Click for next page ( 24


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 23
3 Opportunities in Information Processing INTRODUCTION At the heart of today's information society is the capability to use the enor- mous amounts of information we are able to send from one place to another. In using information, we filter, digest, consolidate, reorganize, add, and share; i.e., we process information before deciding to selectively retain it or send it to other users. All of this processing adds value to the information by making it more useful for various applications. In some cases, value is created by feeding the input information into a computer program that is designed to diagnose a condition or to predict or calculate a result based on such input. The importance of information processing to our society is pervasive. It is vital for commerce, manufacturing, national defense, finance, education, transportation, environmental monitoring and control, efficient energy utiliza- tion, biomedicine, and all areas of scientific research. The worldwide market for information processing hardware and software, including computer systems, office automation equipment, and input/output (I/O) peripherals (but not including displays, data storage devices, communications equipment, and con- sumer audio, video, and personal products) is around $100 billion per year. The U.S. market share is around 55 percent, Japan's share is around 20 percent, and Western Europe's (Germany, United Kingdom, France, and Italy) combined share is 25 percent. This market is currently growing at around 13 percent per year and is projected to grow threefold by the year 2000.~2 Today, almost all information processing is done electronically. Electronics brings an ever-enlarging range of applications to all segments of society. 23

OCR for page 23
24 PHO TONICS However, the use of lightwave technology is an exciting extension of the electronic technological approach. While electronics combines speed, control, and precision with low cost, it has shortcomings in the bandwidth (amount of information) that can be carried on an electronic channel, and it suffers from susceptibility to electromagnetic interference. Although bandwidth con be effec- tively increased by implementing many side-by-side electronic channels, this approach requires that special attention be paid to isolating each channel from the others, resulting in increased size and higher cost. The result is that electronics suffers from a mismatch between the speed of handling information within a processor and the rate of sending information between processors or from a processor to an outside user. Thus there is a communications bottleneck inherent in electronics. Photonics offers some significant advantages that can greatly expand the performance capability of information-processing systems, such as nearly limitless bandwidth and immunity from electromagnetic inter- ference. In addition, light is a natural vehicle for conveying information in many side-by-side channels, an aspect known as connectivity. These advantages are technically available, and the existing widespread commercial use of photonics in long-distance telecommunications systems has provided a seed from which the use of photonics in information processing might grow. CURRENT TECHNOLOGICAL STATUS Digital Technology In digital applications, information is coded into binary form, where a voltage near one standard value represents a 1 and a voltage near another standard value represents a 0. A robust digital information system is one that can recognize the value of a bit (1 or 0), carry out logic operations on bits, and cascade bit-processing operations without a significant probability of making an error, thereby permitting numbers to be represented and manipulated over many orders of magnitude with ultrahigh precision. Present-day digital electronics uses transistors both to carry out logic operations and to store the bits. Transistors for digital applications feature three terminals, standard system-wide voltages, and high gain with large noise margins. Three terminals ensure isolation of the output from the input; stan- dard voltages are established by the system-wide bias voltages and system ground, not by the transistors themselves; high gain with large noise margins keeps the output values standardized despite the variability (10 to 20 percent) in the characteristics of the individual transistors. These features also combine to permit a single device to drive many other devices (fan-out)- or many devices to drive a single device (fan-in), while maintaining the standard voltage levels. A robust circuit of transistors maybe designed that allows for some degradation

