Much work is being done toward the optical implementation of traditional electronic processing and computing methods. Many of the proposed methods may not be the optimal way to utilize the benefits of optical techniques. We introduce here a new optical gate - the Fredkin gate - that is in principle minimally dissipative and its response time in some implementations may be limited only by the duration of optical pulses (i.e. in the subpicosecond range). To indicate the viability of this novel approach, a number of optical implementations of Fredkin gates with some interesting applications are proposed.

The potential and promise of very high-performance spatial light modulators (SLMs) capable of performing logic operations has motivated the investigation of digital computing systems that possess many desirable attributes of optical systems, namely massive parallelism, global communication at high bandwidths, high reliability, many useful degrees of freedom, robustness in the presence of defects, and simplicity. The parallelism of easily realizable optical single-instruction, multiple-data (SIMD) arrays makes them a natural choice for implementation of highly structured algorithms for the numerical solution of multi-dimensional partial differential equations and the computation of fast numerical transforms. A system comprising several SLMs, an optical read/write memory, and a functional block to perform simple, space-invariant shifts on images has enough flexibility to implement the fastest known methods for partial differential equations (e.g. multi-level methods) as well as a wide variety of numerical transforms (e.g., FFT, Walsh-Hadamard transform, rapid transform), in two or more dimensions, and using either fixed or floating-point arithmetic. Performance is projected at greater than 109 floating-point operations/s using SLMs with resolution 1000 x 1000 operating at 1 MHz frame rates.

Nonlinear optical interactions combined with the use of ultra-short optical pulses has certain attractive features for ultra-highspeed optical computing. However, at present, a serious price must be paid in terms of interaction time or energy or both.

Parallel processing models as computational paradigms are discussed and related to optical computing. Two classes of parallel computing models are discussed - shared memory models and graph/network models. These models are used to analyze some of the possible effects of optical technology on parallel computing. It is found that the use of optics potentially provides certain fundamental advantages. In addition, some factors that limit the communication capabilities of optical systems in the case of network models are found.

A two dimensional optical logic operation with three microchannel spatial light modulators (MSLMs) is demonstrated. In this proposed system, it is possible to get 16 functions of two dimensional logic operations. The system and the experimental results are described.

The mathematical and physical basis for realizing compound literals via optical shadow-casting is discussed. The relation between various switching algebras and cubical notation is considered. An emphasis is placed on the physical independence of the shadow-casting implementations with respect to possible algebraic constructs. Shadow-casting systems for content-addressable memories are discussed, and the consequences of compound literal implementation in terms of throughput and energy per switching operation are presented. The advantages of using bistable optical switching elements that employ third order nonlinearities are considered.

A microcomputer-based real-time programmable optical signal processing system utilizing a Magneto-Optic Spatial Light Modulator (MOSLM) and a Liquid Crystal Light Valve (LCLV) is described. This system can perform a myriad of complicated optical operations, such as image correlation, image subtraction, matrix multiplication and many others. The important assets of this proposed system must be the programmability and the capability of real-time addressing. The design specification and the progress toward practical implementation of this proposed system are discussed. Some preliminary experimental demonstrations are conducted. The feasible applications of this proposed system to image correlation for optical pattern recognition, image subtraction for IC chip inspection and matrix multiplication for optical computing are demonstrated.

A new optical logic operation technique is proposed which is based on spatial encoding and superposition of a decoding mask with the coded input patterns. Sixteen logical functions of two logical variables can be realized. In the present method, the multiple instruction-stream multiple data-stream (MIMD) logic operation is realized, that is, different logical operations are performed in parallel. The new method is the MIMD extension of Tanida and Ichioka's optical shadow-casting logic, which is based on a SIMD (single instruction-stream multiple data-stream) logic gate array. Simple examples are demonstrated. An application of this method to the Minnick cellular architecture is discussed.

Ferroelectric liquid crystals (FLOs) in the Surface Stabilized (SSFLO) geometry can be used to make voltage operated light valves with bipolar response, intrinsic memory, and microsecond switching speeds. These light valves, configured as linear or matrix arrays, have a variety of practical optical computing applications. We present a high spatial resolution matrix-addressed SSFLO light valve array suitable for use as a spatial modulator for transmitted light. The array employs the typical SSFLO geometry of the FLO as a dielectric in a transparent capacitor. The capacitor plates are indium-tin oxide (ITO) coated glass with the ITO patterned into electrical strips, the top and bottom strips oriented at 90° with respect to each other, to for a matrix-addressed array of overlapping square areas, the individual addressible shutters (pels).

This paper presents a description and design considerations for silicon/PLZT spatial light modulators (Si/PLZT SLMs) which provide high parallel processing power, dynamic range, cellular resolution, sensitivity and the potential for implementation of "smart" optical devices. Following an overview of Si/PLZT SLMs, we discuss potential performance with respect to fundamental limits. Finally, we predict performance in specific applications. This analysis suggests that Si/PLZT SLMs can play an important role in the implementation of a variety of optical processing and computing systems.

