We report the implementation of a prototype 3-D optoelectronic neural system that combines free-space optical interconnects with silicon-VLSI-based hybrid optoelectronic circuits. The prototype system consists of a 16-pixel input, 4-neuron hidden and a single-neuron output layer, where the denser input-to-hidden layer connections are optical. The input layer uses PLZT light modulators to generate optical outputs which are distributed to an optoelectronic analog neural network chip through space invariant holographic optical interconnects. Optical interconnections provide fan-out with negligible delay and allow the use of compact, purely on-chip electronic H-tree fan-in structures. The scalable prototype system achieves 8-bit electronic fan-in precision and a maximum speed of 640 million interconnections per second. The system was tested using synaptic weights learned off-system and applied to a simple line recognition task.

It has been suggested that massively parallel optical routing networks and digital processors are unlikely to be 100% error free due to the myriad problems encountered in the fabrication, alignment and operation of the active components and their associated optics. This paper presents novel algorithms based on the back-propagation class of neural networks which can be implemented on optical hardware which can allow these networks or processors to correctly perform their functions in the event of failure or misalignment of a particular processing element. The algorithms are of particular interest since they are designed to exploit the particular optical characteristics and dynamics of the hardware, in this case nonlinear interference filters (NLIFs).

We discuss the concept of bifurcating neuron and show it combines functional complexity, comparable to that of the living (biological) neuron, with structural simplicity and power efficiency which are important attributes for its hardware realizations, both optoelectronic or electronic. The functional complexity of the bifurcating neuron is key to removing many limitations of present-day neural networks which employ predominantly sigmoidal neurons that can not account for the temporal relations of firing instances of neurons in a network. In the language of information processing, accounting for such temporal relations is synonymous with retention of phase information.

The complex behavior of the bifurcating neuron is characterized. It is shown to be capable of exhibiting phase-locking and synchronization, and that it exhibits a host of firing modalities that parallel those observed by neurophysiologists in the living neuron including chaotic firing and that it is capable of bifurcating between these firing modalities depending on the nature of its input. The implications of this complex behavior for the introduction of a new generation of bifurcating neural networks, that are capable of using chaos as adaptive intrinsic noise, for self-annealing and directed intelligent search of the phase-space of bifurcating networks are discussed. It is argued that the bifurcating neuron concept is key to building new physical structures (bifurcating networks) in which one can study the roles of bifurcation, synchronicity, and chaos in collective nonlinear dynamical signal processing and is moreover key to modeling and understanding higher-level cortical signal processing such as feature-binding and cognition, and that its ease of implementation in analog hardware promises to offer important technological benefits.

An optical neural network based upon the Neocognitron paradigm is introduced. A novel aspect of the architectural design is shift-invariant multichannel Fourier optical correlation within each processing layer. An innovative bipolar neural weights holographic synthesis technique is introduced to implement both the excitatory and inhibitory neural functions. Multilayer processing is achieved by iteratively feeding back the output of the feature correlator to the input spatial light modulator and updating the Fourier filters. By designing the neural net with characteristic features extracted from the target images, successful pattern recognition with intra-class fault tolerance and inter-class discrimination is achieved. A detailed system description is provided. Experimental demonstrations of a two-layer neural network for space objects discrimination is also presented.

One of the most outstanding properties of artificial neural networks is their capability for massive interconnection and parallel processing. Recently, specialized electronic neural network processors and VLSI neural chips have been introduced to the commercial market. The number of parallel channels they can handle is limited because of the limited parallel interconnections number with one-dimensional (1-D) electronic wires. High resolution pattern recognition problems may require a large number of neurons for parallel processing of the image. The holographic optical neural network (HONN) based on high resolution volume holographic materials is capable of providing 3-D massive parallel interconnection of tens of thousand of neurons. A HONN with 3600 neurons, contained in a portable briefcase, has been developed. Rotation-shift-scale invariant pattern recognition operations have been demonstrated with this system. System parameters, such as signal-to-noise ratio, dynamic range, and processing speed, will be discussed.

