We have designed and fabricated 1 X 8 and 4 X 8 VCSEL arrays at 850 nm and 980 nm operation wavelength, respectively, which are designed for maximum single-mode output power and high frequency applications. GaAs VCSELs in the 1 X 8 array show record high single-mode CW powers up to 4.8 mW. Individual devices of the 4 X 8 InGaAs VCSEL array exhibit small signal modulation bandwidths exceeding 10 GHz.

We demonstrate a novel design that allows fabrication of substrate input/output photodetectors and lasers on the same substrate by selective oxidation of the bottom mirror.

A novel high-speed optoelectronic gate, based on electric field screening and ultrafast electrical recovery, is proposed, and demonstrated in principle with ultrathin barrier quantum wells.

Not Available

Not Available

A high-speed optical photorefractive correlator using a ferroelectric spatial light modulator and a new type of binary filters optimized for the crystal nonlinearity will be presented in terms of characteristics and performances.

A new snake based technique for target segmentation is presented. This approach can be implemented with optical correlators and thus enlarge their field of application.

We compare various techniques for the implementation of correlation filters in the 4f architecture and demonstrate that oversampling certain parts of the filters can improve performance.

Not Available

We demonstrate a 2.5 Gb/s optical intra-MCM data link with a four-channel micro-optical bridge. This bridge was fabricated by deep proton lithography and monolithically integrates cylindrical lenses and micro-mirrors.

Free-space optical interconnects represent a candidate technology to alleviate some of the interconnection bottlenecks currently experienced by high performance electronic systems at the board to board and backplane level [1,2,3]. The two most widely investigated technologies for free-space optical data transmission are multiple quantum modulators (MQW) and arrays of vertical cavity surface emitting lasers (VCSELs). Although VCSELs have many desirable optical attributes, large arrays of VCSELs which can be hybridly attached to CMOS integrated circuits are not yet available. This has motivated the design of modulator-based systems. Recently a ring-based free-space optical interconnect [4] has been designed for optical backplane applications [1]. In this system optical data transmission is achieved by the use of GaAs multiple quantum modulators (MQW) which are used to modulate an externally generated spot array. Here we describe design and assembly of an optical array generator which is used as provide the spot array for this system. This is referred to as the optical power supply (UPS) module. The ring-based system is designed to be scaleable (to 4 or more boards), modular (to allow easy assembly and maintenance), compact and also robust and misalignment tolerant. The UPS for each stage of the system is designed as a separate module that is pre-aligned and then mounted onto the system with minimal additional adjustments. A kinematic mounting technique has been developed to enable the UPS to be demounted and remounted without further realignment.

Low-order adaptive optics is a design option to achieve realistic low-cost free-space optical interconnects. We are investigating the issues using a system demonstrator.

We present the design, fabrication, and test of a 90 degree(s) polarizing beam splitter grating for 633 nm. This component is compact and its fabrication is reproducible enough to be integrated in optical reconfigurable interconnect systems. For TE polarization, the incident beam is undeflected whereas TM beams exhibit a 90 degrees deviation. We used electromagnetic vector theory to optimize diffraction efficiencies and the extinction ratio. We employed direct electron beam writing and reactive ion etching to fabricate a polarizing beam splitter etched in a photoresist layer deposited on a fused silica substrate.

Theoretical and experimental investigations of the intensity- and field-dependent photorefractive response in polymeric composites were presented. Measurements of the decay time constant and the corresponding grating-amplitude were demonstrated in photorefractive composite consists of 1-n-butoxyl-2,5-dimethyl-4-(4'-nitrophenylazo)benzene: poly(N-vinylcarbazole):2,4,7-trinitro-9-fluorenone with a weight ratio 44:55:1. Analysis suggested that two gratings exist simultaneously, one was caused by the photorefractive effect, and the other was caused by the photoisomerization process.

Not Available

Not Available

Not Available

Not Available

Not Available

We propose to use dynamic ternary phase-amplitude modulation for free-space reconfigurable optical interconnections, and demonstrate its significant advantages over the widely- adopted binary-phase scheme.

Features of light are compared with those of electrons and other possible information carriers from a view point of computing and information processing, especially when they are used in future brain-type systems. It is described that the use of wave aspects of light leads to a realization of ultimately parallel and adaptive information processing systems. Two architecture examples utilizing such lightwave features are presented and discussed: (1) coherent optical neural networks and (2) coherent high-frequency clock delivery.

We propose a new spatial-phase-code division multiple access system based on Fourier-holography switching. The proposed scheme was experimentally verified with double-multiplexed hologram.

