by Nicholas G. RAMBIDI, Dr. Sc. (Chem.), M. Lomonosov Moscow State University

Nanotechnology of today largely determines the ways of further progress in computer engineering. Continuous improvement of digital computers is putting an end to miniaturization of their primitives, when individual molecules are used as switching elements. At the same time, the implementation of increasingly complicating computational problems requires the elaboration of fundamentally new devices realizing biological principles of information processing. Just based on these systems, efficient solution of the machine intelligence problems can be achieved.

Possibilities of using natural microscopic objects or designing their artificial analogs capable of performing certain macroscopic actions have been actively discussed by research community beginning from the mid-20th century. One of the first scientists in this quest was virtually Ervin Schrodinger - outstanding Austrian physicist, Nobel Prize winner in 1993 (foreign honorary member of the USSR Academy of Sciences since 1934), who offered in 1943 the idea of aperiodic crystal - a system at the molecular level capable of storing enormous genetic information. However, it was prominent American physicist, Nobel Prize winner in 1965 Richard Feyman, who should be called a genuine prophet of nanotechnology. He put forth a statement (fantastic for that period) that there are no physical restrictions for recording the entire Encyclopedia Britannica on a pinhead and creating various engineering devices, including computers, at micro- (nano-)level.

Thirty years later, in the 1990s, it was exactly nanotechnology* that became one of the main instruments promoting intensive developments in a variety of human activity fields. It turned out that consistent miniaturiza-

* See: V. Bykov, "A Microscope Scans Atoms", Science in Russia, No. 4, 2000; V. Bykov, "Russia's Nanotechnological Potential", No. 6, 2003; B. Kolbasov et al., "New Nanomaterial", Science in Russia, No. 1, 2005; V. Sleptsov, M. Dantsigher, "Nanostructures of New Quality", Science in Russia, No. 2, 2005. - Ed.

Articles in this rubric reflect the authors' opinion. - Ed.

стр. 38

Molecular rectifier of Aviram and Ratner.

tion of devices and their individual elements, up to nanometer dimensions, resulted in the appearance of new, often unique, properties. It became clear also, that it is possible to control the properties of a macroscopic object by means of directional change in its structure at micro level.

Within a short period, miniaturization has become an urgent need for industry. This was largely the result of rapid development of semiconductor computation technology that has been moved from transistor-based printed circuits to micro circuitry. Their micrometer dimensions achieved currently are already close to limits tolerated by fundamental physical principles. These insuperable limitations made it necessary to search new approaches to constructing digital integrated circuits and their physical realization that would be conceptually different from those dominating today, i.e. to the development of new primitives. Most active were attempts to use natural switching systems-individual molecules or systems built on their basis.

The first real proposal of this kind was put forth in 1973 by Americans Ari Aviram and Mark Ratner. They considered the process of electron moving through a molecule representing a combination of its two fragments: the first-with the properties of an electron donor, the second-an acceptor dopant*. The fragments are connected with a grouping through which, like a tunnel, electrons could pass. Having placed such molecule between two electrodes, these researches showed that with one of the directions of an applied field, this device behaves as a unidirectional electron conductor.

This work gave rise to a flow of proposals abroad and in Russia concerning the use of molecular systems in electronics engineering. However, after the initial period of euphoria, the hour of disillusion struck. It became clear that by no means each molecule is capable of serving as an electric signal switch that might respond in a unique manner to an applied stimulus. Molecules are essentially quantum objects, and the processes in them are of probabilistic nature. Therefore, an attempt was made to use a great assemblies of molecules-a microscopic volume of matter, rather than individual molecules as elements of an electronic circuit.

Experiments carried out in this field have resulted in creation of working prototypes of molecular storage devices. Probably, most applicable for practical use have become the versions of random access memory for computers developed in the Syracuse University (USA) by Robert Birge in the late 1980s-early 1990s based on unique protein: bacteriorhodopsin. Its molecules (each of them is a cyclic combination of seven polypeptide spirals containing a photosensitive fragment-chromophore) form a photosynthetic center of halophilic bacteria (Halobacterium Halobium). While absorbing a light quantum, this protein acts as a proton pump promoting the synthesis of adenosine triphosphate (ATP) - a nucleotide which (in all living organisms) acts as a universal accumulator and transmitting agent of energy. With absorption like this, a molecule undergoes structural transformation.