We describe an optical holographic interconnection method for neural network implementations in which each weight is distributed among many angularly and spatially multiplexed photorefractive gratings. This approach greatly reduces crosstalk and results in efficient use of the input spatial light modulator. Experimental results for neural networks implemented using this approach are also discussed.

We describe a two-layer neural network using holographic optical disks as the interconnection weights. Such a system can be used to implement one two-layer with large number of hidden units, or several two-layer networks with smaller number of hidden units.

This paper describes recent results obtained during the experimental development of a holographic optical neural network based upon the spectrally selective recording properties of spectral hole burning materials. This general architecture has been initially tested as a bi-directional associative memory system (a subclass of neural networks). The results obtained clearly demonstrate the fundamental ability to fully connect two 2D planes of digital information. Expectations are that this architecture can be extended to capacities of 10¹² interconnects or greater in a modest form factor system.

The visual recognition system in animals is an inherently parallel system. Existing very large scale integrated circuit implementations have serious limitations on power consumption and computational speed. We propose a highly parallel optoelectronic system combining elements of the Grossberg and Dobbin vision models and Fukushima's Neocognitron model. This high performance machine vision hardware is capable of distortion invariant recognition. Modeling and implementation of such systems provides insight into the visual recognition function, and also is useful for designing optoelectronic hardware that can be used for many applications.

Radially nonlinear neurons are introduced, and back propagation learning for multilayer networks of these simple hidden units is derived and simulated. The nonlinear transformation performed by a hidden layer of radial units can be represented as a simple multiplication of the summed net input to each neuron by a single value which is only dependent on the total input to the hidden layer. This allows a simple optical implementation, in which a single modulator/detector is able to act as an entire hidden layer by multiplexing the neuron net inputs and processed outputs.

A modified model of artificial neural networks is suggested, the interconnective matrix of which is consisting of summation of matrices T₁ and T₂, where T₁ is the outerproduct matrix of memory vectors and T₂ is the complementary vectors one, respectively. Since all of elements of the interconnective matrix are non-negative values, the model is more suitable for implementation in optics. An optical neural network system is set up in laboratory to implement the model, which is composed of a green LED array for writing, a red LED array for read out, a PROM for recording interconnective matrix, a photodetector for measuring the output and a feedback system. In the meantime it is also executed by computer simulation. The experimental results of associative memory of two, three and four vectors with eight bits show that the model is of more capacity of associative memory, especially for non-orthogonal vector's memory.

In this paper we propose a Feature Enhanced Interpattern Associative (FEIPA) optical neural network. The common part of the stored patterns is regarded as redundance and its contribution in the association process is discarded. Therefore, the output before thresholding is more uniform, and it is more easier for the thresholding performance and increase the iteration speed. Furthermore, the optical implementation is much easier because all the elements of the interconnection matrix are non-negative and unipolar. The theoretical description and the experimental results are presented.

This paper presents new results of our data association (DA) neural net (NN) on measurement-to-estimate data (rather than measurement-to-measurement data). It also uses our detection unit. Our new jitter model is discussed and initial results are presented. This also includes tests of our fixed coefficient estimator and initial optical laboratory results.

We find that: the use of an estimator improves DA MN results (since no clutter is present in the estimate frame), reduction of jitter by using the detection system in tracking helps, our system handles measurement noise and jitter and clutter with no loss of track, and our fixed coefficient estimator suffices.

Proposed is a modification of simulated annealing method to minimize the cost function in combinatorial optimization problems. The modification differs from the traditional method in that, first, a random fluctuation of the cost function feasible at the given effective temperature is evaluated, and then, a few random rearrangements of system elements are executed. Accepted are those rearrangements for which the real change in the cost function does not exceed this random fluctuation. As shown, the probability of finding an optimal or near-optimal solution is improved by increasing the number of rearrangements for one fluctuation, the annealing time being the same. The proposed method is more adapted to parallel computing devices.