OCR for page 23
OPPORTUNITIES IN INFORMATION PROCESSING 25 of voltages throughout the circuit while still providing bit values that are recognizable without a significant probability of error. The silicon (Si) integrated circuit is the transistor technology of choice for general purpose digital electronics. The worldwide market for Si integrated circuits exceeds $20 billion.) Si microelectronics is low-cost because of mass fabrication methods. The simultaneous fabrication of many millions of devices on a large Si wafer, coupled with the lack of need to test, adjust, or repair individual devices (except for quality-control testing), ensures low cost. Si integrated circuits reliably carry out logic and memory functions despite the variations of the characteristics of individual devices that mass production entails. Examples of successful, mass-produced Si integrated circuits are micropro- cessors and dynamic random access memory (DRAM) chips. Microprocessors can now be produced with circuit densities of 200,000 devices/cm operating at cycling rates of 100 MHz, while dissipating only 0.05 pJ/switching cycle. This means that only 1 W/cm2 of heat is generated at the maximum operating rate, a heat load easily dealt with using today's cooling technology. The cost of readily available DRAM chips puts the cost per device at less than 0.01 cent, packaged, tested, and ready to plug in. DRAM chips with 4 Mbits of storage are already being produced in pilot line facilities. These numbers are by no means static, with improvements being developed at a rapid pace. These standards must be not only matched but also exceeded by alternative technologies. Devices that are going to compete with Si transistors must be able to reproducibly switch between two (or more) easily recognized states, must show gain if they are to be cascadable, must be immune to undesirable feedback from the output to the input, and must be inexpensively mass produced. Furthermore, in applications that call for rapid, repeatable cycling, the power dissipation must be kept within cooling capacity limits. Turning to alternative materials for transistor electronics, a semiconductor that has received serious attention as a competitor for Si is gallium arsenide (GaAs). Since 1970, GaAs electronic devices that operate at higher speeds than Si devices have been fabricated and studied. The higher speed is a conse- quence of higher carrier mobility and lower intrinsic carrier concentration in GaAs. Other advantages over Si include potentially lower heat dissipation, higher temperature operation, greater resistance to ionizing radiation, and optoelectronic capability. However, there are disadvantages with GaAs that are rooted in present fabrication technology. Gallium arsenide is chemically less stable than Si, it does not form a stable self-oxide as a good insulator, it is harder to control the stoichiometry, and there are difficulties with fabricating reproducible contacts and interfaces. These features make it more difficult to realize low-cost, reliable, mass production of GaAs integrated circuits. Never- theless, the advantages, particularly in speed, have enabled GaAs to find wide- spread use in certain select applications. For example, discrete devices such as

OCR for page 23
26 PHO TONICS radio frequency oscillators and amplifiers used for communications purposes are made from GaAs or related III-V materials. Current medium scale in- tegration (MSI) GaAs technology can produce 1-kbit memory chips. For applications that require more speed than Si can provide, GaAs integrated circuits represent the most mature alternative and are commercially available. The integrated circuit has clearly experienced widespread success. But as engineers have devised methods of placing more and more devices on a single chip and have figured out efficient ways to interconnect more complex chips, they have come up against serious weaknesses that limit further improvements in this technology. While the fastest Si transistors can switch in less than 0.1 ns, the fastest cycling time for a Si microprocessor is only~ 10 us. This limitation is imposed by the limited capacity of electrical interconnects on the chip. Electrical striplines have to be large enough to reduce attenuation of high-speed signals, and they have to be placed far enough apart to minimize cross talk. Other problems are ground loops and clock skew (the problem of synchronizing pulses distributed over many circuits). Equally serious is the problem of electrically connecting chips to other chips or modules of chips to other modules. The number of I/O electrical pads that can be fabricated on a chip is limited by the space they occupy and the power consumption they demand. Furthermore, the number of wires that can be bonded to these pads is limited by yield. In addition, the number of I/O connections that can be simultaneously switched is limited by the resulting rate of change of current (di/dt), a factor that produces noise in other parts of the circuit. The resulting I/O bottleneck, limited by the number of connections (connectivity) and the bandwidth of each connection, limits the throughput of a chip. These connectivity and bandwidth problems become even more difficult to overcome when dealing with intercon- nects at the board-to-board and processor-to-processor level. Taken together, these problems place practical limits on the performance of all-electronic information processing systems. Analog Technology While digital information processing is preeminent when precision over many orders of magnitude is required, there are many applications where analog information processing is used to handle large amounts of information at speeds that cannot be achieved with digital technology. These applications are already using photonic technology. Much progress has been made in specialized coherent processors that employ laser sources and utilize both amplitude and phase information on the optical beam.3 In particular, specialized preprocessing of data is carried out in an analog fashion for imaging, mapping, pattern recognition, and spectral analysis, applications that are primarily of value for military systems. For example, synthetic aperture radar (SAR) utilizes

