This project presents the development of a new process flow in the Washington Nanofabrication Facility to create a magnetically neutral Nuclear Magnetic Resonance (NMR) test device for use at the Department of Energy (DOE) Pacific Northwest National Labs (PNNL). At PNNL, within the Environmental Molecular Sciences Laboratory, the devices are used to characterize environmental contaminants, like radioactive isotopes, using NMR spectroscopy. The devices being developed are to be superior in uniformity and layering a stack of thick copper sandwich with a precisely deposited aluminum layer. The project includes computer-aided design (CAD) and fabrication of a precision mask set for fine resolution that limits irregularities in the test structure that could adversely impact the NMR measurement. The fabrication process starts with an electroplating seed, photolithography, and thick electroplating of copper. Following plating, chemical mechanical polishing is performed to yield an atomically smooth surface. After this, a precision thickness aluminum layer, that cannot be conventionally electroplated, is deposited and patterned via a lift-off process. This project will also be a primer for more industry-focused capabilities for the WNF with eventual development of through-silicon vias (TSVs) for research applications.

The purpose of this research is the development of a precisely electroplated set of vacuum capacitor electrodes for inclusion in the MAJORANA Demonstrator (MJD) - the first stage of the MAJORANA project. Funded by the Department of Energy the MAJORANA program is part of an international effort to study neutrinos, dark matter, and axions led by the UW Center for Experimental Nuclear Physics and Astrophysics (CENPA). The project focus is on making the capacitors with a physical dimension tolerance (<2 microns) that is far less than what can be obtained via machining or other comparable methods of manufacturing. The desired capacitors are made using high-aspect ratio patterns in silicon as an electroplating mold. Silicon wafers are coated with an electroplating seed layer and then bonded to each other using thermcompression wafer bonding. Thermocompression bonding is a technique used to bond wafers by applying heat and force simultaneously. Following bonding, the wafers are patterned, using a photolithography process and Deep Reactive Ion Etching (DRIE), to form the mold cavaties. Photolithography is a multiple step process that includes application and patterning of a light-sensitive material called photoresist. Next, bottom-up copper electroplating is performed until excess deposit of copper called over-plating is achieved. The over-plated structures are polished back to establish dimensional uniformity. The final step is releasing the electroplated structures via a blanket DRIE process. Successful completion of this research project will allow assembly of MJD which in turn will be used to search for neutrinoless double-beta decay – one of the rarest forms of radioactive deacy. This project will also be a primer for more industry-focused capabilities for the Washington Nanofabrication Facility (WNF) with eventual development of through-silicon vias (TSVs) for research applications and a potential foundry process for precision part creation without the limitations of machining.

Language models characterize the probability of a sequence of words for use in a variety of natural language processing applications, including automatic speech recognition. Web text is a key source of data for developing the language models used in such systems because of its sheer volume, and the fact that much of it is freely available to the public. However, web text varies broadly in genre, depending on purpose, audience, formality, and other characteristics. For example, the difference between the formal exposition of a Wikipedia article and the informal banter between two Twitter users. When using web text to train language models for conversational speech recognition in particular, it is important to find text that resembles that speaking style. When training a language model using various web text sources, we use a measure of worth called perplexity. By selecting web text that matches the conversational style of speech, the language model perplexity improves, which generally leads to higher accuracy speech recognition. I work on selecting text taken from Twitter. This involves research on what kinds of Twitter searches will result in tweets containing useful text, as well as how to identify Twitter users who post useful text.

Disfluent speech has typically been regarded as noise in spoken language processing: aspects of speech that should be ignored in order to automatically detect and understand the intended meaning of spoken language. However, it has also been suggested that disfluencies provide information about a speaker's cognitive processes and efforts to manage the discussion. Most work on disfluencies has investigated very controlled scenarios or conversational speech. In our work, we are studying high-stakes debates, including Supreme Court oral arguments and congressional hearings on the financial crisis, in addition to informal conversations among strangers and family members. By studying disfluent speech in both high and low stress situations, we can better understand the impact of stress on disfluency rates. We use text alignment techniques to understand the relationship between the perceived mistakes and correction. We subdivide and partition disfluencies into categories by classifying inserted, deleted, and substituted words according to their semantic similarity; thus, making correlations between social context and disfluency rate more evident in addition to providing a better interpretation of speaker opinions. Preliminary results show that lawyers have a tendency to model their speech patterns off of the justices they are speaking to, but not vice versa. That is, lawyers tend to be more disfluent when the justices are speaking disfluently, which indicates that disfluent speech reflects social context. Thus, we hope to develop ways to automatically detect and understand deeper meaning in spoken language, such as defensiveness and strength of opinions, through our study of disfluent speech.

