. IIE Transactions, 28, 1011–1021.