Algorithm implementation and techniques for providing more-reliable overlay measurements and better tracking of the shallow-trench isolation (STI) process

Abstract

The intent of Chemical Mechanical Polishing (CMP) is to minimize topography across a wafer. Topography is essential for edge contrast in optical overlay metrology'. Many ofthe advanced CMP processes cause a reduction in step height to much less than 20 Angstroms2. Conventional Overlay Metrology measurement optics require step heights in excess of 50 Angstroms to provide sufficient signal and contrast for algorithms to function properly. CMP further complicates the measurement task by causing significant step height and film thickness (contrast) variations as well as introducing asymmetry across a single overlay measurement target3. Such targets can be difficult to measure in standard metal and contact CMP applications let alone more modem front end CMP processes like Shallow Trench Isolation (STI). The registration marks become much more difficult to observe. Robust broadband optics with spatial filtering improves the contrast of the image; but the granularity of the substrate, the variation in edge detail, and the asymmetry of the registration mark can overwhelm a standard measurement algorithm. The registration mark measurement precision, accuracy and success rate can suffer due to the process variation in an aggressive front end CMP application such as STI.This paper presents the results of a new algorithm designed to improve the success rate, precision and accuracy of the measurements for low contrast targets produced by STI. The paper will also review the algorithm and discuss the results of target design optimization. Results will be provided from multiple lots with multiple wafer analyses demonstrating the effectiveness of the algorithm. Measurement yields improve from the 35%-50% success rate using current algorithms to 99% -1 00% success rate using the new algorithm. Precision was improved from I Onm to 3nm, and as low as I .2 nanometers 3cr. The true success of the algorithm is not just the improved measurement success, precision and accuracy; but it is in the verification that the edges are detected and measured accurately. Many current algorithms are giving estimates.

<title>Subresolution placement using IR image to CAD database alignment: an algorithm for silicon-side probing</title>

“…Measurement error in registration accuracy is proportional to the square root of the noise [10]. Therefore, one obvious reason for the relative unreliability of the FIB-to-Optical alignment is that both images contribute in lowering the signal-to-noise ratio of the correlation function, in contrast to the other cases where one of the images (CAD) is entirely noise-free.…”