Copper electrochemical deposition (ECD) is utilized in various fields of advance manufacturing ranging from complex multilayer flex and rigid printed circuit boards (PCB), to semiconductor copper plating including damascene and through-silicon vias (TSV) processes for the production of ULSI circuits. Monitoring of only the deliberately added bath constituents concentrations does not guarantee satisfactory performance of the plating solution. Prompt detection and diagnosis of disturbances leading to an out-of-control process behavior is critical for reducing losses in process operations. As ECD processes, like many chemical processes, are becoming more measurement-rich, conventional univariate process control methods become inadequate. Chemometric techniques have been applied to process problems (as opposed to analytical chemistry). These applications can be roughly divided between those directed at maintenance of process instruments, e.g., calibration and its transfer, and those that are concerned with the maintenance of the process itself, e.g., statistical process control (SPC) and fault detection. A system was developed utilizing a novel approach that combines electroanalytical methods including DC- and AC-voltammetry with various chemometric techniques using the same hardware for both applications of chemometrics: calibration and SPC. The focus of this presentation is on the latter area, specifically on statistical discriminant data analysis techniques based on chemometric factor analysis. All measurements and calculations were performed using instrumentation and software custom-developed for this application.The electroanalytical techniques employed were designed to provide a response strongly affected by the presence of foreign contaminants [1,2], accumulated degradation products, out-of-target concentrations of bath constituents and out-of-target physical conditions of the plating process (i.e. temperature). A modeling-power [1] based condition was defined and implemented to determine which portion of the voltammetric response can be effectively decomposed by factor analysis prior to discriminant analysis. The shape differences between deformed and training set voltammograms were quantified using various outlier-detection chemometric techniques including PCA [1], MD/PCA, MD/PCA/R [1], SIMCA and Fs-ratio [1].The effectiveness of the system was proven by detection of foreign contamination with iron in a copper damascene bath caused by unnoticed dropping of the carbon steel chapman bit into the plating chamber while changing the anode in between of high-volume production runs. Thanks to the prompt warning, significant production losses were prevented.This presentation focuses also on retrospective study leading to a development of an analytical model capable not only of detection but also of diagnosis of the specific type of contamination.A. Jaworski, H. Wikiel, K. Wikiel, ECS Trans. 25 (3), 199 (2009).A. Jaworski, H. Wikiel, K. Wikiel, Electroanalysis, 23 (1), 253 (2011).
Read full abstract