A 2D VCSEL/CMOS array and a GaAs smart detector array are used in the design and implementation of a high performance optoelectronic ATM switch based on 3-stage cross-point switching architecture.

Optical intercotmects have the potential to overcome the limitations encountered in present electronic backplane technology in providing the information throughput required in ever faster multi-processor computers and lelecommunication switching systems[lJ[2]. A major challenge in this area is the design of robust, misalignment tolerant and field-serviceable optomechanical hardware for optical and optoelectronic components. Components often need to be precisely positioned to tolerances in the micron range laterally and fractions of a degree angularly. Alignment must be maintained despite vibrations, temperature variations and the occasional breakdown and maintenance cycle inherent in an industrial environment. This paper describes the implementation of a novel four-stage clustered optical interconnect designed for use in optical backplane applications[l]. The system optical design is first reviewed (more details can be found in[3]) followed with calculated and measured optical power throughputs. A tolerancing analysis illustrates the benefits of partitioning the system in pre-aligned modular building blocks. This is shown to minimize the number of critical alignment steps and considerably simplifies system assembly. It is then shown how proper optomechanical design allows for the passive insertion of modules to complete system assembly.

We show how to calculate exactly the number and strength of the connections between two arbitrary volumes, and derive and apply a novel sum rule.

We present a morphological algorithm for compression of images. It is a modification of the morphological subband decomposition algorithm proposed by Pei and Chen.

With the prospect of a "billion transistor" microprocessor chip becoming a reality in the next decade, VLSI photonics will be an important technology for high bandwidth, chip level, I/O. In this paper we describe a system which demonstrates the use of free space optical channels to connect the functional units in a super scalar microprocessor. This approach enables architectures in which the number of functional units available for parallel instruction execution is significantly larger than can be implemented in a purely electronic design. The design is implemented with SEED devices, flip-chip bonded on a O.5prn CMOS silicon chip1 and is currently being fabricated as part ofthe 1997 CMOS-SEED Coop program2. In a super scalar microprocessor, high performance is achieved by executing multiple instructions in parallel. The architecture consists of multiple functional units, each capable of independent execution, with source and result operands delivered via local interconnection busses. During program execution, a control unit works on a buffer filled with instructions that are eligible for execution and selects those instructions that can be executed without a conflict for resources or data. For example, two instructions may be in conflict over a specific functional unit, a bus, or a port to memory or a register file. These types of conflicts are collectively referred to as structural hazards. There may also be a conflict caused by data dependence within a sequence of instructions. In other words, the operand of one instruction depends on the result of another. These are called data hazards. Other conflicts can be caused by uncertainty over the outcome of a conditional branch instruction. In this case it may be unknown whether or not a particular instruction will be executed at all. This is called a control hazard. These conflicts place a limit on the performance of super scalar machine by limiting the number of instructions that can be executed in parallel. In contemporary designs, architects have attempted to circumvent this limit by building additional functional units. This has an obvious impact on structural hazards but can also be effective on data and control hazards when speculative or redundant execution techniques are used3. For example, if a control hazard introduces uncertainty about the outcome of a conditional branch, both execution threads are allowed to proceed until the uncertainty is resolved. At that time, the computation from the untaken branch is simply discarded. Similarly data dependencies can be resolved by speculating as to the result of a dependent computation and discarding an execution thread ifthe guess was wrong. In general, the more speculative instruction execution that is possible, the greater the effective level of instruction parallelism. However, the number of functional units that can be built and connected on a single chip limits electronic designs. As an alternative, we are suggesting a design which implements free space optical channels as the interconnection busses in a multi-chip super scalar system. These high-speed interchip busses allow us to create systems where the number of functional units is significantly larger than can be implemented in a purely electronic design. In this paper we describe a prototype system which implements six integer functional units and three registers files in a three-chip super scalar ALU design. The control unit implements a subset of the MIPS RS-2000 instruction set architecture and is capable of full dynamic (runtime) scheduling ofALU resources. The rest of this paper is organized as follows. We begin with a logical description of the optical bus structure between the chips and the optoelectronic interface. This is followed by a description of the internal organization of the ALU chips. The optical system used to implement the optical interconnect is presented followed by simulation data showing the performance of the optical interconnect. Finally, we give a briefoutline ofour future research.

An optical VanderLugt correlator for real time recognition of bytes at 1.55 micrometers is presented. The discrimination results with phase only filter are shown.