The optical implementation of a locally interconnected systolic architecture of a neural network is considered in this paper. In the design presented here, the Hopfield model, one of the widely researched artificial neural network, is formulated as a consecutive matrix-vector multiplication problem with some prespecified threshold operations. The multiplication array structure is derived from a cascaded dependence graph with nonlinear assignment. By the same nonlinear assignment, a locally interconnected systolic array with bidirectional communicational links is then obtained. Each processing element in the systolic array is treated as a neuron and the synaptic strengths are stored in it. The optical design employs a liquid crystal light valve (LCLV) structure to implement the matrix-vector multiplier. The paper will show that the optical and systolic implementation of the neural networks achieves a higher precision in computation.

Optoelectronic realization of adaptive filters and equalizers using fiber optic tapped deIay lines and spatial light modulators has been discussed recently. We describe the design of a single layer fiber optic Adaline neural network which can be used as a bit pattern classifier. In our realization we employ as few electronic devices as possible and use optical computation to utilize the advantages of optics in processing speed, parallelism, and interconnection. The new optical neural network described in this paper is designed for optical processing of guided Iightwave signals, not electronic signals. We analyzed the convergence or learning characteristics of the optically implemented Adaline in the presence of errors in the hardware, and we studied methods for improving the convergence rate of the Adaline.

A neural network pattern classifier is presented. Its decision boundaries are formed from segments of conic sections which allows it to achieve improved performance over piecewise linear neural network classifiers, such as our earlier adaptive clustering neural network (ACNN). We discuss an optical realization that uses complex-valued weights, optical intensity detectors, and an additional input neuron to achieve piecewise conic decision surfaces (rather than the piecewise linear surfaces that the ACNN produces).

Presented are experimental results that demonstrate bipolar optical incremental update in an amorphous silicon/ferroelectric liquid crystal (PLC) optically addressed spatial light modulator. The results are understood in terms of charge compensation in the PLC producing a nonlinear capacitance behaviour. Describing the FLCin this way, the experimental results have been accurately modelled. Also, an optoelectronic neural network, in which this device is incorporated as an adaptable weight storage device, is described and experimentally demonstrated with parallel optical updating of the interconnection weights

Two broad approaches to computing are known - connectionist (which includes Turing Machines but is demonstrably more powerful) and selectionist. Human computer engineers tend to prefer the connectionist approach which includes neural networks. Nature uses both but may show an overall preference for selectionism.

"Looking back into the history of biology, it appears that whenever a phenomenon resembles learning, an instructive theory was first proposed to account for the underlying mechanisms. In every case, this was later replaced by a selective theory." - N. K. Jerne, Nobelist in Immunology.

Recently, there has been some effort to using selectionist methods to optimize neural networks. Farhat (1) used optically generated noise to assist in fast simulated annealing. Shamir et al. (2-4) used Genetic Algorithms (GA) to optimize pattern recognition masks for Fourier processors (a very limited type of fully interconnected neural netwo rk). Both built selectionist-designed connectionist systems.

I hope to show that a true hybrid selectionists/connectionist system of useful proportions can be put together optically. That is multiple, parallel neural networks can be assembled to work in parallel optically. Each seeks to learn the task independently, but those which do best are allowed to exchange information among themselves to produce a new generation which replaces the old.

I will then argue that a parallel, evolving, coordinating neural network population is superior in principle to single, even "optimum," neural network for real world problems and that optics is far better suited for such a system than electronics.

Holographic disk based scheme of the photonic neural network with outer product implementation is described. Some aspects of two-dimensional fourier- holograms registration on the photothermoplastic disk under infrared laser thermodevelopment are considered. Developed approach allows to supply the high information recording density (~10⁵ bits/mm²) for the holographic disk, the high quality of IWM fourier-holograms registration (η=5-7% end contrast ratio ~70:1 ) in addition to the possibility of image reconstruction from the hologram identical to the original image without aberrations.

A new architecture for performing outer product Hopfield Model associative memory has been proposed by using a polarization encoding method. In the architecture, multiplication is performed by rotating the polarization of the transmitted light, and optical associative memory with bipolar binary inputs and bipolar analog interconnect weights can be realized. Also no electronic differential amplifiers need to be used in the architecture, so it may be realized by all optics.