OCR for page 23
OPPORTUNITIES IN INFORMATION PROCESSING 27 conventional optical elements (e.g., lenses) for coherent optical processing to scan through large amounts of data or images previously recorded on photo- graphic film. Another example is that of real-time, wide-bandwidth spectral analysis, in which the signal to be analyzed drives an acousto-optic (Bragg) cell that deflects a light beam that then passes through conventional lenses before being detected by a detector array. Bragg cells use traveling acoustic waves to create an index grating in a crystal that diffracts and shifts the frequency of an opticalbearn passing through the crystal. These cells are available commercially for operation in the gigahertz (GHz) range. In these applications, after the analog preprocessing, the data are digitized and handed off to conventional digital computers. Such analog processingis now carried out photonically rasher than electronically, because the inherent coherent parallel processing capability of optics allows greater throughput, system simplicity, speed, and a high time- bandwidth product, while maintaining the necessary accuracy (dynamic range and resolution). A commercial application of analog technology is in laser printers and scanners, where a laser beam is deflected across a target to be written on (for printers) or read (for scanners). A rotating mirror or a spinning holographical- ly ruled disk is used for large angle deflection. For example, in laser printers that have a single laser as the light source, a multifaceted rotating mirror provides a raster scan, while the rotating photosensitive drum that captures the image provides an orthogonal scan. The laser diode is turned on and off, and the lenses keep the beam focused on the surface of the drum (Figure 3.1~. PHOTONICS LEVERAGE As stated in the introduction, the potential leverage of photonics on infor- mation-processing stems from inherent advantages of optics, including ultrawide bandwidth, freedom from electromagnetic interference, connectivity, and coherent parallel processing capabilities. Fiber-optic technology featuring ultrahigh bandwidth and freedom from interference is already available com- mercially and can be harnessed for information-processing applications. De- velopment activities in interconnection/communications that utilize these advantages do not require a change in current analog or digital processor archi- tectures, thus allowing the technology to be more immediately useful in infor- mation-processing systems. However, the ultimate leverage of photonics will be exploited if pragmatic ways of capitalizing on the full parallelism of optics can be found. This calls for intensive applied research on new techniques and devices, including new processor architectures and optical nonlinear (i.e., logic) device arrays.4 The inherent advantages of photonics are now discussed in more detail. The leverage offered by the nearly limitless bandwidth of optics relative to

OCR for page 23
28 / ~----- I ~ DEVELOPER - PAPER ~ TRANSFER ELECTRODE FIXER PHO TONICS CHARGING ELECTRODE ROTATING PHOTOSENSITIVE DRUM \, LASER BEAM \ - LASER DIODE / COLLIMATING LENS FOCUSING ' LENS - - it\ . / ROTATING MIRROR SCANNER FIGURE 3.1 Schematic of a laser~lectrophotographic printer showing a many-faceted rotating mirror for raster scanning and a rotating photosensitive drum that provides the orthogonal scan. electronic technology can be viewed in many ways. Wide bandwidth implies capability for ultrahigh-speed channels, on the order of 10~ bits/e by todays research results. This means that the entire text of the Encyclopaedia Bntan- nica can be transmitted over an optical fiber in just 1 s. This same bandwidth can be parceled out to many parallel users (e.g., 104 users at 106 bits/e with time-division multiplexing). Furthermore, light as a signal carrier offers the potential for increasing the information-carrying capacity from today's 10~ bits/e to more than 10~4 bits/x.5 The immunity from electrical interference is another major source of leverage of photonic technology. Here, problems with ground loops, pick-up, and susceptibility to electromagnetic interference and radio frequency inter- ference are significantly reduced because the optical carrier signal is generally at least 4 orders of magnitude removed from electrical interference frequen- cies. Such basic considerations are already leading to use of fiber optics rather than electrical cables to interconnect computers. Present advanced develop- ment activities are probing how far into a computer this optical interconnect ap- proach will penetrate; many people feel that optically interconnecting printed circuit boards will become a widespread reality. A closely related advantage of optics for interconnects is freedom from cross talk; i.e., two beams of photons can pass through one another with

