Self-Organized Physical Intelligence System
A new research project entitled "physical intelligence" has been launched at the Department of Mechanical and Aerospace Engineering at UCLA in Feburary, 2010. The goal of the project is to establish a universal physical intelligence theoretical model that explains how self-organized structures with intelligent functions spontaneously evolve in a complex system. The theory will unify a number of innovative theories in the fields of statistical mechanics, thermodynamics, complex systems, game control theory, and critical phenomena, etc. The theory will also be implemented in a self-organized physical intelligence system (SOPHIS) with two interacting sub-systems: a physical intelligence unit based on neuronal electronic circuit and a functionalized chemical reaction-diffusion system. The project will not only lead to fundamental understanding about various physical and biological self-organization systems and origin of intelligence, but also practical applications such as the creation of next generation of electronic circuits with intelligent behaviors, and dynamic interactions/control of biological systems. The multiple-million dollar project is supported by Defense Advanced Research Projects Agency (DARPA) and will last for three years. The project team is led by Prof. Yong Chen at the Department of Mechanical and Aerospace Engineering at UCLA, and also the research groups led by Prof. Chih-Ming Ho at UCLA, Dr. Tad Hogg at Complex System Consultant, Dr. Richard Rohwer at Stanford Research Stanford Research Institute International, and Prof. Jeff Shamma at Georgia Institute of Technology..
Bio-inspired Intelligent Sensing Materials for Fly-by-Feel Autonomous Vehicle
Knowing the state of an aerospace vehicle in real-time is crucial for maximizing its performance, assuring its reliability, and completing missions successfully missions. This is especially true in harsh or combat environments. Despite the importance of vehicle awareness, however, the current state of the art in aerospace vehicles is primitive as well as prohibitively heavy, expensive, and complex to scale to an entire aircraft. A departure from traditional technology is necessary to increase the performance and capabilities of next generation aircraft, particularly UAVs. In this Multidisciplinary University Research Initiative (MURI) project led by Professor Fu-Kuo Chang at Stanford University, we plan to develop innovative multi-scale design, synthesis, fabrication and integration techniques for building and functionalizing intelligent sensing aerospace materials intended for large-scale applications. We plan to invent a new approach to integrate materials, sensors, electronics, signal/data processing and algorithms, as well as a multi-scale fabrication technique that will allow devices in nano/micro scales to be integrated into macro-structures and materials. Specially our group plans to develop integrated synaptic circuits that enable an unprecedented neurologically inspired system with signal parallel processing, real-time pattern recognition, adaptive learning, thereby breaking the paradigm of conventional monitoring and control systems based on preprogrammed computers. The circuits will be based on synaptic transistors, and integrated with Si-based electronic circuits. Neurologically inspired theoretical models and architectures will be directly integrated and applied to establish the circuit architecture. The integrated synaptic circuits can (1) promptly process a large amount of signals in parallel to recognize exogenous threats accurately and effectively, (2) implement real-time learning autonomously, and (3) provide a prognosis for appropriate response. This project is supported by Materials and Microsystems Division, US Air Force Office of Scientific Research (AFOSR).
Fabrication of 3D Nanocircuits
The rapid downscaling of Si-based circuits is accelerating the introduction of new nano-patterning technology and nano-engineered materials. Although a wide range of nano-patterning techniques has been developed, they all have different intrinsic problems and limitations. Optical lithography, such as extreme UV and X-ray lithography, will face tremendous difficulties as it approaches the ~22 nm critical dimension. Moreover, their optical projection systems also become expensive (at > $50 M/tool). Although scanning lithographic techniques such as electron beam lithography (EBL) and scanning probe (AFM/STM) lithography can potentially reach ultra-high resolution (<10 nm), these serial processes are limited by their slow production rates.
The recently developed method of nanoimprint lithography provides a new avenue for nano-patterning with high resolution, high throughput, and low cost. It has been listed in the ITRS as one of the future technology candidates for nano-manufacturing. Previously we have fabricated nanoscale devices and crossbar circuits by nanoimprint lithography. The major challenges for the imprint lithography are to reduce defect densities and improve yield. In this project we plan to develop a nanoimprint lithographic process that can fabricate 3D overlay crossbar circuits which will be high-defect tolerant and also self-aligned. This is important because high defect count and overlay are two of the key challenges for nanoimprinting. These crossbar circuits have already been proposed in 2-D to complement curcuits by providing a high degree of programmable interconnectivity between logic cells. Such networks are highly defect-tolerant, as defects can be routed around efficiently.