This paper defines the important parameters in an on-module interconnect tradeoff analysis and shows how optics compares to electrical interconnects over the next decade. The analysis is based on latency, power consumption and circuit area, given that bandwidth-distance and electrical isolation requirements can be met by both electrical and optical interconnect in an on-module environment. As a result of the analysis, the paper provides a set of requirements that optics must meet when competing with electrical interconnects at the on-module interconnect level.

An overview of recent work on optical computer design at Heriot-Watt University is presented. The role of parallel electronic machines in the optical machine design process is addressed at both the modular and the system level. Example results of system simulation on both a distributed array processor and a transputer array with implications for general parallel processing are discussed. The systems considered here consist of highly parallel optical units of simple processing capability, running under the control of more sophisticated but less parallel electronic processors. Stress is laid throughout on the use of highly parallel, non-local, free-space, optical interconnects to maximise performance.

This paper describes a scheme for implementing a digital optical matrix processor including an all-optical control unit, using free-space interconnected optoelectronic integrated circuits (OEICs). Many digital optical processors have been proposed in the last decade or so, but if they involve any complexity, the control of the processor is either not discussed, or is relegated to an electronic host computer. An all optical processor has advantages in cost, complexity, and robustness over an optical processor that requires an electronic host computer. We have already developed the basic principles of operation, and several novel matrix manipulation algorithms, including one that accomplishes boolean matrix multiplication of two NxN matrices in N processor cycles, and arithmetic addition of N pairs of N-digit words in N cycles. We estimate the system throughput at over 10¹¹ operations per second.

The system design of a large (512x512) replicated banyan switch for broadband applications at 622 Mbit/s is presented and the physical limitations that arise in electronic implementation outlined. Optics seems the key to solve the problems that make all-electrical implementation impossible for high speed and large size switches. Possible solutions that adopt optics for board-to-board interconnections are here proposed and discussed for their introduction in the switch.

A reconfigurable optical interconnection technique via complex amplitude computer generated holograms (CGH) in electrically-addressed spatial light modulators (ESLM) is presented. Several networks including one-to-many (either regular or irregular) and strength-adjustable interconnections can be implemented by the new technique.

The cost of components for optically interconnected systems, such as optoelectronic device arrays and computer generated holograms, depends on their method of manufacture. In this paper, we build analytic models that measure component cost based on the cost and yield of each manufacturing step. Using these models we show that for large optoelectronic arrays, the most cost-effective implementation is hybrid, not monolithic. Moreover, we show that the most cost-effective interconnection hologram is monolithic.

ImSyn™ is an image synthesis technology, developed and patented by Essex Corporation. ImSyn™ can provide compact, low cost, and low power solutions to some of the most difficult image synthesis problems existing today. The inherent simplicity of ImSyn™ enables the manufacture of low cost and reliable photonic systems for imaging applications ranging from airborne reconnaissance to doctors office ultrasound.

The initial application of ImSyn™ technology has been to SAR processing; however, it has a wide range of applications such as:

•Image Correlation

•Image Compression

•Acoustic Imaging

•X-ray Tomography (CAT, PET, SPECT)

•Magnetic Resonance Imaging (MRI)

•Microscopy

•Range-Doppler Mapping (extended TDOA/FDOA)

This paper describes ImSyn™ in terms of synthetic aperture microscopy and then shows how the technology can be extended to ultrasound and synthetic aperture radar. The synthetic aperture microscope (SAM) enables high resolution three dimensional microscopy with greater dynamic range than real aperture microscopes. SAM produces complex image data, enabling the use of coherent image processing techniques. Most importantly SAM produces the image data in a form that is easily manipulated by a digital image processing workstation.

In this paper, the optical butterfly interconnections have first been implemented in theory and experiment by using the special reflected interconnect gratings and liquid crystal light calve(LCLV) , and two most primitive optical logic operations (AND and OR) have been completed on the basis in parallel . Hence , this work makes the fundamental for more complex digital optical computings.