OCR for page 23
OPPORTUNITIES IN INFORMATION PROCESSING 29 negligible interaction. The same is obviously not true for electrons flowing in crossed wiring. This advantage can lead to numerous interconnection advantages, especially with the use of optical elements such as lenses, holo- grams, and waveguides to direct beams. Freedom from cross tale leads to another advantage of optics, connectivity. For example, a lens can transmit a large spatial array of optical signals through free space onto a matched array of receivers. The resolution is limited by diffraction effects, and one may easily conceive of transmitting millions of parallel signals using 1-micron-wavelength light, f/1 optics, and source and receiver arrays having cross sections of centimeters. Implementing the equiva- lent connector in electronics is inconceivable. Finally, freedom from cross talk leads to the last important leverage point for photonic technology, coherent parallel processing capability. For example, by passing a coherent beam of light from an object (e.g., a photographic plate) through a lens, one obtains, at a particular location, the spatial Fourier trans- form of the input. This important signal-processing transform dissects the input image into spatial frequency components. The lens is the sole element respon- sible for this parallel processing transformation, whereas to do the equivalent operation electronically would require a relatively large processor and numerous interconnections and switches. Powerful optical analog processing of this type is already being harnessed in practical systems, as described under "Analog Technology" in the "Current Technological Status" section. AREAS RIPE FOR PRODUCT DEVELOPMENT Digital Interconnect Technology As pointed out above, progress in electronic digital computers is being limited by communications considerations. The fastest transistors switch in 5 ps, whereas the fastest systems run with a cycle time of =5 us. The source of this 3-order-of-magnitude disparity can be traced to communications constraints such as bandwidth and connectivity. Photonics can ease these communications constraints through the use of optical interconnections exploiting the large bandwidth of optics. Optical fibers are already being used to provide high- speed, computer-to-computer communications. Several computer manufac- turers are, or soon will be, using optical fibers to connect their central processing units and their storage systems (e.g., disks and magnetic tapes). Some manufacturers are already using optical fibers to interconnect modules within their processors. The evolution of this process is illustrated in Figure 3.2. In assessing what is ripe for development in photonic interconnect tech- nology, it is important to note that there are distinct differences in photonic technology needs for telecommunications and information processing. These

OCR for page 23
30 cpu ~ Node ~~ Y i A; Small Area ~ Ne t wo rks Airs ~ AN a. - Lase r //j/~i///// Detectors . . . / \ , ~ ~ V -I /}~d~, / ~///oo a a ~ |, On-Chip Clock Distribution PHO TONICS Board Opto-l.C. ~ A ~c: ~ I\` _ 1~c~o I. ;.... AL Fiber _ ~ O O O | \Board to Board Chip Opto }~[- Chip to Chil? FIGURE 3.2 Scenario for the evolution of optical interconnects within computing systems. differences stem primarily from the short distances ~ < 1 km, usually < 100 m) of information-system interconnection links, where the emphasis is on multiple parallel channels, many components, and affordability. Furthermore, photonic devices for information processing systems must be compatible with high- packing-density integrated circuit processing and package operating tempera- tures. Long-distance telecommunications optoelectronics is dominated by indium phosphide (InP)-based lasers and detectors because they operate in the spectral region (1.3 to 1.5 micron) where fibers have lowest loss and highest bandwidth (i.e., low dispersion). Since the wavelengths emitted by these InP- based sources are not effectively detected by Si detectors, the long-term ability to form totally integrated interconnection circuits is severely limited, as will be discussed further. On the other hand, GaAs-based sources, which emit at shorter wavelengths (approximately 0.85 microns) and are compatible with optical fiber characteristics for the great majority of applications in information- processing interconnects, can be effectively utilized with Si detectors. In light of the relatively well-developed GaAs materials growth and electronic circuit technology and the development of inexpensive sources for the commercial market (e.g., compact disk and laser printers), it is inherently reasonable to concentrate much of the photonic product development effort on GaAs tech- nology.

