Brian E. Thompson

Continuum modeling of argon radio frequency glow discharges

A continuum model has been developed which successfully describes the concentration, movement, and energetics of charged particles within a rf discharge. This model includes continuity equations for all charged particles, Poisson’s equation to determine the local electric fields, and an electron energy balance to determine the ionization and energy‐loss rates. All input parameters (diffusivity, mobility, ionization, and energy‐loss rate) were defined using reported values determined in dc field experiments.

Ion bombardment energy distributions in radio‐frequency glow‐discharge systems

The measurement and interpretation of ion bombardment energies in rf discharges of SF6, CF3Cl, and CF3Br from 0.2 to 1.0 Torr, are discussed. Errors in ion sampling due to disturbances of the electric field and neutral density around the sampling orifice are shown to be most important at higher pressures and with larger orifices. The dependence of the ion bombardment energy distribution on the electric‐field‐to‐pressure ratio is reviewed. Combining this relation with a simple electrical model of a plasma gives estimates of the ion bombardment energies in collisional sheaths. The bombardment energy is proportional to (rf current)/(electrode area×frequency ×pressure) with the proportionality constant for a particular system depending on the collision cross sections and relative ion‐to‐netural mass ratio. The constants found for the three gases studied experimentally are close to theoretical estimates.

Monte Carlo simulation of ion transport through rf glow‐discharge sheaths

The transport of ions through rf glow‐discharge sheaths was simulated with a Monte Carlo method to determine the distributions of ion‐bombardment energy and angle of impact. Several sheath parameters were varied and their effects examined: (1) the type of ion‐molecule scattering (hard sphere, potential field interaction, charge exchange), (2) the ratio of ion and neutral masses, (3) the ratio of the sheath width to collision mean free path, (4) spatially uniform, spatially linear, and time‐dependent (rf) electric fields in the sheath, and (5) the frequency of an rf component in the sheath. The results of the Monte Carlo simulations indicate that the type of elastic scattering (hard sphere or soft sphere) does not significantly change either the impact angle distributions or the scaled ion‐bombardment energy distributions. Charge‐exchange scattering produces a much greater ion‐bombardment directionality and a different shape of the ion‐bombardment energy distribution. The fully developed distributions depe...

Atomic chlorine concentration measurements in a plasma etching reactor. I: A comparison of infrared absorption and optical emission actinometry

Atomic chlorine concentrations in Cl2 and CF3Cl plasmas have been measured using both infrared absorption spectroscopy and optical emission actinometry. These measurements were made over a range of plasma conditions including plasma excitation frequencies of 72 kHz–13.5 MHz, power inputs of 10–100 W, and pressures of 200–800 mTorr. In Cl2 plasmas, the technique of optical emission actinometry misrepresents atomic chlorine concentration changes by nearly an order of magnitude. The errors in the actinometry technique are believed to result from excited state Cl production by electron impact dissociation of Cl2. A simple model for Cl emission is in good agreement with the experimental observations. In CF3Cl discharges, the technique of optical emission actinometry is shown to accurately represent variations in atomic chlorine concentration with changing process conditions.

Comparison of measured and calculated SF6 breakdown in rf electric fields

Measured breakdown characteristics of SF6 at 4–14 MHz are compared with the breakdown predicted with an electron‐balance model. This balance includes terms for the formation of electrons from ionization of neutral molecules and electron losses by attachment to neutral molecules, electric‐field‐driven fluxes, and diffusion. With model parameters obtained from existing literature on electron properties in dc electric fields, the model predicts radio frequency (rf) breakdown voltages in agreement with the experimental values. The electron balance model is also expressed in a nondimensional form to aid in generalizing rf breakdown to other gases. The electron concentration profiles formed when including the electric‐field‐driven fluxes are compared with the sinusoidal distribution achieved for diffusion‐controlled breakdown. The success of this model suggests that glow discharges can be modeled in a similar fashion.

Nucleation on SiO2 during the Selective Chemical Vapor Deposition of Tungsten by the Hydrogen Reduction of Tungsten Hexafluoride

A horizontal hot-wall chemical vapor deposition (CVD) quartz reactor with rectangular cross section was used to study the effect of different process conditions on the nucleation of tungsten on SiO 2 during selective WCVD by the H 2 reduction of WF 6 . The experimental procedure included placing a metallic surface at the center of the reactor, and small samples of SiO 2 at different positions both upstream and downstream with respect to the metallic surface. Digitized scanning electron microscopy micrographs were used to determine the particle size distributions of nuclei on the SiO 2 surfaces

Comparing a high and low-level deep neural network implementation for automatic speech recognition

The use of deep neural networks (DNNs) has improved performance in several fields including computer vision, natural language processing, and automatic speech recognition (ASR). The increased use of DNNs in recent years has been largely due to performance afforded by GPUs, as the computational cost of training large networks on a CPU is prohibitive. Many training algorithms are well-suited to the GPU; however, writing hand-optimized GPGPU code is a significant undertaking. More recently, high-level libraries have attempted to simplify GPGPU development by automatically performing tasks such as optimization and code generation. This work utilizes Theano, a high-level Python library, to implement a DNN for the purpose of phone recognition in ASR. Performance is compared against a low-level, hand-optimized C++/CUDA DNN implementation from Kaldi, a popular ASR toolkit. Results show that the DNN implementation in Theano has CPU and GPU runtimes on par with that of Kaldi, while requiring approximately 95% less lines of code.

Continuum Modeling of Charged Particle Transport: RF Breakdown and Discharges of SF 6

Continuity equations for the concentration of electrons, positive ions, and negative ions were constructed and solved to predict rf breakdown voltages and the electrical properties of SF, discharges.These balances for the three types of charged species include terms for convection (electric field-driven fluxes), diffusion, and reactions (ionization, electron attachment, and negative-positive ion recombination).The mobilities, diffusivities, and reaction rate coefficients necessary for the rf discharge model are based on reported measurements and calculations of these parameters in dc electric fields.The electric fields developed in the rf discharge are calculated from Poissons equation and applied voltage conditions.Predictions based on this model are compared with measured rf breakdown characteristics of SF 6 .

Discrimination between singing and speech in real-world audio

The performance of a spoken language system suffers when non-speech is incorrectly classified as speech. Singing is particularly difficult to discriminate from speech, since both are natural language. However, singing conveys a melody, whereas speech does not; in particular, a singers fundamental frequency should not deviate significantly from an underlying sequence of notes, while a speakers fundamental frequency is freer to deviate about a mean value. The present work presents a novel approach to discrimination between singing and speech that exploits the distribution of such deviations. The melody in singing is typically not known a priori, so the distribution cannot be measured directly. Instead, an approximation to its Fourier transform is proposed that allows the unknown melody to be treated as multiplicative noise. This feature vector is shown to be highly discriminative between speech and singing segments when coupled with a simple maximum likelihood classifier, outperforming prior work on real-world data.