The PRIMO optical processor is an analog outer product processor based on 1-D arrays of electrooptic modulators. PRIMO is capable of performing matrix-matrix multiplication, convolution, correlation, and other linear computational algorithms. In this paper we discuss techniques for representing bipolar and complex data in PRIMO using square law detectors. It is shown that by utilizing a bias and properly sequencing the data the square law detection nonlinearity can be fully compensated.

A, new optical switching device, the two-port Sagnac interferometric switch (TPSIS) is proposed. The TPSIS uses a polarizing beam splitter and half-wave plate to auto-isolate beam retroreflections. Since the polarizing beam splitter separates the interferometer input and output signals the TPSIS can be used as a two-port ultrafast optical logic device. The applications of TPSIS to binary, and multiple-valued,logic and arithmetic all-optical computation are discussed. Using the TPSIS, various Post, Residue and Webb multi-valued logic and arithmetic functions can be optically implemented. A number of implementation examples are presented.

The use of stimulated light scattering as a means for achieving optical control functions directed toward an optical computer is described. Stimulated thermal Rayleigh scattering is seen to be a preferred nonlinear optical mechanism when compared with the more familiar stimulated Raman and stimulated Brillouin scattering. It possesses the highest gain, the lowest threshold, and scattered radiation which is approximately the same frequency as the inducing radiation. Optical control functions such as optical bistable switching, optical amplification and optical limiting or clipping are described.

The utilization of optical computing will ultimately depend on the development of very efficient optical nonlinear materials that can process optical information rapidly using very low optical switching energy per operation (of the order of 10 to 100 fJ). We will describe both an all-optical and an electro-optical approach to optical bistable devices. In the all-optical approach, the nonlinearities associated with excitons bound to neutral donors in good quality CdS platelets are utilized. Oscillator strengths of about 7 with a consequent radiative decay of 500 ps have been measured. Under ideal conditions it becomes possible to approach the fundamental limits to optical switching energies with this system. Switching energies of the order of 10 femtojoule are predicted. Experimentally, we have demonstrated a 4 pJ optical switching energy, the lowest ever reported for an all-optical bistable device. The switching time was measured to be less than 1 ns and was limited by the response time of our detector . In the electro-optical approach, we use a bistable InGaAsP/InP diode laser amplifier. An optical switching energy of less than 1 femtojoule, the lowest ever reported (about 3000 photons), is sufficient to cause the device to switch. In addition to the low optical switching energy, these nonlinear optical devices also offer the following advantages: low total energy consumption, high gain, room temperature operation and wavelength compatibility with diode laser sources. The switching-on and off occurs in about half a nanosecond. Optical gain has been measured to be 10-20 dB. The gain is a very important characteristic that enables simple cascadability. The observed switching results from a change in the index of refraction of the order of 10-4 Large two-dimensional arrays of these bistable devices would open new possibilities in the area of digital optical signal processing.

This paper discusses a means of optically interconnecting a pool of processor-memory pairs. The network consists of n optical transmitters (laser diodes), n deflectors (acousto-optic devices or mirrors) and n photo-sensitive receivers (photodiodes). Characteristics of the deflectors and their effects on network performance are developed. One method of using the proposed network is to fix the communication links between processors at the start of algorithm execution. Alternatively, the links between processors may be permitted to change dynamically during algorithm execution. Both of these methods of utilization are briefly explored in the paper.

Optical interconnects have the potential to increase the throughput of future computing systems. Development of optical interconnects at a larger scale for telecommunications and local area networks have demonstrated the capabilities of optical interconnects. To incorporate this technology in computing environments, further development is required in four areas - optoelectronic devices, routing and switching mechanisms, packaging, and architecture development. In this paper, we will discuss specific directions for the development and give examples of research directed toward the insertion of optical interconnects into high throughput computing systems.

This paper introduces new applications and design tradeoffs anticipated for free space optical interconnections of VLSI chips. New implementations of VLSI functions are described that use the capability of making optical inputs at any point on a chip, and take advantage of greater flexibility in on-chip signal routing. These include N-port addressable memories, CPU clock phase distribution, hardware multipliers, dynamic memory refresh, as well as enhanced testability. Fault tolerance and production yields may be improved by reprogramming the optical imaging system to circumvent defective elements. These attributes, as well as those related to performance alone, will affect the design methodology of future VLSI ICs. This paper will focus on identifying the design issues, their possible solution, and their impact on VLSI design techniques.

Advances in computational speed and system complexity are proceeding at a rapid pace. As these systems become more complex and computation speed demands increase, parallel processing becomes the viable solution to meeting the requirements for future systems. Optical signal processing and interconnection provide parallel signal processing capability. A concerted effort is underway to develop optical signal processing and interconnection under the impetus and direction of DARPA. This paper discussed the design, development, and performance of GaAs optoelectronic integrated circuits for application in optical interconnection control and signal processing. Performance of optical transmitters and receivers are given as well as data obtained on 8:1 multiplexed transmitters and 1:8 demultiplexed receivers. Optical data transmission and reception at data rates of 1.5 GHz has been obtained.

