Laboratory for Manufacturing and Sustainability

Increasing systemic error during copper CMP (Chemical Mechanical Planarization) is due to the uneven surfacetopography generated during the process. A mechanistic model based on a fundamental understanding of the processconstituents was proposed to predict material removal rates and the post CMP topography. Two synergisticmechanisms were proposed: 1) chemically dominant behavior is explained by the repetitive removal and formation ofa protective layer on copper surface and chemical dissolution during the process, 2) mechanically dominant removalmechanism is due to the material behavior of copper at the nano-scale and subsequent oxidation and removal of theplastically deformed copper. As a step forward to optimize the process and the manufacturing system, this model wasextended to explain pattern dependent variability during copper CMP.

Millisecond scale benzotriazole (BTA) adsorption kinetics in acidic aqueous solution containing 0.01 M glycine and 0.01 M BTA have been investigated. Chronoamperometry was used to measure current densities on the surface of a micro-copper electrode in pH 4 aqueous solutions containing 0.01 M glycine with or without 0.01 M BTA. In the presence of BTA the current density decreased as the inverse of the square root of time for a few seconds due to adsorption of BTA. At potentials above 0.4 V saturated calomel electrode the current leveled off after a second or so due to the formation of a Cu(I)BTA monolayer on the copper surface. Based on these data a governing equation was constructed and solved to determine the initial kinetics of BTA adsorption. Analysis shows that material removal during copper chemical mechanical planarization (CMP) in this slurry chemistry occurs mostly by direct dissolution of copper species into the aqueous solution rather than mechanical removal of oxidized or pure copper species and that each interaction between a pad asperity and a given site on the copper removes only a small fraction of the Cu(I)BTA species present at that site.

CIRP has had a long history of research and publication on the development and implementation of sensor monitoring of machining operations including tool condition monitoring, unmanned machining, process control and, more recently, advanced topics in machining monitoring, innovative signal processing, sensor fusion and related applications. This keynote follows a recent update of the literature on tool condition monitoring and documents the work of the cutting scientific technical committee in CIRP. The paper reviews the past contributions of CIRP in these areas and provides an up-to-date comprehensive survey of sensor technologies, signal processing, and decision making strategies for process monitoring. Application examples to industrial processes including reconfigurable sensor systems are reported. Future challenges and trends in sensor based machining operation monitoring are presented.

In this study, we performed a specific precision grinding process in an attempt to improve the mirror-quality finish and tribological characteristics of titanium nitride based coating films (TiN, TiCN, and TiAIN). The ground surfaces were highly smooth with no evidence of cracking, chipping, or peeling, demonstrating that the hard coating films were finished uniformly. For the TiAIN coating, a significant high level mirror-quality finish was achieved with an average roughness Ra of 4 nm. In addition, for all films, the employed grinding process led to superior tribological characteristics. In the case of the TiN film, the precision grinding process produced a carbon- and copper-rich surface layer, as well as higher compressive residual stress.

The material removal characteristics of a silicon wafer were experimentally investigated with respect to the chemical dissolution and mechanical abrasion of the wafer during silicon chemical mechanical polishing (CMP) using an alkali-based slurry. The silicon surface without native oxide is rapidly dissolved by the slurry containing an amine agent, which effectively leads to the reduced hardness of the loaded silicon wafer due to Si–Si bond breaking during polishing. The abrasive particles in the slurry easily remove the reacted silicon surface, and the removal rate and wafer non-uniformity for abrasive concentrations of 1.5–3 wt% are better than those for other concentrations because of the low and steady coefficient of friction (COF) owing to the evenness of abrasive particles between the wafer and pad. Also, it was found that a high slurry flow rate of 700–1000cm3/min improves wafer non-uniformity owing to the reduced temporal variation of temperature, because the slurry acts as a good cooling source during polishing. However, the removal rate remains almost constant upon varying the slurry flow rate because of the effective dissolution characteristic of the slurry with abundant amine as an accelerator, regardless of the reduction of average temperature with increasing slurry flow rate. In the break-in process used to stabilize the material removal, the viscoelastic behaviors of the pad and the ground wafer surface with native oxide and wheel marks cause a temporal change of the friction force during polishing, which is related to the removal rate and wafer non-uniformity. As a result, the stabilization of removal rate and wafer non-uniformity is achieved through a steady-state process with elevated temperature and reduced COF after a total polishing time of 60 min, based on the removal process of the wafer surface and the permanent deformation in the viscoelastic behavior of the pad.

