Feed-forward, as a method to control the Lithography process for Critical Dimensions and Overlay, is well known in the semiconductors industry. However, the control provided by simple averaging feed-forward methodologies is not sufficient to support the complexity of a sub-0.18micrometers lithography process. Also, simple feed-forward techniques are not applicable for logics and ASIC production due to many different products, lithography chemistry combinations and the short memory of the averaging method. In the semiconductors industry, feed-forward control applications are generally called APC, Advanced Process Control applications. Today, there are as many APC methods as the number of engineers involved. To meet the stringent requirements of 0.18 micrometers production, we selected a method that is described in SPIE 3998-48 (March 2000) by Terrence Zavecz and Rene Blanquies from Yield Dynamics Inc. This method is called PPC, Predictive Process Control, and employs a methodology of collecting measurement results and the modeled bias attributes of expose tools, reticles and the incoming process in a signatures database. With PPC, before each lot exposure, the signatures of the lithography tool, the reticle and the incoming process are used to predict the setup of the lot process and the expected lot results. Benefits derived from such an implementation are very clear; there is no limitation of the number of products or lithography-chemistry combinations and the technique avoids the short memory of conventional APC techniques. ... and what's next? (Rob Morton, Philips assignee to International Sematech). The next part of the paper will try to answer this question. Observing that CMP and metal deposition significantly influence CD's and overlay results, and even Contact Etch can have a significant influence on Metal 5 overlay, we developed a more general PPC for lithography. Starting with the existing lithography PPC applications database, the authors extended the access of the analysis to include the external variables involved in CMP, deposition etc. We then applied yield analysis methods to identify the significant lithography-external process variables from the history of lots, subsequently adding the identified process variable to the signatures database and to the PPC calculations. With these improvements, the authors anticipate a 50% improvement of the process window. This improvement results in a significant reduction of rework and improved yield depending on process demands and equipment configuration. A statistical theory that explains the PPC is then presented. This theory can be used to simulate a general PPC application. In conclusion, the PPC concept is not lithography or semiconductors limited. In fact it is applicable for any production process that is signature biased (chemical industry, car industry, .). Requirements for the PPC are large data collection, a controllable process that is not too expensive to tune the process for every lot, and the ability to employ feedback calculations. PPC is a major change in the process management approach and therefor will first be employed where the need is high and the return on investment is very fast. The best industry to start with is the semiconductors and the most likely process area to start with is lithography.
|