At the same time, bacteriorhodopsin, as a protein, features a unique stability. It is capable of preserving

* Acceptor dopant - (in semiconductors) a dopant atom, which may "trap" electrons from donors' valence band thus forming holes participating in electrical conductance. - Ed.

стр. 39

Structure of the bacteriorhodopsin molecule, its photo-cycle and principal scheme of the random access memory based on this protein.

Switching of rotaxane molecule states (A), rotaxane molecule (B) and memory device (C) developed on its basis.

unchanged properties for many years in the form of dry powder or polymer films. A fundamental property of its molecule consists in its photochemical cycle: after absorbing a quantum of light, it passes through a series of excited states and takes inadvertently its initial form. In this case, optical characteristics of protein are changed. Thus, the natural bacteriorhodopsin in conditions of indoor temperature behaves as a photo chromic medium with a fast time of information storage (not exceeding microseconds). At the temperature of 77 °K, the photo cycle breaks, and the molecule acquires properties of a system with two stable states. Transitions between them could be initiated by visible light in the range of rather broad absorption bands with maximum values at 540 and 412 nm. Using this phenomenon Birge's team designed the random access memory of 25 Mbytes in capacity. To continue this work they started creating a volume memory that would act at indoor temperature.

Estimations show that in about 3 cm³ of polymer information of hundreds of gigabytes can be recorded, stored, and also read out with the use of laser beams. However, elaboration of such system is a most complicated engineering challenge. Therefore, until now, basic solutions are tried out on service models with storage capacity of 1 to 2 kb.

Development of devices based on bacteriorhodopsin has made closer the industrial use of molecular media in computer engineering. However, a genuine breakthrough took place at the very end of the 19th century due to the appearance of unique systems - rotaxanes. They are built of fragments devoid of chemical bonds and capable of moving relative to one another. These molecules have two stable states changing from one state to another by means of electric signal.

In storage device developed jointly by the research lab of Hewlett-Packard Company and California University in Los Angeles (USA) use is made of the so-called "crossbar" architecture simplifying its manufacture, and "bottom-up" construction principle. In a traditional technology based on an opposite "top-down" principle, the formation of an integrated circuit (microchip) starts from thin layers of material on the surface of a

стр. 40

Structure of human cerebral cortex (A), scheme of McCallock-Pitts formal neuron (B) and elementary neural network (C).

Various realizations of the distributed reaction-diffusion systems.

work piece with subsequent formation of a preset pattern on them and eliminating unnecessary parts. But while using the "bottom-up" approach, the first thing to do is to create a system of parallel electrodes - extended metal threads. Their thickness in a micro-miniature device should be minimal - to 10 nm. Then a monolayer of oriented rotaxane molecules is formed on the work piece. Applied on top is the second system of electrodes, which are perpendicular to initial ones. Molecules of rotaxane contacting both systems (their quantity in the monolayer ranges from several pieces to tens) act as switching elements. This model was used for the elaboration of the working prototype of a random access memory chip with the density of 7 10⁹ bit/cm², formed by blocks of 64 bits each, and the density of elements in each state amounted to 5 10¹¹ bit/cm². The figures are rather impressive, since in devices manufactured presently this density does not exceed 10⁸ bit/cm².

New approaches have given rise to revision of former concepts. It turned out that single molecules, and not only their assemblies, can be the elements of computation devices. Thus, systems based on chiropticene will probably feature high-speed performance. This optically active model was designed in the USA (CALMEC company) in the latter half of the 1990s as an element, which is switched over very fast-within femto-seconds (1 fs=s^-15 ) by a simultaneous action of light emission and electric field. Designers are planning to use chiropticene as a base of storage devices of super high capacity and to develop a prototype in which volume of 1D³ (25.4 mm³) up to 16 Tb of data will be stored. New architecture has been already developed for this device, which will make it possible to read out or record data by packages with a capacity of about 1 Mb at an exchange rate to 2000 of such operations, i.e. 2 Gb for 1 s.

