Correlation of CMP slurry properties including large particle counts (LPC) to semiconductor manufacturing defect and yield metrics has long been a promise of proper implementation of undiluted slurry process monitoring [1]. Continuous, real-time metrology validating this correlation has now been in use in volume production for over three years in fabs in Asia and the US [2]. In addition to wafer defects, such metrology has also proven useful in identifying slurry lot and tank changes, filter changes, and other operational events in the fab [3]. Process monitoring metrology often shares many attributes with the laboratory analytical methods on which it is based. This is the case for slurry particle analysis using a variety of single-particle optical sensing (SPOS) technologies that have been available over the past 30 years. SPOS is a laboratory analysis technique that has been adapted to provide periodic monitoring of the particle size distribution in CMP slurry. Due to small sample volume, periodic sampling of the production slurry stream, delay time, and chemical and electrostatic disturbance of the suspension due to sample dilution, users can obtain only a gross record of slurry characteristics with low statistical inference power. Correlation of SPOS data with yield and defect events has proven to be unsatisfying, with consecutive data points exhibiting a high level of random variation that masks trends and slurry events. Nonetheless, because SPOS has been the only option available, there is a vast historical database and an adequate comfort level in the semiconductor industry despite known shortcomings. Recent advances in sample delivery and dilution methods have attempted to address these issues, but the underlying problem of statistical significance due to small sample volume [4] remains a major contributor to SPOS data inconsistency. To address these fundamental shortcomings of SPOS as a manufacturing process monitor, the Vantage SlurryScope™ system was designed from its inception to be a continuous, real-time, undiluted slurry process monitor. As such, a primary requirement is to detect changes in the characteristics of the CMP slurry for any reason as a function of time. Less importance is attributed to precise reporting of the characteristics of isolated particles, although this option is available under some conditions. An SPOS-like data conversion becomes physically untenable in flowing, undiluted slurry, where resolution of individual particles is confounded by massive particle coincidence, with concurrent light detection from thousands to millions of particles on each sensor element, and with the cumulative light scattering from small particles that are individually below the detection threshold. A solution which provides continuous, real-time process monitoring can be more effective without depending on individual particle resolution. To enable robust process monitoring under these constraints, a novel, computationally efficient method has been developed to extract meaningful information about CMP slurry characteristics from the large set of light scattering observations that SlurryScope produces. This method does not depend on the resolution of individual particles or on a preset minimum particle size resolution. While these parameters are crucial for analyzing and reporting particle size distributions, they are not essential for process monitoring. The new Vantage method presented here uses the entire distribution of light power scattered by the flowing slurry and integrated by SlurryScope sensors to detect changes in that distribution which correspond to physical changes in the slurry. More than 107 light scattering integrations per second are collected from a highly controlled sampling volume. This circumvents SPOS sampling limitations, and opens new directions in process monitoring of CMP slurries. An initial implementation of such monitoring metrics for high density ceria slurries, called “HD Mode,” is described in this paper. The SlurryScope system has demonstrated a high sensitivity to slurry properties due to lot-to-lot differences, tank changes, filter changes, dilution and mixing errors, mechanical failures, and violation of turnover limits. In one production fab study [2], the correlation of SlurryScope monitoring data to process yield and defects provided a 3x improvement over traditional SPOS methods. Evaluation of the new algorithm is continuing, with benefits including the use of a single hardware sensor and computational engine for all slurries from low refractive index silica to high refractive index ceria across a full range of fab production concentrations. The elimination of dependence on a minimum particle size detection limit and a continuous, real-time sensitivity to changes in slurry conditions allows the user to achieve process defect objectives in sub-10nm device technologies. REFERENCES [1] K.A. Barry, APC XXV Symposium, 2013. [2] J. Bennett, M.A. Fury, ICPT, 2014. [3] C.D. Aparece, M.A. Fury, Semicon West, 2013. [4] M.A. Fury, NCCAVS CMPUG, May, 2012.
Search from 250 M+ research papersSearch from 250 M+ research papers
Open Access
May 26, 2016
Dawei Yu + 8