Photorefractive materials are capable, at relatively low optical power, of recording volume phase holograms in real time with very high diffraction efficiency and of phase conjugating one or more laser beams (either mutually coherent or incoherent). These unique capabilities offer tremendous potential for applications in optical storage, interconnection and signal and image processing. This paper gives a brief review of some device concepts in each of these areas. Selected examples are given to qualitatively explain the basic principle and their potential merit. Technical issues that need to be addressed to improve the performance of various devices are also discussed.

We present the write light intensity threshold characteristics of an OASLM incorporating a hydrogenated amorphous silicon photoconductor and a surface stabilized ferroelectric liquid crystal modulator, LAPS-SLM.

In this paper, we describe a new implementation which applying the two threshold characteristics of the LAPS-SLM to perform all Boolean logic by one device.

A waveguide holographic optical interconnect technique is used for signal distribution operations in a compact 3D wafer packaging configuration. This paper describes the design, fabrication and results of an integrated optical device which utilizes this waveguide holographic optical interconnect technique.

It is desirable to design an optical interconnect so that it has high bandwidth, signal-to-noise ratio, and alignability. Examining the relationship between the alignability and bandwidth and the alignability and signal-to-noise ratio shows that increasing the alignability can lead to a simultaneous increase in the signal-to-noise ratio, but a decrease in the bandwidth. To find a design which balances these opposing effects, the product of alignability, signal-to-noise ratio, and bandwidth is plotted versus the normalized detector size of the interconnect. The peak of this curve yields the optimum detector size for the given physical interconnect design. Using this detector size, the interconnect's signal power and receiver parameters can be altered, as needed, to yield the required bandwidth and signal-to-noise ratio.

We motivate our interest in examining how optics can be used in a class of general purpose parallel computer architectures called Distributed Shared Memory (DSM). We describe an abstract DSM architecture called Beehive that incorporates a weak memory model called Buffered Consistency. We propose a specific optical implementation of Beehive called OBee. This optical implementation uses optical waveguides to implement an interconnection network called Optical Broadcast Rings (OBRs). The OBRs are used in OBee as part of the hybrid electronic/optical hardware support for cache coherency and three types of synchronization (locks, barriers, and combining F&OPs). We also use the OBRs to propose purely optical hardware support for the locks and barriers.

An optical backplane is introduced which is capable of high data rate transmission. It consists of a stack of circular light-guiding plates arranged one on top of another and is therefore called Optical Parallel Plate Stack (OPPS).

In this paper we present the results of an analysis of optical power transmission and signal dispersion within the OPPS system. Measurement results and a hardware realization ofan 8-bit parallel bidirectional optical backplane using the OPPS are reported, too.

A new optoelectronic parallel bus system for fast computer-internal data transmission was developed. This parallel bus system basicly consisting of laser diode arrays (or LED arrays), light-guiding stripes and photodiode arrays allows the substitution of galvanic backplanes nowadays used as computer-internal bus systems. All stripes optically isolated against each other are spatially arranged such as to form an optical parallel stripe plate (OPSP).

The optical transfer function of the single stripe, the corresponding system transfer rate as well as the S/N ratio and the bit error rate (BER) of the bus system have been calculated. The relevant results of calculations, corresponding system simulations and measurements are presented. These results show that the new optoelectronic parallel bus is capable of data transfer rates above 1 Gbit/s on each bit channel.

A 3-D holographic optical memory is described that combines spatially and angularly multiplexed storage to yield a storage capacity of approximately 10¹² bits in a crystal with volume less than 100 cm³. A non-mechanical scanning mechanism, consisting of acoustooptic deflectors and a segmented mirror, retrieves any stored hologram in a time equal to the acoustic delay through the aperture of the acoustooptic deflector

The computational power of current high-performance computers is increasingly limited by data storage and recall rates. In existing sequential-access electronic memories, a hierarchy of devices from cache memory to secondary storage provides a performance continuum, allowing a balanced system design. Here we discuss the use of 3-D volume storage based on two photon materials to bridge the gaps in the storage hierarchy of parallel-access memories.