Header Error Control in all-optical ATM switching nodes is discussed. An architecture of an error detection subsystem is designed suitable for free-space parallel optical implementation.

Characteristics of large fan-out interconnection are studied in terms of signal propagation delay and power consumption. Four types of interconnection models are considered and their properties are formulated. The obtained result supports the effectiveness of optical digital computing schemes based on large fan-out interconnection such as optical array logic.

We propose an optoelectronic parallel-matching architecture enhancing processing capability of parallel computing systems. The parallel-matching architecture provides concurrent data search and synchronization as well as fast communication capability.

An architecture for optical CAM-based parallel database processing is presented. It features compact design and constant-time parallel processing of database operations using wavelength and polarization division multiplexing techniques.

The increasingly high performance of electronic processors will place a burden on data communications in future systems. High speed and dense interconnections will be needed at various levels of a system hierarchy: among gates on a chip; among chips on a multi-chip module (MCM); among chips or MCMs on a board, and among boards via a backplane.

We demonstrate a reconfigurable data transparent optical fanout operation and a switchable digital optical logical inverter between planes of optical thyristors using polarization-selective diffractive optical elements.

The reconfigurability of optical interconnects is often translated in polarization-selectivity. To this aim, polarizing highspatial frequency gratings in GaAs [1], polarizing holographic optical elements in DCG [21, polarization-selective computer-generated holograms in LiNbO3 [3] and polarization-selective diffractive optical elements in calcite [41 have been developed. Polarization-selective, or anisotropic, diffractive optical elements (ADOEs) implementing calcite in combination with an index-matching polymer have proven to have several advantages over LiNbO3 [41. Recently we implemented ADOEs in a free-space reconfigurable optical interconnect using liquid crystal retarders (LCR) to control polarization of light [5]. The reconfiguration rate in this case was limited by the switching speed of the LCR, which is 20Hz in the case of a nematic LCR, and 20kHz in the case of a ferroelectric LCR. Very recently electrically controlled polarization switching in VCSEL's has been demonstrated up to 50MHz [6,7] with a polarization contrast ratio of 20:1 [7]. Here we report on the implementation of these polarization-switching VCSEL's in a reconfigurable interconnect demonstrator based on ADOEs. We present reconfigurable interconnects at 30MHz, and we demonstrate the proof-ofprinciple of a data transparent reconfigurable interconnection scheme with a bit rate of 1MHz and a reconfiguration rate of 40kHz. Both the bit rate and the reconfiguration rate are limited by the available electronic source of modulation.

We present a high speed compact optical system with a large field of view, dedicated to optically interconnect planes of microemitters and receivers.

We review recently reported radiation effects on photonic devices intended for space application: LEDs, laser diodes, VCSELs, liquid crystal modulators and acousto-optic Bragg cells.

We present an optoelectronic stochastic artificial retina for binary image restoration, combining a spatial light modulator, a CMOS circuit and an optical random number generator.

Here, a prototype for free-space optical interconnection in massively parallel processing is presented. Physical and optical models of the core of the prototype-image distributor are discussed. The general features and the performance of the image distributor are also described. From the digital simulation and experiment, it is found that the multi-faceted free space image distributing technique is quite suitable for free space optical interconnection in massively parallel processing.

Since the arrival of new devices such as the spatial light modulators based on ferroelectric liquid crystals' , optical pattern recognition has known a revival. Almost all applications in optical pattern recognition are concerned with the recognition of objects in real images with industrial and military applications. For these applications there are major problems to make filters which are able to cope with rotations and scale-changes in the input-image2 along with the presence of noise in the image under investigation. Another recent application is in the field of securing information3. In this article we present a completely new application : we use optical pattern recognition for the recognition of a particular DNA-sequence in a very large DNA-string. We aim to code a genome sequence which is composed of DNAsequences, polymers build up by 4 different organic bases called : Adenine (A), Thymidine (T), Cytosine (C) and Guanine (G). The sequence of these bases in the polymer compose the coding of all biological information necessary to reproduce and grow as a living organism. The human genome is composed of 2.8 billion such bases, a typical virus like HIV contains about 9000 bases. The occurrence of these bases in the large DNA-strings is depending on the kind of DNA but in many cases this occurrence is almost random for the 4 bases. At this moment large efforts are done to resolve the complete sequence of the human genome and a lot of other living organisms. An important topic in genetics is the search for functional sequences in very large DNA-strings to acquire a better understanding of life and diseases. Nowadays, these searches are mainly performed by artificial intelligence techniques4 . In this article we show that optical pattern recognition can be a very effective and very fast alternative for these searches.

