Microelectronic Basis for Pattern Recognition Systems
Research and Development of the Neuron-Like Microelectronic Component Basis with Data Compression for fast Pattern Recognition Systems in Experimental High Energy Physics, Atomic Stations Robotics, Interactive Computer Systems (NEURON)
Tech Area / Field
- INF-ELE/Microelectronics and Optoelectronics/Information and Communications
- INF-OTH/Other/Information and Communications
3 Approved without Funding
MIFI, Russia, Moscow
- NPO Molniya, Russia, Moscow\nAll-Russian Research Institute of Automatics, Russia, Moscow
- GSI, Germany, Darmstadt\nLawrence Berkeley National Laboratory, USA, CA, Berkeley\nUniversity of California, USA, CA, Berkeley\nBrookhaven National Laboratory, USA, NY, Upton\nUniversity of California, USA, CA, Los-Angeles\nOptima Handels- u.Konstruktions Gmbh, Germany, Arnoldstein
Краткое описание проектаThe main planned scientific result within the limits of the project is the developing of a new neuronlike component basis with high effectivity for fast and reliable data processing and pattern recognition systems.
The commercial importance of the project is defined by the planned developing of industrial versions of certain pattern recognition systems on the basis of existing today following laboratory versions of such systems:
· Trigger system for high energy physics international experiments capable of fast 3D tracks processing and recognition.
· Important friction units monitoring and diagnostic system for atomic power stations on the basis of acoustic emission analysis.
· PCbased software and hardware complex for 2D image processing in laser measuring system for precise monitoring of space position and shape of modern high energy physics detectors.
· Scanning probe microscope for surface mechanical properties and topography investigations with nanoscale resolution.
· Medical system for 3D reconstruction of cavity behavior and blood flow structure (velocity field) in the transport segment of circulation system.
· Set of interactive computer systems for speech processing, finger prints processing, financial analysis, planning and forecasting.
Later a serial production of the mentioned above systems and neuronlike components designed is planned to be carried out according to orders of Russian and foreign scientific centers, institutions and universities. The demand on such systems and components is confirmed by the management and specialists of Russian and foreign leading scientific centers: Central Scientific and Research Institute of Control, Economics and Information of The Atomic Energy Ministry of the RF – (Moscow, Russia), Acoustic Institute named after N.N.Andreev – (Moscow, Russia), Scientific and Technological Center "Superhard Materials" (Troitsk, Russia), GSI (Darmstadt, Germany), Lawrence Berkeley Laboratories (USA) and others.
So, for instance, in accordance with the international project "STAR" (BNL, LBL, USA) supposed is development and delivery of fast neuronlike multichannel laser image recognition system for TPC detector.
The development of the similar system is planned for the international project "HADES" (GSI, Germany).
In collaboration with the Russian Scientific Center of Cordial Surgery named after A.N.Bakylev (Moscow, Russia) and laboratory INCERM306, University Bordo2 (Bordo, France), supposed is the development of medical system for 3D reconstruction of cavity behavior and blood flow structure (velocity field) in the transport segment of circulation system.
In collaboration with the company HTE (Zelenograd, Russia) it is planned to develop a new version of scanning probe microscope for surface mechanical properties and topography investigations with nanoscale resolution and neuronlike image processing for Fraungofer Institute of Semiconductors Technology and Integrated Circuits (Erlangen, Germany).
In collaboration with the Summer Holdings Inc. (London, Ontario, Canada), OPTIMA Handelsu. Konstruktions GmbH (Arnoldstein, Germany) and Algontex Trading Inc. (Nepean, Ontario, Canada) it is planned to develop a new version of fast laser projection system with the hardware neuron nets image processing.
In accordance with the agreement with STAR Research Inc. (Ney York, USA) supposed is the joint development, production and marketing of computerbased system for human psychophysiological parameters processing and recognition.
The approach to neuronlike component basis design and recognition systems development proposed in this project will differ both from the one of first of all electronic firms, not directly engaged in nuclear research centers (Intel Corp., Ampteck Inc.), and from the one of nuclear research centers (for example, LBL, GSI, SLAC, BNL).
The first it will differ from in the possibility of higher orientation towards a specific pattern recognition task. From this point of view the main advantages of the proposed neuronlike component basis are:
· More high functional possibilities (the possibility to realize neuralnet with lager size and with higher speed) due to the exploitation of data compression methods and optimum data presentation forms in each neuron layer.
· More high specialization, orientation on the 2D and 3D pattern recognition systems design and hence a more preferable parameter aggregate.
· The possibility to change the neuralnet structure in order to achieve more high task processing effectiveness.
· The possibility to perform structural and parametrical adaptation of neuralnet on the set of examples.
The second it will differ from in an employment on the first step programmable logic (FPGA) and a special semicustom arrays oriented on realization of neural nets with data compression on the second step of system design, what offers at equal other conditions the advantage of shorter design and manufacture times.
The problem states is of complex character and supposes not only a creation of specialized neuronlike microelectronic component basis, but also developing certain pattern recognition systems with it help as well.
It is supposed that mentioned above approach will be realized mainly by the personnel of the following three institutions:
· MEPhI – conception development of the neuronlike microelectronic component basis and pattern recognition systems, problem oriented library creation, schematic and structural design of the programmable logic and semicustom arrays, design and manufacturing of the PCbased neuralchips testing set.
· VNIIA – design, developing and testing of the important friction units monitoring and diagnostic system for atomic power stations on the basis of acoustic emission analysis.
· SE PA MBP "Molnia" – circuit and mechanical design of hybrid microasseblies and PCB units employing a set of neuronlike components, their preproduction, testing on the measuring sets, participation in testing on actual physical research installations.
Participating in the project are foreign collaborators from the physical research centers GSI (Darmstadt, Germany), and LBL (Berkeley, USA). Their role consists in an experts evaluation of the works and a following study of the designed and manufactured in limits of the project samples of the neuronlike components and pattern processing and recognition systems.
The qualification of the scientists and engineers being ready to participate in the project and render it not less than 10% of their working time is rather high. Many of them are in Russia the leading specialists regarding matters of electronics and microelectronics, neural nets, CAD and etc.
The project is important from the viewpoint of an alternative peaceful employment for scientists and engineers engaged in design and manufacture of armaments. The project realization promotes reorientation of a part of the VNIIA's, SE PA MBP "Molnia"'s and MEPhI's scientific and manufacturing potentials, specialized in military equipment design and radiation hardness problems, toward investigation and design of equipment's for physical research, nuclear reactor safety provision. The works at the project realization will employ during 3 years tens of specialists, having a many year experience in design, implementation and application of components and units, used in military equipment and installations. More than 23 of the total work's volume (in mandays) will be carried out by scientists and engineers, having experience in armaments design and manufacture.
All planned works in limits of the project are based on the developed and existing today at MEPhI laboratory samples of neuronlake pattern processing and recognition systems. Supposed also is patent protection of all original project designs.