We describe the first experiments to validate the concept of a high-speed, high-density random access memory. This technique is based on a hybrid of the time-domain stimulated echo and the frequency domain scheme. The basic approach we have developed is the partitioning of the absorption frequency domain into smaller bins, so that each frequency bin stores a smaller portion of information independently. The advantage of the approach is the flexibility it offers in tailoring the memory system to match the computational processor. We have demonstrated storage and retrieval of data for long time in frequency bins separated by 750 MHz. We have achieved data rates of 40 MHz.

Holographic recording for data storage typically utilizes angular multiplexing for effectively accomplishing a 3d packing of data in the media. For large storage systems this results in very complex optical paths. Multiplexing by employing an orthogonal set of phase encoded references is an elegant answer to this problem. The results of experiments where as many as forty holograms are multiplexed in an SBN crystal using this technique are reported. Efficiency versus number of pages for different write times and crosstalk for various configurations is shown.

There is growing interest in optical volume storage because of the potential for terabit/cm3 memories with highly parallel access. At this time however, the technology is in an early stage of development and no clear winner has yet emerged with regard to storage material, geometry or input/output architecture. A major design question concerns the tradeoff between "volume" storage implemented in bulk materials (for example, photorefractive crystals) versus films or stacks of films (for example, bacteriorhodopsin or eleciron trapping materials)'. The latter possibility, which we call quasi-volume storage, opens a much wider class of storage materials for consideration.

One approach to volume storage is bit oriented, similar to magnetic memory but higher in density. The second approach, which concerns us here, is essentially holographic in nature; each data write or read is distributed throughout the material volume on the basis of angle multiplexing or other schemes consistent with the principles of holography. A variety of addressing schemes may be contemplated, but their practicality depends to a large degree on the interface devices available.

In an earlier paper(2), we discussed a recently demonstrated family of dynamic holograms which offer a possible basis for not just one but several programmable diffractive devices required for various data input/output schemes. Dynamic fixed holograms are not themselves erasable storage materials, but rather programmable diffractive elements which complement holographic storage materials by providing interferometric interfacing tools. In this note, we suggest that such dynamic holograms may be particularly adapted for integration with thin planar storage materials to achieve high data storage densities approaching those of bulk materials, but with a much more flexible geometry and optical readin/readout architecture

The steady increase in volume of current and future databases dictates the development of massive secondary storage devices that allow parallel access and exhibit high I/O data rates. Optical memories, such as parallel optical disks and holograms, can satisfy these requirements because they combine high recording density and parallel one— or two—dimensional output. Several configurations for database storage involving different types of optical memory devices are investigated. All these approaches include some level of optical preprocessing in the form of data filtering in an attempt to reduce the amount of data per transaction that reach the electronic front—end

A means for writing and reading information in a 3D volume storage memory is presented. Two structures of a photochromic material are utilized as the write and read forms of an optical storage memory induced by two photon processes. The spectroscopic properties and physical characteristics of these materials and the stability of the two forms are discussed.

The potential use of optical phased arrays to address individual bits within a volumetric memory medium is examined by using phase sensitive ray tracing procedures. We demonstrate that a 200 element linear array with cylindrical microlens collimation is capable of addressing two-photon (λ≈1tm) activated irradiated volumes of approximately 64 μm³, comparable to the volumetric elements that can be addressed via standard crossed-beam two-photon excitation. The two key advantages of using phased arrays is speed of addressing and the elimination of cleaning pulses. Both advantages dramatically improve read/write speeds as well as signal-to-noise performance of two-photon volumetric memories. We examine four different configurations including collimated linear and spherical arrays as well as binary and pure phased two- dimensional arrays. Although phased arrays with the necessary characteristics have not as yet been constructed, we hope that the present demonstration of their potential advantages will enhance interest in the development of phased arrays for application in volumetric memories.