Study of pattern recognition technique using optical correlation has a long history. However, the technique has not been put to practical use yet. The main reason is that the amount of information included in temporally and spatially changeable images from the real world is too large to be processed by a single optical correlator. Another reason was a lackoffunctional optical or optoelectronic devices, such as spatial light modulators, micro lens arrays and smart pixel devices. However, functional optical devices have been developed andbecome available in optical systems. Nowadays it becomes possible to realize an optical computing system surpassing electronic system in processing speed, by applying the speed and parallelism of light. We think it is possible to realize a machine vision system which shows an adequate ability ofpattern recognition au! processing speed by a multiple optical correlator' using a set of optimized correlation filters and functional optoelectronic devices. In this paper, we designedsets ofoptical correlation filters fordetection andclassification of road signs in a image of real world scene, in order to evaluate the ability of machine vision system using multiple optical correlator. The correlation filter set is designedas to have partial invariance fordistortion to adaptto the change of aspect of road signs. Computer simulation result shows that the combination of multiple optical correlator and partial invariant correlation filter can indicate high performance of pattern recognition.

A non-linear pre-processing operation for pattern recognition is described. This operation is shown to provide excellent suppression of clutter and superior results to the linear counterpart.

The reduction of geometrical distances in optical correlators, in order to build miniaturized systems, has become a necessity for realistic applications [1]. In general, the scope of the studies carried out is only focused in some parts of the problem. Nowadays, liquid crystal devices (LCD), usually removed from commercial videoprojectors, are used to display images in optical setups. As a result of using pixelated panels, the viability study of building an optical correlator in a reduced space should not be only limited to a design of optical systems. It must include an analysis of the behaviour of the displaying devices. Another issue that has not been raised yet is the reduction of the material needed to control simultaneously two LCDs in a Vander Lugt correlator, which involves a duplication in the driving electronics. We analyze in depth several engineering problems regarding the construction of a Vander Lugt correlator with LCDs at the input and Fourier planes. In order to reduce the length of the setup we have designed two telephoto systems. An original method to control both devices with a single 8-bit frame grabber and a single videoprojector electronics is presented. We have solved the problems related to pixel-by-pixel control when displaying the data from the frame grabber on the LCDs. An accurate analysis of the phase modulation capability of the panels is also discussed. Experimental results obtained with this setup are presented.

The application of a phase-encoded joint transform correlator to decryption of an optical image encryption system by a double-random phase encoding is described.

Intensity invariant pattern recognition is achieved by a normalization of the correlation signal. Fourier for segmented and multiobject input are derived with the Holder's inequality.

Multi-class correlation filters (POF and MACE) with carrier frequency, which are adequate for shift invariant associative memory are analyzed and compared with classical filters.

In this work we analyze the effects of signal-dependent noise in the input scenes of optical correlators. We propose several algorithms to process this noise and we evaluate their performance. In particular we compare the strategy of restoring the input scene with noise and perform the correlation, with the alternative of correlating directly using filters designed to take into account the noise model in the input image.

A method based on the angular correlation of the Fourier spectra of fabric images is proposed to automatically evaluate web resistance to abrasion.

A new encoding technique for Joint Transform correlators is presented, with no loss of information for correlation filters displayed on non-ideal SLMs, and narrower autocorrelation peaks.

A nonlinear correlation detection method for improving the selectivity in pattern recognition is presented. The new approach is based on morphological correlation. We propose a localized slicing carried on low and high component Fourier domain information.

The iterative generation of holograms for its specific use in optical correlators is presented. This permits a optimal utilization of the available resolution in the holograms, and the optimization of the desired figures of merit. Computer simulations shows that holograms generated by direct binary search and simulated annealing exhibit a better performance than cell oriented methods when used in a correlation process.

We present a compact optical correlator with Internet access. Users can remotely download images and get the optically computed correlation results back on their monitor.

A compact optical novelty filter based on two-wave mixing in a non-linear photorefractive crystal is implemented. The performances of the crystal, a cobalt doped barium titanate (BaTiO₃:Co), in terms of depletion efficiency and response time are investigated. It is shown that the spatial light modulator involved in the system causes a drop of the wave-mixing. The system capability to detect cyclic new events, to track motions and to enhance the edges is demonstrated. In the case of a cyclic novelty detection, the cut-off frequency of the system is measured for several incident intensities. The outline characteristics of a moving object are studied versus its velocity and for a set of incident intensities and show, for the first time to our knowledge, to be in good agreement with the theoretical predictions.