Neural Electronic Materials, Devices, and Circuits
Over six decades, modern electronics has evolved through a series of major developments (e.g., transistors, integrated circuits, memories, microprocessors) leading to the programmable electronic machines that are ubiquitous today. However, owing both to limitations in hardware and architecture, these machines are of limited utility in complex, real-world environments, which demand an intelligence that has not yet been captured in an algorithmic-computational paradigm. As compared to biological systems for example, today's programmable machines are less efficient by a factor of one million to one billion in complex, real-world environments. We plan to create a new materials, devices and circuits that can break the programmable machine paradigm and define a new path forward for creating useful, intelligent machines.
A neurological system requires a tremendous amount of synapses (an estimated 1014 in the human brain) for signal processing, memory, and learning. Si electronic circuits and/or programmable computers have had very limited success to date in the field of neurological control systems for one primary reason: the lack of low-cost and low energy consumption circuits that can emulate the essential properties of synapses. We are developing devices that can have the functions of synapses and are integratable with conventional Si MOS transistors. The devices can be reconfigured flexibly in analog mode by input signals. It integrates logic, memory, and learning functions that can fully emulate biologic synapses while only occupying a tiny fraction of the size of the smallest CMOS representation of a synapse; moreover it is a nonvolatile passive device and thus requires little power to operate.
We plan to explore electronic configurable ionic conductive materials, such as conjugated polymer and perovskite manganite, to achieve desired synaptic characteristics. The neural circuits will be built on a large number (>106) of low-power consumption, low-cost, and high-speed synaptic devices integrated with CMOS circuits, which enables the circuits to dynamically process signals from a large amount of distributed sensors (>108) in parallel, and make real-time dynamic interactions and decisions to reconfigure the sensors and actuators. We will design the circuits with dynamic reconfiguration and adaptation capabilities similar to biological neuronal circuit architectures. The sensory information from the sensors will be spatially mapped to the collective model in the circuits and aggregate locally to produce a topographical mapping of the sensory information. A multi-layer cortical network will provide the necessary processing functions and ultimately result in a robust control strategy for the actuator system.
Single-Molecule Biomarker Detection Platform for Mycobacterium tuberculosis (mTB)
Currently available diagnostic approaches for tuberculosis are limited by low sensitivity and time factors. The present gold standard for diagnosis is isolation of positive culture from clinical specimens, which generally takes up to 8 weeks for results. Unfortunately, even with rapid culture techniques, the results take at least 1 week to return, which increases the difficulty of identifying infected patients and ensuring that they follow-up for treatment. Nucleic acid amplification techniques are limited by their sensitivity; they are presently widely recommended only for AFB-smear (+) patients. The EliSpot assay, which detects IFN-g producing T cells specific for mTB, cannot reliably distinguish between active and latent TB given that the test is performed on peripheral blood samples. Therefore, rapid, highly sensitive tests are desperately needed in order to identify patients infected with TB and limit the transmission of this potentially fatal infection.
We have developed an innovative “lock-in” fluorescence assay that can detect a single biomolecule with ultrahigh sensitivity and specificity. In the “lock-in” fluorescence assay, we designed two probes that can bind to the target molecule. For example, probe 1 is a DNA aptamer grafted on an Au nanoelectrode and can capture the target protein. The target will bind with the second probe labelled with fluorophores. When an external electric field is applied on the Au electrode, the probe1-target-probe2 complex can be attracted towards or repelled from the Au electrode. Due to the quenching effect of the Au surface to the fluorophores, the fluorescent signal can be modulated by the electric field. The molecular modality can be unequivocally correlated with the modulated fluorescence, which enables the modulated specific fluorescence from a single target biomocule to be unambiguously distinguished from background noise and nonspecific fluorescence. With the improvement of signal-to-noise ratio by the modulation, the technique has enabled the detection of target molecules with the sensitivity down to the single molecule level.
The overall goal of the proposed project is to develop an ultrasensitive, reliable, rapid, simple, and economically feasible single-molecule biomarker detection platform for TB surveillance. As for the proof of principle during the first year, we plan to reach the following milestones in the project: (1) The proposed platform can detect mTB biomarkers (such as genes) with an ultrahigh sensitivity down to single molecule level without PCR. (2) The platform can detect mTB DNA biomarkers from the sputum samples of infected human patients. After we successfully test and demonstrate the platform within the one-year duration of the project, the platform can be incorporated into the TB surveillance analytic system, and also has the potential to be applied in the area of TB biomarker discovery. The platform provides a practical and affordable solution for TB surveillance, and can also effectively improve the treatment for the TB patients most in need within the developing countries in the future.