Transparent electron trapping (ET) thin films are optical storage media recently developed by Quantex. A 3-D erasable optical memory based on stacked ET thin film layers is presented in this paper. The page memory is written by properly imaging the page composer onto a specific layer with blue laser light at 488 nm. To read out the memory on a specific layer, a slice of 1064 nm infrared (IR) beam is guided into the ET layer from the side of the stacked layer media. The orange emission (around 630 nm) corresponding to the written page memory at that layer is emitted as the result of the JR stimulation. The feasibility of this novel 3-D optical memory has been demonstrated by preliminary experiments.

Nonlinear properties of chacolgenide glass A52S3 thin films have been measured with a self-diffraction (Z-Scan) technique. Photostructural changes and dynamical effects have been measured.

A polarization vectorial holographic recording medium has been investigated for the development of high performance three-dimensional optical memory storage architectures. The approach has been to fully characterize the molecular and bulk properties of the polarization vectorial holographic (PVH) recording medium, and to optimize material performance for memory applications. Response time, write/read/erase speed, and fatigue of the PVH medium have been dramatically improved. The response time has been reduced to 80 μs. No material fatigue has been observed in over one million read/write/erase cycles. We also demonstrated high-speed (< 1 kHz) write/read/erase cycling. These improvements are the result of a detailed investigation of the molecular photophysics which allowed the optimization of material and recording parameters.

Within digital memory applications, laser scanners are used to either write information on a medium or to interrogate a medium. It is our intent to address the issues of reading high volume optical mediums of digital information at nanosecond random access speeds. Current applications include digital opto- electronic computing, parallel readout of optical disk, volume holographic memory readout, as well as some of the more conventional applications such as robotic vision systems and inspection systems.

High speed digital memory (ECL,GaAs) currently used in high performance computer technologies rely on expensive and power consumptive memory chips. These memories provide only marginal densities per chip (4-l6Kbit). However, through the integration of optical interconnects and volume storage technologies coupled with alternative GaAs structures, an increase in speed as well as a decrease in power consumption can be realized. Thus, increased storage densities and faster access rates can be achieved. The laboratory prototype discussed in this paper has demonstrated that random beam deflection in the nanosecond regime is possible using tunable laser diodes and a dispersive medium.

A possible approach to a volume optical memory system is a page addressed or bit plane architecture. Bit planes of data are stored in the memory medium analogous to pages in a book. Each bit plane consists of a matrix of spots like letters on a page. A two-photon memory mechanism is used and the memory medium is assumed to be in the shape of a cube. Orthogonal beams will be used to locate a point within the cube. The memory architecture would consist of an input spatial light modulator, dynamic focus lens system, optical memory medium and detector array. Both read and write operations require that a bit plane of data be imaged to a specific plane within the memory medium. Read out would be accomplished in a similar way using the dynamic lens, second orthogonal beam and a detector array. The dynamic focus lens used here is the liquid crystal adaptive lens (LCAL). The LCAL focuses light by electrically grading the refractive index across its aperture. By changing voltages of the discrete electrodes the focal length is changed, thereby creating a dynamic lens. This presentation will discuss the LCAL's perfomance as applied to a page addressed volume optical memory.

We present the design, implementation and performance of a holographic dynamic focusing lens (HDFL). This device was developed to focus a 2D array of point sources to any of the many discrete planes inside the volume memory material. The lens utilizes two components: a spatial light modulator (SLM) as an active element which allows fast switching (≈ 1 μsec), and a diffractive optical element which allows for high SBP. The unique design feature of this device is that the resolution of the lens is not dependent on the resolution of the SLM. Simulation and experimental results are presented.

We discuss the different contributions to aberrations in the liquid crystal adaptive lens, specifically those due to errors in voltage settings at the electrodes (which could be dithered) and those due to quantization errors of the electrodes and meshing errors between electrodes (which are fixed upon device construction). For application in a large optical memory, the dominant errors are the fixed errors. Linear current meshing is introduced as a solution to these errors because a current-carrying plate joining adjacent electrodes provides a linear interpolation of index between the electrodes.