Timing constraints for state-of-the-art very large scale integrated circuits (VLSI) in silicon are rapidly approaching communication limits available with layered two-dimensional metal and polysilicon wiring approaches. For such communication-limited systems, reliable clock distribution is a key concern. The range of finite differences in signal delays over clock wires of various lengths for large chips creates a timing skew that is significant when compared to the switching time of transistors in the circuit. The high bandwidth and three-dimensionality of imaging optical systems suggest that optical clock distribution systems have the potential to overcome the timing barriers presented by planar wiring. Clock signals can be holographically mapped to detector sites within small functional cells on a chip surface. Within each functional cell, the clock is distributed via surface wires with negligible delays, reducing skew effects to the variation in reaction times of the photodetectors on the chip. Experiments simulating the response of optical clock detection circuits in standard 4 micron CMOS technology have been performed.

This paper discusses the need for dynamically reconfigurable interconnection networks in parallel computing systems and describes some of the important physical parameters, limitations and tradeoffs of optical and electronic interconnection architectures. Several optical matrix-vector realizations of generalized crossbar networks are given and compared. These systems may utilize acousto-, electro-, or magneto-optic spatial light modulators as their active switching elements, and moderately large crossbar networks (64x 64 to 512 x 512) appear feasible. Finally, the use of general optical sequential logic architectures to implement crossbar and multistage networks is considered.

A new optical interconnect technique, suitable for very high packing densities, is proposed for implementation in VHSIC/VLSI circuits. The approach allows vertical bonding.

A generalized bilinear transform is introduced. Implementations of programmable optical interconnections(POI) based on the bilinear transform are presented. Applications of the POI to digital computer logic are described.

In this paper we deal with optical interconnects. This is an active area both in optical computing, and in the electronics area with the innovation of the optical coupling of electronic chips.

Design considerations for an optical holographic associative memory for implementation of fuzzy cognitive maps are presented. The holographic associations are implemented in a photorefractive material to provide adaptive connection strengths. The difference equation governing adaption is developed based on the physics of hologram formation in photorefractive materials. An architecture and operating system that accommodates the requirements of fuzzy cognitive maps while utilizing the capabilities of photorefractive holography are presented.

The concept of Attentive Associative Processing trades the spatially delocalized and self-organizing features of traditional Associative Architectures for improved hardware efficiency and increased speed of adaption. Optical computing may prove to be ideally suited to implement these architectures.

Shift invariance in associative memories is discussed. Two optical systems which exhibit shift invariance are described in detail along with computer simulations showing their effectiveness. It is shown, however, that shift invariance cannot be achieved with systems that employ only linear interconnections to store the associations without an accompanying decrease in the storage capacity equal to the number of shifted versions that are recognizable.

Parallel-processing architectures consisting of multiple optical adaptive associative modules, cascaded and interconnected in particular configurations, are being developed for applications requiring the manipulation of massive amounts of symbolic information. Specific optical implementations of the individual adaptive heteroassociative modules (elements) are presented and their operational theory and behavior are discussed. These modules adaptively learn and store a series of associations in the form of electronic charge distributions in an optical control device called a microchannel spatial light modulator.

An all-optical fully parallel associative memory system is described which utilizes a holographic data base. Phase conjugate mirrors are used to provide feedback, thresholding, and gain. The memory is compared to the Hopfield neural network model of associative memory and preliminary experimental results are presented.

We present a theory of and a design for a new kind of optical computer employing a phase-conjugating resonator loaded with a volume-holographic recording medium. Coupled Mode equations are derived and solutions presented showing error-correcting content-addressable and associative memory. Multiple scattering in the volume enables the reconstruction of associative chains and cycles, permitting the sequential readout of time-sequences of images and the classifications of images into groups. Applications to invariant pattern recognition, goal seeking by optimization of a path product, and sentence disambiguation are discussed.

Optical correlators provide the unique capability to simultaneously search all regions of an input scene with distortion-invariance, high performance in noise and with the ability to handle multiple objects. This paper examines initial concepts and architectures for correlators in the decomposition of an entire scene to achieve the identification and location of its major constituent parts. The role for such processors in symbolic Al will be emphasized.

Artificial intelligence problems are solved on electronic computers by techniques which make heavy use of address calculation and dynamic management of data storage space. Optical computing is normally associated with numerical problems in which the size of the data space is fixed and addressing may be handled in a predictable manner not affected by actual data values. A criterion is presented for determining the amount of dynamic storage management required for an expert system problem and several methods are discussed for eliminating unnecessary address manipulation by careful choice of data representation. Major emphasis is placed on the implementation of the mathematical technique of resolution. Various resolution strategies are analyzed and the impact of these strategies on storage management is assessed with a view to minimizing the complexity of processing. Finally, novel uses of electro-optical/electronic hybrids are considered for problems in which the state space grows drastically or where reversible control strategies are required to implement search methods.

A rule-based optical symbolic and inference processor concept using a relational graph is described. The architecture and its design are discussed and detailed for one example. The system uses a Fourier/polar/Mellin/Fourier transform representation space and the case studies addressed involve aircraft identification and classification. However, the basic rule-based optical-inference relational graph processor concept and architecture are quite general purpose.