Motion control in high-speed micromilling processes requires fast, accurate following of a specified curvilinear path. The accuracy with which the path can be followed is determined by the speed at which individual trajectories can be generated and sent to the control system. The time required to generate the trajectory is dependent on the representations used for the curvilinear trajectory path. In this study, we introduce the use of subdivision curves as a method for generating high-speed micromilling trajectories. Subdivision curves are discretized curves which are specified as a series of recursive refinements of a coarse mesh. By applying these recursive properties, machining trajectories can be computed very efficiently. Using a set of representative test curves, we show that with subdivision curves, trajectories can be generated significantly faster than with NURBS curves, which is the most common method currently used in generating high-speed machining trajectories. Trajectories are computed efficiently with subdivision curves as they are natively discretized, and do not require additional evaluation steps, unlike in the case of NURBS curves. The reduced trajectory generation time allows for improved performance in high-speed, high-precision micromilling. We discuss the use of several metrics to quantify the quality of the subdivision interpolation, and apply them in calculating the error during trajectory generation for the test curves.

Robust interoperability methods are needed in manufacturing systems to implement computeraided process planning algorithms and to verify their effectiveness. In this paper we discuss applying MTConnect, an open-source standard for data exchange in manufacturing systems, in addressing two specific issues in process planning and verification. We use data from an MTConnect-compliant machine tool to estimate the cycle time required for machining complex parts in that machine. MTConnect data is also used in verifying the conformance of toolpaths to the required part features by comparing the features created by the actual tool positions to the required part features using CAD tools. We demonstrate the capabilities of MTConnect in easily enabling process planning and verification in an industrial environment.

The strong Ce-O-Si bonding between CeO2 abrasives and SiO2 film surface; i.e., the chemical tooth effect, improved planarization efficiency in CMP using ceria-based slurry as a result of nonlinear behavior of the removal rate. Removal rate is a power function of pressure and relative velocity (i.e., RR = kPαV β ). In particular, the high dependency of removal rate on pressure when α >1 results in a much higher material removal rate in the upper pattern than in the lower pattern. Therefore, the planarization efficiency of ceriabased slurry is better, from initial polishing time to the completion of the polishing step, than that of conventional silica-based slurry with an exponent value of α ≈1.

Today the requirements for reducing the impact of our manufacturing activities are increasing as the world awakes to and addresses the environmental impacts of our society. Energy consumption, greenhouse gas emissions, materials availability and use, environmental impact levels, etc. are all topics of interest. Semiconductor manufacturing in general and process steps such as CMP are not exempt from this and, in many cases, the industry has led the efforts in reducing impacts. This paper will first review some of the drivers for sustainable manufacturing, then define some of the terms that will be useful for determining the engineering aspects of sustainability and sustainable manufacturing, as well as metrics for assessing the impact of manufacturing in general and CMP in particular. An assessments of CMP will be given to illustrate the potential for “design for the environment” in CMP and related processes. Consideration will be given to research opportunities, including process modeling, that this focus provides to CMP researchers, consumable suppliers and industry.

Copper CMP is a corrosion-wear process, in which mechanical and chemical-electrochemical phenomena interact synergistically. Existing models generally treat copper CMP as a corrosion enhanced wear process. However, the underlying mechanisms suggest that copper CMP would be better modeled as a wear enhanced corrosion process, where intermittent asperity/abrasive action enhances the local oxidation rate, and is followed by time-dependent passivation of copper. In this work an integrated tribo-chemical model of material removal at the asperity/ abrasive scale was developed. Abrasive and pad properties, process parameters, and slurry chemistry are all considered. Three important components of this model are the passivation kinetics of copper in CMP slurry chemicals; the mechanical response of protective films on copper; and the interaction frequency of copper with abrasives/pad asperities. The material removal rate during copper CMP was simulated using the tribo-chemical model, using input parameters obtained experimentally in accompanying research or from the literature.