| Summary: | 博士 === 臺灣大學 === 化學工程學研究所 === 96 === The quality control of integrated circuit (IC) processing is becoming more and more important as the wafer becomes larger and the feature size shrinks. However, an advanced IC fabrication process consists of 300+ steps with scarce and usually difficult quality measurements. Thus product yield may not be realized until months into production while in-line measurements are available on the order of a millisecond. The series production nature and measurement setup lead to a unique process control problem. In this work, typical disturbances are explained and the possibility for disturbance rejection to each type is explored. The disturbance types can be categorized into three types: tool-induced, feed-induced, and context-induced disturbance.
For the tool-induced disturbance, feedback run-to-run controllers are usually used to deal with. To ensure stable process operation and ultimately meet the exacting requirements on final product quality, the typical advanced IC fabrication process requires many on-line sensors and off-line metrology tools for acquiring process and product information necessary for effective monitoring and control. However, the high cost associated with these measurement devices has made the economics of metrology a major factor. Various run-to-run controllers, carry out stability analyses, and analyze control system performance are derived. These results are then applied to the problem of rational metrology strategy selection where the effects of various metrology strategies on control system performance are systematically analyzed. In particular, if control performance takes priority over economics, results for determining maximum tolerable sampling intervals, maximum tolerable delay, and measurement priority are also presented.
Usually, pre-measurement is available before wafers are processing in an equipment tool. Thus, capability of the run-to-run control by sequencing the incomings (feed-induced disturbance) such that improved control performance can be achieved. The frequency domain explanation is: a negative feedback system is effective to reject low frequency type of disturbance. From the feedback property, then the answer to the feed sequencing problem becomes clear: rearrange the feed in such a way that it gives a low frequency characteristic. The issues of controller tuning, model mismatches, and time-varying (slow drifting) parameters are also explored and the analyses reveal the robust performance of the proposed feeding policies. The feed sequencing problems are tested for systems with different dimensions, e.g., SISO, SIMO, and MIMO systems which include the model of an experimental CMP process.
Besides tool-induce and feed-induced disturbance, there are possible other factors such as metrology tool bias, product type, or chamber that may induce disturbance. To involve all other types, the context-based state estimation method is usually used. The most important feature of a context-based system is rank deficiency, and the proposed method unbiasedly estimates relative status of each context and process output by state transformation. The transformed states are straightforward and physically meaningful. Furthermore, a solution of planning paths with guarantee of output performance is also investigated.
|
|---|