We introduce a system for optical encryption of 2D page- oriented data. The system used is an anamorphic optical correlator. The data is scrambled by means of a code on a 1D hologram which is also used for descrambling the data. The principle is somewhat similar to a 2D Rubik cube or a 2D combination lock: the page is sliced first into horizontal slices which are shifted laterally with wraparound. The same process is then carried out vertically and repeated. Computer simulations and optical results are presented.

Memory interconnect has become increasingly important for the electronics community since memory access times have not kept pace with the increase in speed of the central processor. The gap in relative performance between processors and DRAM memories has grown unabated by 50% per year over the last five years. Although cache memory can partially hide the main memory access time, the main bottleneck at the memory level remains unaffected. DRAM is to stay as the main component for high-capacity memory but its cycle time is not likely to decrease by much due to the very nature of the storage process (capacitive charge/discharge). The parallelisation of DRAM chips has increased the aggregate bandwidth but the latency time of the first word reaching the memory has decreased only slowly. A common solution chosen by all memory manufacturers and computer designers has been to widen the bus between the core processor and the (cache) memory in order to satisfy the bandwidth exhibited by the processor I 1 1. This strategy might not be the most efficient one. To illustrate this statement, consider a block transfer of 512 bits, say, which corresponds to the transfer of a data block between a core processor and its cache memory. An hypothetical new technology allows us to trans!èr data at a high-bit rate such as 500 MHz or 10 GHz for example. In the case of a cache, the address of the requested block would have to be submitted to bus arbitration before being sent to memory. Table I shows this typical case-study for a bus arbitration of IOns, an address to be transferred to the memory of (N)=40 bits, a memory read (and write) time oi(RW)=4Ons and a data block transfer of(M)=512 bits.

In this paper an application of optical signal and energy transmission in the field of biotechnology is presented: A retinal implant, which is powered and provided with digitally coded information simultaneously by free space optical interconnection.

Spatial solitons are self-trapped optical beams that propagate without changing their spatial shape, since the diffraction and the nonlinear refraction balance each other in a self-focusing . In this paper we study the behavior of a soliton beam in a waveguide which, in the plane between the cladding and the substrate, has a distribution of refractive index that follows a trapezoidal curve considering the possibility that the soliton propagates in an oblique direction.

In this paper an optoelectronic system for fast greylevel image decomposition on binary slices is presented. A frame rate decomposition of 6 greylevel image has been obtained at 1.8 kHz.

Photonic-based A/D converters have received and continue to receive considerable attention as an alternative approach to providing increased resolution and speed in high-performance applications.

Ultrashort pulse laser technology has recently experienced significant advances, producing high peak power pulses of optical radiation few femtoseconds in duration, corresponding to only few cycles of its fundamental frequency. The future progress of this area is inevitable due to the unique properties of ultrashort laser pulses that are crucial for various science and engineering application areas including optical signal processing, secure optical communications, storage, medical and biomedical imaging, chemistry and physics. A common feature of these applications relies on our ability to control the shape of the ultrashort pulses15, store and retrieve them6 as well as, conversely, our ability to detect the shape of the ultrashort pulses7. In this paper we discuss examples on utilizing ultra-fast space-time optical processing for implementing data formats suitable for direct interface and transmission through an optical fiber. Examples on using femtosecond laser pulses for information storage and memory access as well as modulation and detection of information will be addressed. Specifically, we present our recent results on femtosecond pulse imaging by nonlinear three wave mixing and 3-D nonvolatile spectral domain storage of femtosecond pulses. There exists a bandwidth capacity mismatch between optical fiber and electronic devices, which can be used to increase the speed, reduce latency, increase security and reliability in the transmission and distribution of information. To implement these applications, an alloptical multiplexer performing space-to-time (i.e. , parallel-to-serial) transformation at the transmitter and a demultiplexer performing time-to-space (i.e., serial-to-parallel) transformation at the receiver will need to be constructed810. For efficient bandwidth utilization, these processors need to be operated at rates determined by the bandwidth of the optical pulses. Such space-time optical processors have been constructed and applied for pulse shaping, filtering, and space-to-time multiplexing and time-to-space demultiplexing11'12. Another example exploits applications benefiting from an optical memory that will store and retrieve information in a format suitable for direct interface and transmission through an optical fiber network, thereby, providing optimal performance in terms of hardware complexity, memory and network capacity, bandwidth, and latency. In this example a spatial image information need to be converted into time domain. The corresponding data sequence in time is stored employing spectral domain 3-D volume holographic recording13. When the stored data is read out of the spectral domain storage system, the output spectrum is converted back into time sequence and sent through the all-optical fiber network to the user node. At the user node the time sequence is converted to lower rate parallel channels in space domain for optical or electronic filtering and detection.