OCR for page 23
OPPORTUNITIES IN INFORAL4TION PROCESSING 31 One important area is the fabrication of components that provide both electronic and optoelectronic functions. Both hybrid and monolithic approach- es show promise for this purpose. In the hybrid approach, optoelectronic chips are mounted along with electronic circuits on a common board. The advantage of this approach is that present and future yield and reliability of individual components can be maximized separately. Since envisioned interconnection applications ~11 require relatively few connections between the electronic and photonic devices, economics will favor this approach for the relatively near future. Both present and advanced hybrid schemes should tee further developed with emphasis on utilizing electronic packaging advances and reducing cost, especially in the attachment of multiple fiber pigtails to a single module. In this hybrid approach, one could imagine InP- as well as GaAs-based modules being developed, but for reasons cited previously regarding cost and Si compatibility, it is felt that GaAs is the appropriate material system for information-processing product development. While hybrid components can provide the functionality desired, the scaling of production for large-volume applications and the use of photon~cs In a broader class of information-processing applications will require the develop- ment of monolithic circuits. These devices, referred to as optoelectronic integrated circuits (OEICs), could be formed in GaAs and would take full advantage of the state of development of GaAs electronic integrated circuits and optoelectronic components (Figure 3.3~. The OEICs that could be most effectively developed today include detector/amplif~er/multiplexer arrays. Cir- cuits involving sources could also be pushed to the product level, but for rea- sons involving heterojunction materials compatibility between laser/LEDs and electronic circuits, further development will be necessary. These issues are Electronics Detector ~ Laser IAL: Electra - optic modulator ~ / Laser ~ Electro-optic switch ~7Y:;4 / /GaAs substrate V L Single mode fibers Waveguide FIGURE 3.3 Schematic of a model optoelectronic integrated circuit (OEIC).

OCR for page 23
32 PHO TONICS further discussed in the following section, where areas of sustained research are described. Closely related components that are, however, ready for product development are individually addressed laser arrays and detector arrays. A technological driver for incorporating OEICs into existing computer architectures comes from the potential for reduction In power requirements and power dissipation. Arguments have already been advanced for the increase in communications bandwidth offered by an optical channel over an electrical channel. But even with the same bandwidth, optical interconnects would lead to a reduction in the drive power requirements for the I/O connections when compared to that required for electrical lines that need to be terminated to prevent reflections. Thus, when power requirements are a limiting factor, optical interconnects have an advantage for chip-to-chip communication at data rates of around 100 Mbit/s or higher. Analog Technology Advances in analog optical processing in many cases will be linked to advances in photonic devices. For example, a spatial light modulator (SLM) provides the equivalent of real-time fun or image planes and can be electrically or optically addressed.7~8 Devices that are reliable, economical, and provide at least TV frame rates are under development at a number of companies. The SLM is a key component that Fill greatly add to the capabilities of analog optical processors and should receive accelerated development. The SLM will improve the practicality of vector-matrix multipliers and other optical analog information processing systems that provide enhanced real-time processing. Advances in two-d~mensional detector arrays (already commercial products) will also enhance the performance of such systems. Guided-wave or integrated optic devices are also attractive for analog processing. These structures consist of optical circuits on a chip (analogous to electronic circuits or to planar versions of bulk-optic circuits) that provide high speed, high reliability, potential low cost (due to batch processing), and low drive power.9 Impressive laboratory and developmental demonstrations have been made in information processing. Examples of their use include forming a monolithic spectrum analyzer, a high-speed (gigahertz) analog-to-digital con- verter, and high-speed modulators for optical-microwave applications (e.g., fiber-optic-based phased-array radar signal distribution). These devices, usually formed in lithium niobate (LiNbO3) or GaAs, are just becoming commercially available and have applications in telecommunications (switching) and sensors (gyros), as well as in information processing.~

