This proposal outlines a one-year research program that continues an ongoing research effort supported by the Pittsburgh Digital Greenhouse on smart microchemical systems. The focus of the program is on demonstrating a prototype miniature methanol fuel reformer for in-situ hydrogen production in micro-fuel cells. Compact micro-fuel cells are currently being considered as alternative energy sources for portable power. The following four main components will constitute the integrated miniature fuel processor: (a) mixer/vaporizer; (b) catalytic steam reformer with Cu/ZnO composite catalyst; (c) water gas shift reactor assisted by a palladium membrane supported on a porous Cu/ZnO catalyst; (d) integrated resistive heaters/sensors and control electronics. These components will be microfabricated on a single silicon wafer thereby leading to an integrated "microplant-on-a-chip". The proposed research effort will focus on the issues of microchemical device integration; microchemical device characterization; thermal and energy management and optimization; and feedback control. Current projections indicate that micro-fuel cells will likely replace primary and rechargeable batteries in the near future, as indicated by the increased research activity on this topic in numerous corporate groups of microelectronics companies.

Magnetic Random Access Memory (MRAM) is based on the integration of CMOS technology with magnetically addressable thin film structures. This combination provides high speed, low power, high density, and nonvolatile memory. MRAM devices are, in fact, being considered as alternatives to DRAM and FLASH memories. The magnetic structures may utilize the giant magnetoresistive (GMR) effect or spin-dependent tunneling (SDT). In both cases, it is important to be able to reliably control the switching of the cell. This becomes increasingly challenging as the dimensions are scaled down to achieve higher densities. The purpose of this research is to exploit our extensive experience in micromagnetics and magnetic field management to explore alternative MRAM designs that will enable scaling to higher densities and data rates. While the primary goal of this work is to develop optimal MRAM designs, a secondary outcome will be the development of design tools for simulating the operation of MRAM systems.

The CHIPS Center at Carnegie Mellon University proposes a critical project to enable a new SoC storage technology that would radically alter mobile, network, and embedded computing. The CHIPS Center is pursuing a non-volatile, IC compatible mass storage vision centered on a MEMS based probe array. The idea is to store data in nanoscale magnetic marks on a uniform medium, using an array of several thousand probes. This whole structure will be compatible with CMOS transistors and fabrication methods, and will result in a one square centimeter chip storing up to 10 Gbytes, with sub millisecond latency, a 100 MHz data rate dissipating about a milliwatt per MHz, at a cost of about ten dollars. Successful development of this new type of non-volatile storage would dramatically affect SoC, embedded and mobile computers, set-top boxes, personal digital assistants, wireless phones, MP3 players, and digital cameras, by reducing costs, and enabling memory intensive functions such as voice recognition. The aim of the project is to reduce or eliminate nonlinearities and instabilities in the probe actuation system of MEMS based mass storage.

The proposed research is to develop a micro-fuel cell based power generation system for portable electronic devices. Our efforts will be directed primarily to realize this goal, a micro proton-exchange-membrane (µPEM) fuel cell that outputs power from tens of milliwatts to a few watts will be designed, fabricated and demonstrated. Proton exchange membrane (PEM) fuel cells deliver high power density, operate at low temperatures, and allow for fast start-ups and immediate response to changes in the demand for power. Hence they are ideally suited for potable electronic applications. Two different (µPEM) fuel cell structures and associated fabrication methods will be systematically examined and compared: one employs micromachined silicon as the cell substrate, upon which pre-fabricated electrolyte membrane/electrodes/backing layers assembly are stacked to form on-chip fuel cells. A stacked multi-cell structure can be fabricated to achieve high output power or voltage. The other design utilizes thin alumina ceramic wafers with micromachined micro gas channels as the substrate for fuel cell fabrication. This planar fuel cell arrangement can provide a hybrid energy source powering the on-chip electronic devices. Optimal down-selection and recommendation based on this dual-track research approach will be made at the conclusion of this project.