Phased array antennas (PAAs) offer many advantages including steering without physical movement, accurate beam pointing, increased scan flexibility in two dimensions, precise PAA element phase and amplitude control to obtain low sidelobes, and reduced power consumption and weight. The complexity associated with controlling the many thousand array elements, while handling the broad bandwidth required of a shared antenna, makes the marriage of photonics and microwave radar attractive. To satisfy the simultaneous requirements of wide bandwidth and large antenna scan angle, true-time-delay (TTD) steering techniques are required so that efficient elemental vector summation (in the receive mode) or distribution (in the transmit mode) can be obtained that is independent of frequency or angle. We describe here a new lower complexity and cost photonic TTD system which we call WDM Delay Broadcasting TTD. This approach (1) encodes fixed delays with wavelength, (2) makes available all of the delays at each of the subarrays, and (3) selects the appropriate delay using an optical filter (with correction via the element electronic delay line). The results of this effort show that it is indeed possible to design and construct a photonically controlled TTD network that will exhibit the performance and cost benefits needed for widespread deployment in government and commercial phased array radar antenna systems.

An efficient optical architecture for implementing a time delay optical neural network utilizing acoustooptic devices, scrolling CCD detector arrays, and photorefractive weight storage is presented.

We propose the novel optical 2D-space-to-time-to-2D-space conversion technique by using an ultra-short pulse laser, in which the time-space conversion technique and the time- frequency transform and their inverses plays an important role.

One can obtain useful approximations of linear systems by implementing them in the form of multi-stage or multi- channel fractional-Fourier-domain filters, resulting in space-bandwidth efficient systems.

We demonstrate a polymer fiber based optical clock distribution circuit on a chip-populated printed circuit board. Various performance parameters are presented.

We demonstrate an OPTOBUS-based opto-electronic cross- connector. The maximum optical insertion loss of the compact 100 X 100 cross-connect box was measured to be 2.8 dB. BER of < 10^-12 can be maintained at 900 Mb/s per channel bandwidth.

A 64 channel optical array interconnect based on 125 micrometers polymer optical fibers arranged in an 8 X 8 array with dimensions of 2 X 2 mm² is presented. Typical insertion losses of 1.5 dB (660 nm) and 5 dB (980 nm) over 20 cm interconnection length have been measured.

We present a design for an opto-electronic bus systems based on commercial parallel fiber ribbon transmission systems and opto-electronic VLSI smart pixel arrays. We describe the design of the individual components, their integration, the performance limits and potential applications.

Parallel optical data links using high efficiency Microcavity LEDs and small diameter POF ribbons are proposed. MCLED performance, coupling efficiencies, alignment tolerances, POF end facet termination techniques and first link experiments are studied.

Liquid crystals (LC) are frequently used as active holographic elements for beam deflection applications. Thanks to their small response time, ferroelectric liquid crystals are mostly studied for telecommunication applications [ 1 ,2] .However, nematic liquid crystals (NLCs) have a very good technological maturity and easily allow multi-phase level diffractive structures with high diffraction efficiencies [3,5]. Thus, NLCs constitute an attractive option for applications like routing where relatively slow (tens of milliseconds) reconfiguration times are sufficient. Previously, we reported devices operating in the 1.5 tm window, with a diffraction efficiency of 70% for deflection angles up to 4° [61. This communication deals with the design of N x N reconfigurable interconnects using this type of devices. The proposed solution satisfies the two major constraints such system has to fulfill, namely the accurate control of the deflection angle needed when connecting single mode fibres, and the finite precision achievable with electrically addressed holograms. Even though the highest interconnection capacity will be afforded by using two-dimensional deflection, our analysis will mainly focus on onedimensional deflection because the key elements of the systems (arrays of single mode fibres and high resolution NLC spatial light modulator SLM ) are presently easier to realise.

