An Adaptive Magneto-Optic Thin-Film Based Architecture
For Producing Self-Structuring Sensing and Recognition Devices
Martin Dudziak*, a, Andre Chervonenkisb, Vladimir Chinarovc
aSilicon Dominion Computing, Inc. & Moscow State University
cSilicon Dominion Computing, Inc.
This is an abstract
for a presentation given at the
Foresight Conference on Molecular Nanotechnology.
There will be a link from here to the full article when it is
available on the web.
We describe an architecture for producing arbitrary-sized sensor/processor devices capable of acting as sensing and recognition elements, individually or in an array, for a variety of different physical states and conditions that can be characterized magnetically or optically. The architecture is aimed at enabling a new class of pattern recognition and testing to evolve that depends less upon conventional digital semiconductor and microprocessor technology with concomitant conventional algorithms for image processing and/or other forms of object recognition (neural networks, fuzzy logic) that must be executed in one or multiple computing systems based upon traditional computing architectures. The new offers more opportunity for employing variable geometries of extremely small, customizable, redundant, and even disposable sensor units that make use of magnetic and optical field effects to trigger events that in turn are translated into higher-scale information objects for input into conventional microprocessor systems.
The heart of this work, based upon converging research efforts in magneto-optic thin-film chemistry and materials engineering, nanoengineering and scanning probe fabrication, and adaptive pattern recognition, rests in the ability to measure and to generate extremely fine-grain magnetic domains and optical patterns using a family of Fe-Ga thin-film crystals developed for this purpose. The approach toward developing devices capable of producing high-resolution images of magnetic fields and sensitivity to magnetic field of the order 10-6 to 10-8 G is based upon a field visualizing film (FVF). This consists of a transparent ferromagnetic layer of Bi-substituted iron-garnet grown by LPE technique on a non-magnetic substrate. The composition of the FVF is characterized by the formula (R Bi)3 (M Fe)5012, where R is a rare-earth ion (Y,Lu,Tm for example) and M is generally Ga or Al. The specific Faraday rotation of 104 deg/sm and absorption coefficient less then 103 cm-1 are available in a generic composition (Tm Bi)3 (Fe Ga)5012. The role of Bi substitution provides for a high Faraday effect and ample transparency aiding in the creation of high contrast domain structure patterns caused by domain wall shifts due to magnetic fields normal to the FVF plane. These domain structures can be easily observed using a polarizing microscope.
Two types of sensor technology had been tested and developed and both employ similar thin-films but with a difference in the application. The first class of device was directed at the detection and measurement of surface flaws and defects in materials without the use of high frequency eddy currents as have been employed traditionally in non-destructive testing of materials. This class of device has been used successfully with two forms of magnetic field enhancement depending upon the application and the nature of the observed material. The second class of device was directed at the detection of intensity and localization of very weak (<= 10-8 G) magnetic fields within a material including biological substances and claims to offer some alternatives to both traditional magnetic force microscopy (MFM) as well as SQUID-based or magnetostrictive based magneto-encephalography (MEG) and related techniques in medicine.
The outgrowth of this work in magneto-optics led to a design enabling the MODE (Magneto-optic Detection and Encoding) thin film, as it is named, to be incorporated into a working design for spatial light modulation and construction of adaptive neural-like networks with submicron domain sensitivity as well as extremely fast (@ 103 m/s) domain shift velocities. Combined with the sensitivity magnitude of the thin film this offers a critical building block for implementing sensor configurations of arbitrary geometries that can be used to react to particular magnetic domain configurations or to particular dynamics in magnetic fields in a sensed sample. Optical configurations can be processed by an intermediary layer of devices that transforms the optical pattern (e.g., hologram) into a magnetic pattern of sufficient local intensity and duration to be processed by the MODE device. However, the attention of this research program has been upon the sensing and registration/recognition of very-low-strength magnetic fields.
The magneto-optic sensor/switching array produces in turn as its output a pattern that can be optically transmitted through a series of switching elements configured as an asynchronous multi-layer array of non-linear oscillators, operating as a biological-like neural network, to produce digital output trigger-functions that are correlated to specific observable and detectable states or conditions in the sample environment, be in one singular sample of micron-scale or smaller, or a large configuration that may cover a large surface, such as on an engineering component or machine. Ultimately these trigger-functions would in a working application interface with conventional digital processing hardware and software and perhaps thereby to some process control or mechanical systems that would act upon the sensed and recognized (categorized) information.
The current research effort has focused upon the refinement of the magneto-optic thin films and the development of experimental devices for testing spatial light modulation and optical computing interfaces, the refinement and testing of the pattern recognition logic (known as Holographic Associative Learning), and furthering the work on synchronizing and stabilizing the oscillator networks that are critical to processing the output from the magneto-optic sensor and switching arrays. This work is jointly funded by two private companies and is a continuing present effort.
[We do not at present have suitable digital (JPG, GIF) images or figures of the current project work to submit but we include the following images as substitutes in support of the abstract text, since an extended abstract with figures was requested. The first two figures (1, 2) show, in purely digital PC-based software, snapshots of the progressive performance of the HAL (holographic associative learning) algorithm for reconstruction and identification/matching of a very badly distorted and noisy image, in this case of a human face, using previously trained / learned data sets. Something quite similar is being employed within the current project. The next figure (3) shows a software tool, Verite, that is being developed as part of the project for the comparison and analysis of patterns processed through the system. The next two figures show actual images produced from a working prototype of a magneto-optic scanner with video output that was used on (4) a banknote and (5) a magnetic barcode strip. The final two images (6, 7) illustrate results from software simulation of synchronization control processes with multilayer oscillator networks that are brought into control by external periodic control signals, an aspect of the work effort referenced in the abstract above regarding the processing of output from magneto-optic based switching arrays that are producing input from sampled events/fields. These images will not be the actual images used in the final paper but are presented here in the manner of providing experimental proof-of-concept.]
Martin Joseph Dudziak, PhD
Silicon Dominion Computing
3413 Hawthorne Avenue, Richmond, VA 23222-1821 USA
Phone: (804) 329-8704; Fax: (804) 329-1454
E-mail: firstname.lastname@example.org; Web: http://www.silicond.com/mdcvprof.html