At Carnegie Mellon, the principal investigators have been developing design and fabrication technologies for the monolithic integration of microelectromechanical components with analog and digital CMOS circuits for use in remote integrated sensing systems. Several generations of integrated CMOS micromachined sensors have been demonstrated in the lab, however, two technological obstacles still remain to application of these sensors in the commercial world: complete tetherless operation and automated design customization. Tetherless communication requires power input for remote operation over long periods of time as well as data communication output of the sensed signal. Wireless telemetry enables dual use of the RF signals: for data communication and for dc power generation via rectification. Such inductively coupled power systems are in use in medical, RF ID and smart cards; however, complete integration in CMOS remains a bottleneck. Furthermore, CMOS remote integrated sensors involve custom design of the sensor, transduction into electronics, analog modulation circuitry, inductive antenna, as well as power rectifying and regulation circuitry. Once the wireless link is designed, it need not change; indeed standardization is crucial for use across a family of remote integrated sensors. On the other hand, each individual sensor can be customized to the desired application. As the sensor operation often depends on the analog readout electronics, simultaneous automated co-design of the sensor/electronics interface is needed. This proposal links the research at Carnegie Mellon with software developers at Neolinear, Inc. to attack these issues inhibiting the ubiquitous availability of custom remote integrated sensors. Sensor, circuit and design methodologies will be developed at Carnegie Mellon.

The principal investigator's recent research involved construction of high-resolution video mosaic from a video sequence generated with a video camera. The task may be divided into three phases. First, from the sequence of images, motion parameters between successive frames are estimated to construct a panoramic mosaic. Next, the Region-of-Interest (ROI) is selected for applying a superresolution algorithm. In this phase, preliminary result has been obtained by using an iterative back projection algorithm. Current and future research, which will improve upon the previous two phases already completed, is along directions summarized by 1) error in motion or displacement parameter estimation, 2) blur identification, and 3) channel noise and quantization error.

Web servers are tuned to perform well most of the time. Periodically, however, every popular web site experiences overload: a situation where more requests are arriving than the web server can service. The overload may last for a while or be transient. Either way, it doesn't take more than a few seconds, our research shows, for the damage to be done. Within just a second or two, the number of connections within the server skyrockets, quickly hitting the server limit, no matter how high this limit has been set. The server starts dropping client requests, clients experience timeouts, and client response times soar. Surprisingly, even when the period of overload ends, its effects linger. It takes a long time for the number of connections at the server to drop, in part due to TCP effects and in part due to the scheduling algorithm at the server. Unfortunately, while the number of connections at the server remains high, clients continue to experience high response times. Little work exists on coping with transient overload in Web servers. We propose a very different idea for handling transient overload. We propose, instead, unfair scheduling, in which priority is given to short requests, or those requests which have short remaining time, in accordance with the well-known scheduling algorithm SRPT (shortest-remaining-processing-time-first). Our prediction is that mean response time at the server will improve by a factor of at least 300%.

Electronic and digital consumer products are becoming increasingly more capable. They are also becoming more numerous. Technologies are being developed to support unified interfaces at the application, system and network levels. Implemented and fielded solutions are expected in the near future. However, the future of the user interface to such devices is far less certain. The capabilities that are designed (or could be designed) into wearable and home devices far exceed people's current ability to take advantage of them. For the benefits of advanced technology to become compelling to large segments of the population, our method of interacting with devices must be fundamentally re-engineered. Any solution must (1) reduce the cognitive load on the user, (2) consider interactive form factors, and (3) be mindful of mass manufacturing costs. Speech is uniquely suited to addressing these issues: It does not require any buttons, keyboards or display, and is thus suitable for devices of any size. Relegating recognition to an off-device processor further reduces its hardware cost. Finally, it works for a much wider range of people than typing, drawing or gesture. We thus propose to design, implement and test a universal (i.e. device-independent) interface for speech-based interaction with wearable and home devices. In a universal interface, the help structure, setup and configuration methods, device querying, navigation methods and confirmation/clarification dialogs will all be uniform across all such devices. This will lead to cross-device user skill transference and sharply reduced development effort for new device interfaces. This uniformity will also improve the recognition accuracy.