Advanced Modeling Techniques Expand the Applications of Picosecond Laser Acoustics for RF Process Monitorina

Abstract

Picosecond Ultrasonics (PULSE™ technology) [1] is a first principle, rapid non-contact, non-destructive, technology, for the measurement of single layer and multilayer metal films in semiconductor process control. It has a strong footprint and is uniquely positioned as a tool-of-record for metal film thickness metrology in RF filter monitoring. In addition to thickness measurements, the technique can be used to characterize acoustic velocity values for dielectric and piezoelectric materials, which is a critical parameter for process control. We have previously reported on the advantages of PULSE technology for RF applications and its excellent performance to meet the stringent requirements for process monitoring and control [2]. Most of the RF applications involve multilayer metal stacks or films on oxide that are more intuitive and are easier to measure and model using our standard modeling algorithms. However, at the measurement wavelength of 522nm, measuring thinner metal films directly on Si substrate is challenging and often requires films on oxide or other films as the signal is complex and dominated by Si oscillations. Another situation in which modeling is challenging is when an SiO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> film is present as a capping layer. The oxide film is included as part of the device stack to obtain a low thermal coefficient of the acoustic device. In a typical full stack, when the oxide capping layer is present over a multilayer stack of oxide/ electrode/piezoelectric layer/ electrode/Si, PULSE measurement signal is a convolution of SiO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> oscillations and echoes from the multilayer stacks (most often three or more layers). To expand the application space of the PULSE technology and improve its performance, we have developed advanced modeling techniques for the deconvolution of different components of the signal to reliably model parameters of interest. In this paper, we review one of the approaches we have successfully used to improve sensitivity, accuracy and robustness without impacting the repeatability needed for high-volume manufacturing.