OCR for page 23
OPPORTUNITIES IN INFORMATION PROCESSING AREAS APPROPRIATE FOR SUSTAINED RESEARCH 33 This section first addresses research areas that are extensions of those discussed in the previous section, dealing with devices and systems for intercon- nect technology and analog information processing. Then a discussion is presented of devices and architectures that have as their ultimate goal all-optical information processing. Digital Interconnect Technology In the area of OEICs, sustained work is needed, especially in the area of light-emitting chips. The development of low-cost, high-yield OEICs with lasers, LEDs, and/or modulators is a challenging area because the optoelectronic and electronic circuits have quite different materials and processing needs that complicate device fabrication. Many applications require extremely low- threshold (< 1 mA) lasers and/or surface (rather than edge) emitting lasers. While impressive research results have been achieved ~ all these technologies, much work remains to be done. An attractive alternative to forming on-chip lasers or LEDs is to form on-chip optical modulators. In many cases, this can reduce the drive-power requirements, simplify the geometry, and simplify circuit fabrication. A closely related area that should be supported is the growth of hetero- epitaxial material such as GaAs on Si or GaAs on InP and formation of high- quality devices in this material. GaAs/Si technology is particularly exciting for information processing because it carries the integration ideas described previ- ously several steps further (e.g., formation of a GaAs source, GaAs multiplexer, and Si capacity-coupled metal-oxide silicon (CMOS) circuit). The research status of the technology today is that while, in only a few years of work, elec- tronic circuits in both Si and GaAs are fairly comparable to those formed on conventional substrates, the fabrication of lasers on GaAs/Si is at quite an early stage of development (e.g., short lifetimes comparable to early 1970s results on GaAs substrates). With each of the above technologies, the use of free space for interconnec- tion, rather than guided wave media (e.g., optical fibers or polymer waveguide circuit boards), should be investigated. Success in this area could be particular- ly valuable in interconnecting two-dimensional laser and detector arrays and in parallel connections between very large scale integrated (VLSI) circuits or printed circuit boards. As mentioned previously, optical fibers are being used to connect city to city, computer to computer, in some cases board to board, and eventually chip

OCR for page 23
34 PHO TONICS to chip. Even gate-to-gate connections with optical fibers are being con- templated. There are several difficulties associated with this evolutionary approach that further research might help to overcome, thereby accelerating the introduction of this new technology. One difficulty is that this evolutionary ap- proach takes advantage only of the bandwidth of optics and not the connectivity. The retrofitting of one technology into architectures optimized for another puts optics at a disadvantage. Compatibility issues limit the performance of optics to that of electronics and thus compromise its viability. Economic issues force optics to compete directly in terms of costs with other wide-bandwidth interconnects such as coaxial transmission lines. One of the great unexploited advantages of optics is its connectivity. A lens can easily handle a 100 x 100 array of communication channels, each with more than a thousand times the capacity of a microwave link. The equivalent 10,000-pin connector would be very difficult to implement in electronics. Research should be carried out to find architectures that utilize all of this connectivity. Analog Technology In the area of analog optical processing, several devices are under study that should tee given additional support. High-performance SLMs are attractive. These devices have potentially high frame rates and signal clocking that is particularly attractive for signal processing applications. Holographic optical elements and advanced SLMs could also provide means for optical intercon- nects. Materials and Algorithms The need for additional research in materials for photonics must be em- phasized. Many of the devices already described would be more attractive if they were made from materials having higher electro-optic/acousto-optic/non- linear optic figures of merit than those currently available. Research in al- gorithms is equally important. Available devices will be even more useful if algorithm research intensifies its focus on utilizing the bandwidth, freedom from interference, connectivity, and coherent parallel processing capabilities of photonics. In addition, increased interaction between algorithm and device researchers could lead to invention of devices with special characteristics that algorithm developers request.

OCR for page 23
OPPORTUNITIES ININFORM~4TIONPROCESSING Digital Optical Devices 35 The ultimate benefit of parallel photonic processing could occur if practical optical logic could be developed.7 This would minimize the conversion of information between electronic and optical formats. Present research in optical digital computing concerns the speed, power requirements, and noise margins of optical logic devices. It is clear that there are optical nonlinearities that react in the order of femtoseconds; the question is, how can they be practically exploited? Research efforts are aiming et developing better materials, especial- ly those that would have larger nonlinearities, reducing the optical power requirements through the design of better devices, and exploring better ways of cooling. One direction of device research is to exploit the properties of semiconductors and microelectronics fabrication techniques. One promising optical logic device is based on a photodiode and an electro-optic modulator (Figure 3.4~. Another promising device is a semiconductor optical resonator. Semiconductor-based optical logic gates have the potential to compete with transistors in terms of very high speed and low power requirements, but there is much research and development that needs to be done to make them practi- cal. Organic polymers have been shown to have interesting nonlinear and electro-optic properties. Much research is now going on in this field and should be continued. Another important direction in optical digital information processing involves neural networks, a concept based on analogies to the central nervous system. Synthetic neural networks are being implemented both in electronics and in optics.~3~4 The basic approach models the behavior of a neural network by using a connection matrix between an array of inputs and an array of outputs and nonlinear feedback to the inputs from all of the outputs to modify the strength of the connections. This approach is being applied to pattern recog- nition and optimization problems. In the case of pattern recognition, a neural network can be made to automatically learn to distinguish between various patterns without the need for programming. In the case of optimization, these networks will search in parallel for a minimum in a multiparameter space. In this application, a significant advantage of optics is connectivity. CONCLUSIONS Photonics will have an enormous impact on information-processing tech- nology.- The more conservative, evolutionary pathway is to find ways for photonics to penetrate as far as possible into the interconnect technology of information processing. Such a pathway will enhance, by many orders of