It should be mentioned that the quest in the field of molecular electronics has been actively carried out also

стр. 41

in this country since the early 1980s. Research Institute of Physical Problems under Ministry of Electronic Industry, institutes of the USSR Academy of Sciences and some higher educational establishments were involved in this work. These efforts were coordinated by the Interdepartmental Council for Molecular Electronics problem under the Commission of the USSR Council of Ministers and USSR State Committee for Science and Technology under the chairmanship of Academician Yevgeny Velikhov. A powerful infrastructure has been created that made it possible to implement a whole number of most important tasks. Actively participating in this work were Academician Andrei Mikaelyan, President of the Estonian Academy of Sciences Karl Rebane, other leading scientists of the country.

Computer science of the second half of the 20th century was developing based on von Neymann's paradigm*. Its realization made it possible to create efficient means for solving most urgent engineering challenges and update them successfully up to the present moment. However, as early as in 1943, scientists of the Northwestern University of the State of Illinois (USA) Warren McCallock and Walter Pitts proposed a neural-network approach to this problem based on the data (known at that time) on cerebral cortex structure. The network of nerve cells in the model is essentially a system of processor elements-formal neutrons. Each of them obtains a positive or negative signal from all the others, processes this information and transmits the results to the rest. They perform processing simultaneously, i.e. with a high level of parallelism, which is very far to be reached by today's semiconductor multi-processor computers. A neuron summarizes entering data algebraically (i.e. in consideration of signal sign) and transmits the summarized signal to the network if its value exceeds the preset threshold value. Such mechanisms of data processing are nonlinear. In contrast to von Neymann's computer, the transfer from solving one task to the other is determined here not by a program being entered, but by the initial states of neurons and network structure. As soon as the starting characteristics for it are preset, it is evolving in time. Its final state is the solution of a selected task.

* John von Neymann (1903 - 1957) - American mathematician and physicist. In mid-1940s he put forth a paradigm on whose basis digital computing machines were designed. Among its main principles are the following: all computations should be implemented in a binary system, computational operations should be serial, a program should be stored inside the machine. - Auth.

стр. 42

Main operations of processing black-and-white images by Belousov-Zhabotinsky media.

Beginning from the 1980s, the problems of artificial intelligence* are becoming increasingly topical. They embrace: identification of objects, scenes and situations, selection of optimal solution of complicated multifactor logical tasks, a number of other, no less important tasks. The necessity of increasing the size of computer memory and rate of elementary operations performed has become an urgent challenge. The development of McCallock's and Pitts' ideas has become an alternative to "computational complexity of a task-computer efficiency" race. It resulted in the revival of neural network concepts and development of neuro-computers.

Unfortunately, because of fundamental differences between von Neymann's and McCallock-Pitts' principles, digital semiconductor technologies turned out to be low efficient in creating complicated trainable neural networks. Therefore, nanotechnological principles of controlling object properties at the expense of directional designing of its structure at micro- and nanolevel, gave rise to the attempts of applying different alternatives of these networks physical implementation. One of promising alternatives is related to the attempts of building them on the basis of the so-called distributed reaction-diffusion media. What do they correspond to?

The model of McCallock and Pitts describes one of the probable ways of realizing the distributed dynamic systems. In a general case they are essentially spatially extended media. In each of their elementary volumes physical or chemical reactions take place, which determine a reaction to external influence. The degree of parallelism here is incomparable with the possibilities of computations based on digital data processors. Elementary volumes are bound by a feedback system, and their relationship that shows up in diffusion, causes a situation that the dynamics of a medium, its behavior as a whole, is much more complicated than in microvolumes, i.e. individual cells. (Such formations may well be illustrated by communities of ants or bees demonstrating complex behavior despite the simplicity of actions of an individual species, and bacterial colonies: their participants are grouped spontaneously in such a way, that quite nontrivial spatial structures of round or spiral shape are generated.) So, reaction-diffusion media are similar in architecture with neural networks, where microvolumes act as elementary processors-neurons.