Recent rapid advances in technologies such as modulators, vertical cavity surface-emitting laser (VCSEL) arrays, smart-pixel processing elements (PEs) integrating electronic circuitry with optical inputloutput channels, and micro-optic components has allowed the construction of parallel optoelectronic computing systems. These systems utilise the high bandwidth, high density and global nature of optical communication paths to overcome some of the limitations of conventional interconnections. Globally interconnected parallel optical processors are suitable for applications such as sorting, FFTs, and signal processing, and high level image processing. The latter is of interest in our laboratory, specifically with its application to machine vision. However, our system, to be described below, is a general purpose machine due to the programmability of the processing elements (PEs) and by reconfiguring optical interconnections. We can therefore implement a variety of operations on our system: image processing, arithmetic, matrix operations, sorting, and signal processing.

Self-routing `concentrators' are fundamental building blocks of optical switching systems. An N-to-M concentrator can process and extract data packets from N optical channels and forward the packets to M electrical channels, where typically N M. Terabit Optical Backplanes which exploit free-space optical data links, with bandwidths approaching 1 - 10 Terabits per second will require extremely fast self- routing concentrators which can make routing decisions within a few nanoseconds. In this paper, a VLSI analysis of a new circuit called the `Daisy Chain' concentrator is presented. This concentrator has a regular topology suitable for very efficient VLSI layout, which leads to very high clock rates. The analyses are performed using 0.8 micrometers standard cell CMOS technology with the Synopsys CAD tool. The results shows that the proposed concentrator uses substantially less VLSI area from 20 - 50% less in the control logic and up to 150% less on the switching logic than the previous best known concentrator circuit. It also performs significantly faster, ranging from 20 - 40% faster in the control logic and 150 - 300% faster in the switching logic. Using 0.8 micrometers CMOS technology, the proposed concentrator can be used in smart pixel arrays for optical backplanes with clock rates in the range of 500 Mhz. Using faster CMOS or ECL logic, the concentrator can support clock rates in the several Gigahertz range.

The interconnection network is the communication backbone of a parallel processor system on which all remote data accesses occur and, thus, has a strong influence on the overall performance of the system. While the bandwidth delivered by conventional electronic-based networks has increased slowly in recent years, the bandwidth demanded by processors has increased at a much faster pace, soon causing the network to become a performance bottleneck. Optoelectronic-based networks can potentially provide much higher bandwidth capacity to mitigate this problem [1,2]. However, with the present growth rate of communication demand of distributed multiprocessor systems, even a high-bandwidth optoelectronic network (see Figure 1), could become oversaturated unless advanced routing techniques are incorporated in the interconnect architecture to efficiently utilize the bandwidth.

We present the architecture and optical design for a smart pixel optoelectronic system that performs very high throughput networking and three-dimensional single-instruction multiple-data (SIMD) parallel signal processing. The smart pixel chip in the system is called a TRANslucent Smart Pixel Array (TRANSPAR), and it contains a 2-D array of fine-grain processing elements with additional functionality for performing carrier-sense multiple access with collision detection (CSMAJCD) networking. This chip is currently being fabricated by Lucent through the DARPA/GMU/CO-OP foundry program in 0.5 micron CMOS with flip-chip bonded multiple quantum well (MQW) optical modulators and detectors (II.

The use of inter-chip optical interconnects can lead to dramatic increase in chip-level and system-level performance of high-performance computing and signal processing systems. In this paper, we describe our efforts to build chipsets for these applications using optoelectronic VLSI. The performance advantages of using optical interconnects are compared with the conventional approach.

Presented is a novel design for opto-electronic integrated optic devices utilizing a nonlinear optical polymer core and conductive polymer claddings that will potentially have an order of magnitude shorter interaction length than conventional NLO polymer OE devices operating at TTL voltage levels and allows for in-situ poling.

Threshold analysis of nitride VCSELs is performed. Double- heterostructure VCSELs are found to be much less sensitive to optical losses than their quantum-well counterparts.

A number of approaches have been used to implement programmable add/drop optical filters for wavelength division multiplexed (WDM) optical communications. These include integrated-optic (10) acousto-optic tunable filters (AOTFs) [1], all-fiber mechanically tuned fiber Bragg grating (FBG) devices [2], 10 grating switch with 10 coupler devices [3], array waveguide grating (AWG) multiplexer with 10 thermo-optic switches [4], FBG devices with magnetic field tuning [5], JO Bragg gratings with multi-mode interference couplers [6], NxN wavelength grating routers and lxN mechanical switches [7], AWG multiplexer with manually simulated 2X2 switches [8], free-space diffraction gratings-based filter using a linear array twisted nematic liquid crystal device [9], 10 electro-optically controlled synthesized grating structure based filter [ 10], and bulk dual AOTF-based structures [ 1 1] .It is highly desirable to have a short reconfiguration time (e.g., 10 is), low optical crosstalk (e.g., -40 dB), low drive power (e.g., tens of mWs), low loss (e.g., 5 dB), add-drop filter. In addition, a low cost modular and scaleable filter design that is easily modifiable and repairable is also desirable. So far, to the best of the author's knowledge, no one filter satisfies all the mentioned requirements. In this paper, we propose three filter architectures based on high speed ferroelectric liquid crystal (FLC) devices, that have the potential to meet the above requirements.