In order to increase the accuracy of particle detectors, new sensors and digital acquisition systems must be developed. With the power of reconfigurable digital logic, an acquisition system has been created using a simple and fast architecture in order to test T3MAPS, a prototype sensor for the ATLAS detector. The firmware design of the system is modular to allow for iterative improvements over time and future changes to the sensor. Currently the systems first iteration is being tested using oscilloscopes, and then the system will be connected to T3MAPS for more advanced function tests. In the future the firmware will be ported to multiple platforms with additional functions to allow time over threshold test to be conducted on T3MAPS. T3MAPS is new sensor designed to detect high-energy particles generated by particle accelerators. The ATLAS detector is one of many particle detectors at the Large Hadron Collider, a particle accelerator run by CERN. The firmware for this system is written in a hardware description language, Verilog. This language is used to describe digital designs as a programming language, enabling reconfigurable digital circuits. This offers a vast level of flexibility. As nobody has tried to implement an acquisition system for T3MAPS, great care must be taken to insure that the signal to the sensor is correct and electrically clean. This is verified using a logic analyzer and an oscilloscope to insure the signal meets the specifications required. After signal verification, testing will be conducted on T3MAPS itself using the verified design. Tests will progress from simple configuration of the sensor to full pixel by pixel scans of the chip. Results from these tests will be analyzed to compare the speed and accuracy of this system vs. the existing manual methods.

Sleep apnea is a common syndrome where a person has an irregular breath rate only during sleep, making it difficult to be detected and diagnosed. Yet, in the U.S. about two percent of children are affected by this syndrome and thus at risk of some health issues. There are some prevalent breath detection products on the market but they remain inaccessible due to the undesired prices and limited range of detection. In the breath detection project I was participating in, we intended to detect the breath rate by wireless technology, e.g. home-based WiFi. Our design would ease the cost of production by zero assembly and installation of the hardware components and would achieve the detection for anywhere inside a house (whole-home detection) by wireless signal transmissions in line-of-sight, none-line-of-sight, and through-the-wall scenarios. The detection system broadcasts the wireless signals in a home environment, receives the reflected signals from the targeted person, and then computes the frequency of the target’s breathing over a period of detection. We expected that frequency plots would be uniform for a normal person but irregular for a patient with sleep apnea. One practical implication of performing such frequency analysis in the whole-home environment is that children’s health monitoring can still happen without parental interference.

As renewable energy constitutes a greater portion of the generation fleet, so does the importance of modeling uncertainty as part of integration studies. In pursuit of optimal system operations, it is important to capture not only the definitive behavior of power plants but also the risks associated with system wide interactions. Load forecasting is an area of renewable energy integration studies that is often neglected, chiefly because of a lack of available data. In this research, the dependence of load forecast errors on external predictor variables such as temperature, day type, and time of day was examined. The analysis was utilized to create statistically relevant instances of sequential load forecasts with only a time series of historic, measured load available. The creation of such load forecasts relies on Bayesian techniques for informing and updating the model, thus providing a basis for networked and adaptive load forecast models in future operational applications.

Essential tremor (ET) is a very common and problematic tremor disorder that affects over five million people worldwide. It is not a harmful or fatal disease, but because it can greatly impact the quality of life of older adults. Treatment for essential tremor varies depending upon the level of magnitude of the tremor, and high levels of magnitude could possibly prompt usage of deep brain stimulation (DBS). A device that could collect data could also double as a control for deep brain stimulation in more severe cases of ET. It is because of this reason that a diagnosing wrist device is being developed. Most cases of ET involve tremor in the upper limbs, and so a wrist-mounted or upper-arm-region device would be most logical. My goal was to successfully have a piece of hardware capable of graphing simulated (or actual) tremor in real time into a figure in the program MATLAB. The hardware consists of an inertial measurement unit (IMU) and an Arduino microcontroller hooked up to the computer. Wireless capabilities were also planned in the hardware of the final prototype. This device was proposed to be integrated into fabric of some type and fastened somewhere on the human arm for testing. The prototype developed is a glove with the externally attached sensors and microcontroller, and I was responsible for designing, testing and iterating plans of this device, as well as coding in Arduino and MATLAB to collect and graph data from the device in real time.