In the nature, the media of our interest are found at different levels of the structural organization. At the level of body tissue these are first of all the mechanisms of realizing the functions of cerebral cortex. They are also found in chemical and biochemical systems, in biological membranes and cells, i.e. at the supramolecular level. And finally, nonlinear intermolecular mechanisms may lead to collective excitations-soliton waves extending to great distances along the molecular framework.

A remarkable property of these media shows up in the fact that regardless of a physical realization they manifest the same macroscopic forms of behavior: local concentration oscillations or oscillations in the whole volume of the medium; spreading narrow concentration pulses or extended fields of medium change from its one state to another, formation of time-stable nonuniform spatial distributions of component concentrations (so-called dissipative structures).

Most promising systems for creating neuro-like means of data processing are chemical reaction-diffusion systems and, first of all, those in which reactions of Belousov-Zhabotinsky** take place. What is their es-

* See: S. Shumsky, "Trying to Be Clever", Science in Russia, No. 1, 2002. - Ed.

** Reaction of Belousov-Zhabotinsky - a chemical reaction in which chaotic self-oscillating processes take place. It was discovered in 1951 in the USSR by chemist Boris Belousov and in the 1960s - studied by physico-chemist Anatoly Zhabotinsky. Promoted the appearance of a new science studying self-oscillating processes. - Ed.

стр. 43

Evolution of a half-tone image in the Belousov-Zhabotinsky medium.

sence? Formally it is oxidation of an organic compound (usually methane-dicarbonic acid) by some non-organic one (sodium or potassium bromate) catalyzed by ions of transition metal (cerium, ferrum, manganese, ruthenium). Externally the process looks very effectively, since in the course of the reaction, solution color changes periodically. In reality, it is an aggregate of interlocking intermediate reactions, and their exact number has not yet been determined.

A surprising feature of Belousov-Zhabotinsky media consists in the fact that their dynamics is similar to the so-called specialized neural networks proposed in the 1970s by American mathematician Stephen Grossberg for simulating individual functions of human cortex. Therefore, media of our interest make it possible to simulate complicated logical processes, such as processing and recognition of images, search for the shortest way in an arbitrary labyrinth and a number of others. The very idea of using them for data processing was put forth in 1989 by German physicist Lotar Kunert, Valentin Krinsky, Dr. Sc. (Phys. & Math.) and Konstantin Agladze, Cand. Sc. (Phys. & Math.) (both are from the Scientific Center of Biological Studies of the USSR Academy of Sciences in Pushchino). Later the information capabilities of Belousov-Zhabotinsky media were studied in detail by the author of this article with the workers at the International Research Institute for Management Problems working under the methodological guidance of the Russian Academy of Sciences and at the Physics Department of the Moscow State University.

These media present advantageous initial material for the elaboration of devices for data processing. They are stable and non-toxic. Temperature range and time scales of processes in them are convenient for registering the characteristics with the use of relatively simple physical methods. Chemical components necessary for the formation are accessible and their price is not high.

The catalyst of Belousov-Zhabotinsky reaction changes its electronic state in the course of intermediate chemical transformations. Consequently, the medium changes the color (from red to blue and vice versa). Thus, it is easy to visualize the process and record the spatio-temporal evolution of the system with the use of a video camera into computer memory for further processing. Most convenient are media in which a photosensitive complex of ruthenium is used as a catalyst. In this case information being entered is essentially an image

стр. 44

Processing of multi-view aerial imagery by Belousov-Zhabotinsky media.

(intrinsically - a preset distribution of light intensity) projected by an optical system on the surface of a layer or into the medium volume. Under the action of radiation, the photosensitive catalyst initiates photochemical reactions that cause the change in the content of its basic components. As a result, in all points of the medium changes in their concentrations occur that are determined by the intensity of light radiation in it, i.e. a chemical image of the input data appears.