OCR for page 23
36 CONTACT BIASING PHOTODIODE ISOLATING TUNNEL JUNCTIO INFRA-RED IN ~ MODULATOR - 16 PHO TONICS INTEGRATED DIODE-BIASED SEED ARRAY ~ RED BIAS IN ~ VSUPPLY INFRA RED OUT ~ VOLTAGE SL- SUPERLATTICE GOLD INSULATOR MOW - MULTIPLE QUANTUM WELL FIGURE 3.4 Three~imensional schematic of an array of three-terminal optical logic devices. Each device consists of a photodiode, which absorbs the red bias input light, and a modulator (self- electrooptic effect device, or SEED), which has its infrared transmission controlled by the photodiode. magnitude, the speed of handling and the density of packing information. This route is already having a practical impact at the machine-to-machine level. The more daring, revolutionary pathway is to replace electronic logic and memory devices by photonic devices and devise new architectures to take fuller ad- vantage of the leverage of photonics. This pathway offers the tantalizing, long- term prospect of far more powerful information-processing technology but must be considered speculative until practical optical logic devices are developed. These visions demand that photonics research and development be pursued at a vigorous pace to keep the United States at the forefront of information processing science and technology. REFERENCES 1. 1987. 1987 U.S. market report. Electronics, January 8: 51-74. 1987. 1987 overseas market report. Electronics, January 22: 66-88.

OCR for page 23
OPPOR=NITIES IN INFO~TION PROCESSING 37 2. 1986. The high tech race, who's ahead? Fortune, October 13: 29-37. 3. Goodman, J. W. 1982. Architectural development of optical data process- ing systems. Journal of Electrical and Electronics Engineering, Australia 2 ~ S e p t e m b e r ~ : 1 3 9 - 1 4 9 . 4. IEEE. 1984. Special Issue on Optical Computing. Proceedings of the IEEE 72(July):755-975. 5. Smith, P. W. 1987. On the role of photonic switching in future communica- tions systems. IEEE Circuits and Devices 3(July):9-14. 6. Goodman, J. W., F. J. Leonberger, S-Y. Kung, and R. A. Athale. 1984. Optical interconnections for VLSI systems. Proceedings of the IEEE 72(July):850-866. 7. Bell, T. E. 1986. Optical computing: a field in flux. IEEE Spectrum 23(August):34-57. 8. Warde,C.,andA.D.Fisher. 1987. Spatiallight modulators: Applications and functional capabilities. Pp. 478-524 in Optical Signal Processing, J. L. Homer, ed. San Diego: Academic Press. 9. Tamir, T., ed. 1988. Guided Wave Optoelectronic Devices. Springer Series on Electronics and Photonics, Vol.26. Heidelberg: Springer-Verlag. 10. Hall, D. G. 1986. Integrated optics: The shape of things to come. Pho- tonics Spectra 20(August):87-92. 11. Glass, A. M. 1987. Optical materials. Science 235(February 27~:1003- 1009. 12. Robinson, A. L. 1984. Multiple quantum wells for optical logic. Science 22S(August 24~:822-824. 13. Hecht-Nielsen, R. 1988. Neurocomputing: Picking the human brain. IEEE Spectrum 25(March):36-41. 14. Abu-Mostafa, Y. S., and D. Psaltis. 1987. Optical neural computers. Scientific American 256(March):87-95.