Real-time Monitoring and Optimization of Plasma Etching for Semiconductor Manufacturing

Abstract

For several decades, the semiconductor industry has rapidly progressed by the fast paced improvements in its technology. The continuous shrinkage of the chip size makes sub-20nm device era appear, which enables microprocessors with several GHz multiple-core central processing unit and various types of memory devices with several hundred gigabyte capacity. This technology evolution of the industry has been driven by the Moores law, which has proven to be accurate until the current sub-20nm device era. However, the physical difficulties in materials and patterning technology to cross the 10nm threshold force semiconductor manufacturers to start thinking of more than Moores law. The concept of more than Moores law was firstly addressed in 2007, which is the equivalent scaling by improving the performance of the device, not by shrinking the chip size. Several equivalent scaling approaches are classified into inclusion of process technologies, additional chip and system-level architectural and software design, transition to 450mm wafer, and cost-effective manufacturing through real-time monitoring and control. Semiconductor manufacturing is composed of various processes such as photo lithography, dry etch, diffusion, ion implantation, thin film deposition, cleaning and chemical mechanical planarization. Among these processes, plasma related processes occupy more than 30% of the whole manufacturing steps. Specifically, plasma etching leads technology evolution in plasma equipment for semiconductor manufacturing. The cost-effective manufacturing through monitoring and control in plasma etching has been delayed due to the inherent complexity of plasma, lack of plasma sensors, integration issues from deposition and photolithography even though the plasma etching is one of the core processes for semiconductor manufacturing. As a result, the rapid technology development for plasma etch in terms of the cost effective manufacturing is crucial in winning the competitiveness in semiconductor industry. This thesis has addressed issues in cost effective plasma etch operations and solutions: sensor variable selection and utilization technique, virtual metrology to predict critical dimension, chamber conditioning after wet cleaning, and chamber to chamber matching. All of the developed methodologies were demonstrated in semiconductor manufacturing environments Integrated square response based sensor variable selection technique was introduced for handling the scaling issues from various physical properties of sensors in plasma etching and for helping engineers to intuitively select state variables related to manipulate variables. This technique can be integrated with relative gain array and singular value analysis to strengthen its usefulness in plasma etching. Issues in implementing a robust virtual metrology for plasma etching were discussed: state-of-the art plasma sensors, effective selection of plasma sensor variables responding to individual manipulated variable, sensor data shift across preventive maintenance. In order to handle selection of plasma sensor variables, the integrated square response based sensor selection is refined by the interaction analysis with non-square relative gain array, which can reduce the number of input variables for virtual metrology. With the help of plasma sensor variables and its optimum sensor variable selection, simple linear regression methods such as multiple linear regression and partial least squares regression are successfully applied to predict a metal line critical dimension in plasma etching. The mean absolute percentage error of the virtual metrology systems is less than 5%, which can be maintained by the cost-effective recursive coefficient update technique even under dynamic semiconductor manufacturing environments. A systematic procedure to optimize chamber seasoning conditions with optical emission spectroscopy was suggested to address the process drift after wet cleaning in plasma etching. In order to achieve a quantitative analysis of plasma spectra without being disturbed by noises from optic systems, a self-background normalization technique is introduced. Also in order to automatically determine optimum seasoning conditions, a multiple input multiple output control strategy is applied and the optimum condition is obtained by solving the quadratic optimization problems. The suggested methodology was successfully demonstrated in a dynamic random access memory device manufacturing environment which is suffering from a serious process drift after wet cleaning. The equipment control approach was suggested to solve chamber to chamber performance matching problems. The decomposed etch rate map which enables representation of etch rate profile within a wafer were introduced to design a multiple input multiple output controller with 3 controlled variables. With the 3 controlled variables, the singular value analysis and the relative gain array methods were incorporated with the integrated square response based sensor variable selection technique to find the optimal sets of manipulated variables and controlled variables. In order to find an optimum process condition to match the etch rate performance from the worst to the golden chambers, an optimization problem with constraints was solved. The suggested process condition was applied to the worst chamber, which results in the improvement of the performance matching index from 100.7 to 31.6. Hopefully, the proposed methodologies in this thesis will be disseminated to semiconductor manufacturers who are experiencing similar issues.