Transfer, storage and processing of information with the use of biological systems principally differ from similar operations performed by digital computers. In the former case, complex fragments act as elementary initial units, and not as structures of easiest symbols transmitted by bits (binary digits), as it takes place in the latter case. Fragments may be represented by phonemes in case of processing spoken voice, image, that are operated by vision, and so on. When processing images in various dynamic conditions of Belousov-Zhabotinsky media, two of them (named excitable and oscillating ones) gained acceptance. In this case the nature of an actual operation is determined by both the preset conditions and the state of a "picture" (its positive or negative form).

If a medium functions in the excitable mode, the basic elementary operations of processing a black-and-white image are represented by the contour enhancement of its fragments and their subsequent evolution, whose versions may be conventionally named "contour(+)" and "contour(-)". In the first case (positive process) the appeared contour spreads outermost in the course of evolution in the medium, in the second (negative) - converges inboard. These operations make it possible to reproduce all practically used operations of processing black-and-white images. Related to them is determination of a general form, i.e. exclusion of minor details, and segmentation-division of a complex "picture" into easier fragments. Reaction-diffusion processors may intensify parts of lesser size and, at the same time, eliminate larger ones.

Evolution of images in Belousov-Zhabotinsky medium functioning in the oscillating mode turns out to be much more complicated as compared to the excitable ones. In this case, a half-tone positive is first transformed into a black-and-white negative. In this case the fields with decreasing brightness are step-by-step separated, then-so do the contours of individual fragments and, finally, the "picture" is converted into the original half-tone image.

A most important feature of the media in question that springs out from the considered example, is that they essentially represent the realization of a temporal sequence processor transforming a complex spatial distribution of fragments into their successive separation in time. In practice, this simplifies, for example, processing of aerial photography or space survey. The medium successively separates identical-brightness fragments of positive and negative pictures, which makes it possible to explode them into individual components and simplify a corresponding analysis.

Search for the shortest way in a labyrinth is one of the best known problems of an artificial intelligence. A concrete task in this case (for example, selecting an optimum alternative from the several preset ones) is considered as a labyrinth, and the process of searching a solution-as "wandering" over it. Efforts to find efficient algorithms have been made since the 1960s.

Relatively recently American and Russian scientists have proved that photosensitive media of Belousov-Zhabotinsky might be efficient for solving tasks of this kind. In particular, they have been already used in one of the processor versions developed by our team. The opti-

стр. 45

Search for the shortest way in an arbitrary labyrinth by Belousov-Zhabotinsky media.

mum architecture of such devices should be of a hybrid type, i.e. present a combination of the reaction-diffusion component and a digital computer. In this case a complicated computing operation-simultaneous parallel spreading of the wave over all paths in the labyrinth - is carried out by the medium, whereas the subsequent data processing is performed by a digital processor.

Successive stages (stored in the memory) of wave travel in the labyrinth can be used for finding the shortest way from its entrance to a selected exit. Suppose, a wave propagating from the entrance has passed through the first branching point. At this moment, the digital computer performs the "filling" of labyrinth image with white beginning from the exit. As a result, only a fragment unconnected with the exit remains black. This part can easily be excluded from the image numerically. Successive use of this technique makes it possible to determine the shortest sought way.

Of course, the story of unique capabilities of the molecular elemental base does not imply that in the near or not so near future devices on its basis will fully replace semi-conductor digital computers. They have durably entered many fields of our life and optimally meet their requirements. Nevertheless, new problems arise, and traditional semi-conductor devices, which have come to their physical limit, can hardly solve them or are not capably of doing it at all.

One of them is the development of supercomputers with maximum computation characteristics based on molecular digital circuits, which are necessary, for example, for creating global management systems. Unique properties of the distributed reaction-diffusion media, which are actually an analog form of data processing, leave a hope that they will be able to become a basis for development of a new trend of computer science. Realization of their information capabilities is becoming more and more detailed, as if confirming the forecast of Lee A. Rubel, German mathematician, one of the progenitors of this approach. In the early 1990s he wrote: "The future of analog computing is unlimited. As a visionary, I see it eventually replacing digital computing, especially in the beginning, in partial differential equations and as a model in neurobiology. It will take some decades for this to be done. In the meantime, I believe, it is a very rich and challenging field of research, although (or may be therefore) it is not popular today."

Illustrations supplied by the author