Biofuels are a promising means towards the goal of creating cleaner and more sustainable energy sources, but current biofuels are hampered by incompatibility with current infrastructure and difficulties with using food crops for production. This project addresses these problems with a novel approach to consolidated bioprocessing: genetically engineering a synthetic organism to digest low cost feedstock and produce alkanes, the main constituent of diesel fuel, which can serve as a drop in replacement to traditional petroleum. One challenge of consolidated bioprocessing is that cellulose, the primary feedstock, is far too large for bacteria cells to digest, so engineered strains must be able to break down the cellulose extracellularly. To alleviate this problem, the Klavins lab has implemented a cellular “division of labor” in E. coli which serves to generate specialized cell states: the consumer state and the altruist state. In the consumer state, the cell consumes nutrients and proliferates; in the altruist state, the cells produce large quantities of cellulase and a lysis protein to release the enzyme payload to the culture environment. Cellulase hydrolyzes cellulose into cellobiose, a molecule small enough to be imported and digested by both consumer and altruist cells. This project is focused on the addition of two key genes, previously shown to produce alkanes, into the “division of labor” strain. We extracted these genes using PCR and integrated them into the cellulolytic “division of labor” strain with Gibson assembly. We then identified and quantified alkane output using GCMS (gas chromatography / mass spectroscopy). We also tested the growth dynamics of our strains using cellulose as its sole carbon source through time-lapse cell counts. The results of this project can invite opportunities for industrial scale-up and can lead the way in producing other relevant high value biochemicals such as drugs, chemical intermediates, or other fine chemicals.

The ability to optically isolate single gallium arsenide (GaAs) electron donors has many implications both in quantum information science as well as semiconductor physics. However, conventional methods of fluorescence microscopy are fundamentally limited by diffraction, which sets a minimum resolvable feature size preventing the direct optical imaging of nanoscale structures. At the University of Washington Optical Spintronics and Sensing laboratory we are building a super-resolution optical microscope, based on ground state depletion, to isolate single GaAs donors. A critical component of the microscope is the two-dimensional methodical scanning of laser beams over the sample, for which we have elected to use a piezoelectric fast-steering mirror. I have designed a scanning system capable of under 1 nm mechanical resolution, which will be sufficient for our target optical resolution of about 100 nm to isolate single donors. I will present preliminary results on the performance of the test system using a transmission electron microscopy (TEM) grid as a test sample, which will allow us to evaluate the achieved accuracy and resolution of the imaging due to its regular features. Isolating single donors in GaAs using this super-resolution microscopy technique may lead to new insights in how to localize and evaluate impurities such as donors and acceptors for quantum information as well as classical applications.

Due to small position uncertainties within internal systems, it is difficult to accurately predict the movement of cable-driven robots. By studying what happens within the pulley-cable system, we will be able to propose a model that accounts for the specific properties of the system, such as friction, tension, and viscosity. This model can be used for more accurate control of robot movement.

A pulley board was built with similar mechanical parameters (same pulleys, cables, tensions, and dynamics) as the RAVEN II surgical robot of the UW BioRobotics Lab. The board was fitted with velocity sensors which measured the movement of the cables when predetermined forces were applied to the ends. Position, velocity, and acceleration functions were obtained from these data. A proposed model was formed from physical predictions of the cables, and by fitting the data from the pulley board into the proposed model, functions for a viscous damping coefficient and friction force were found. From the findings, the model was refined and applied to the RAVEN II, transforming sensor inputs into a more accurate estimation of the robot pose.

 The spins of electrons bound to donors are promising candidates as qubits for quantum information processing. The quantum properties of these donor-shy bound electrons are studied in high magnetic fields while immersed in liquid helium at 4K. For stability reasons, the level of liquid helium in the experiment must be monitored with a sensor. An effective way of implementing a sensor is with a superconducting wire. Super conductors have the property of having zero electrical resistance when they are cooled below a critical temperature. Above this critical temperature, the superconductor has significant electrical resistance. Using these characteristics, a superconducting wire can be used to construct a liquid helium level sensor. When current is run through a thin wire submerged in liquid helium, a relationship can be established between the voltage drop across the sensor and the amount of wire submerged in the liquid helium, allowing the liquid helium level to be measured. Results will be presented on the sensor design and performance.