Hybrid SEED technology is used for the realization of a binary neuronal associative memory. The architecture profits from 200 parallel optical data channels available on top of the silicon. The project is in progress and we present first results.

We will descrihe the role of holographic memorv in a current research effort 1 that seeks to combine ariuus advanced technologies to achieve petaflops scale computing within the next decade In addition to holographic memory. the petatlop architecture combines superconductor Rapid Single Flu-.: Quantum (RSFQ) logic. which can operate at I 00 GHz within a cryogenic cm ironmem with power consumption less than 'iO watts. a packet-switching optical network with a multi-le el strncture capable of providing interconnection among tens or thousands of pons '' ith latencies of only I 0 to 30 nanoseconds. Processor-In-Memory (PIM) technology. and a multithreaded hierarchical structure (see Figure I) to allow the processors to access a high capacitv memnrv vhile compensating for the latenc)· problem inherent in such a system.

We develop a novel design of diffractive optical elements fabricated by an electron-beam lithography for free-space optical interconnection. This element is synthesized by multi-functional computer-generated holograms with the over- write grating. Diffractive optical elements including Fraunhofer and image type holograms have been fabricated. The concept and the fabrication process of this element are presented.

Multi-level electronic integration of massively parallel computers leads to latency effects and 110 bottlenecks. Free space optical interconnections might be a good alternative to interconnect processors located on face to face PC boards. When the processors are integrated in MultiChip Modules, an obvious improvement to the latency problem can be making the MCMs to communicate from board to board. We present here part of a project which was led in collaboration with the CEA ILETIIDEIN(Commissariat a l'Energie Atomique ) in Saclay ( France ). The LETI DEIN was in charge of the conception of the parallel architecture, of the design of the MCMs and of the electrical architecture. The Optics Department of ONERA CERT has to make the choice of the optoelectronic components, their caracterization and hybridization.

The potential advantages of volume optical storage were recognized almost 30 years ago and since that time many research efforts have focused on improving the underlying materials and devices for use in such systems. Recent progress in these supporting technologies have made possible several system-level demonstrations of volume optical storage. These recent demonstrations have verified the feasibility of volume memory systems for offering large volumetric storage capacities, fast access times, and very high data transfer rates realized via the parallel two-dimensional (2D) nature of the stored data. The success of these volume storage testbeds have served to ignite additional research into supporting two-dimensional or page-oriented interface technologies such as parallel 2D data detection and error correction. This application therefore provides a strong impetus to study traditional communication theoretic topics such as signaling, equalization, coding, etc. , in the context of highly parallel 2D channels. The storage and retrieval processes associated with holographic optical memory systems are complex. In order to model a specific memory architecture (e.g. , 90 degree Fourier plane photorefractive memory), many physical parameters must be specified (e.g., focal lengths, material constants, modulator and detector details, etc.). We have elected to utilize a simplified model of these systems for equalizer design purposes, in which the recording and retrieval physics are captured through the behavior of a "channel." We consider a coherent optical channel (i.e., linear in electric field) suffering from finite-contrast and post-detection additive white Gaussian noise (AWGN) . This coherent system produces a measured intensity that represents a quadratic function of the stored data, complicating the simple 151 channel model that is familiar to traditional communications applications.

Novel optoelectronic memory characteristics were successfully demonstrated by a symmetric triangular-barrier optoelectronic switch. We have confirmed that stored data can be held about 10 minutes without applied voltage.

The new concept of microholographic data storage allows storage capacities of up to 100 GB on a DVD-sized disc 11/. This concept involves bitwise information storage similar to CD and DVD systems. Instead of using pits, the information is coded in form of holographically recorded, microscopic Bragg-reflectors, located in a thin, photosensitive layer (Fig. 1). Each microholographic Bragg-reflector represents one bit, presuming no coding scheme is applied. Microholograms can be stored overlapping in the same volume by using angle multiplexing, wavelength multiplexing or the combination of both. Such storage of multiple information bits in one single position on the disc increases the storage capacity as well as the data transfer rates by the multiplex factor. In contrast to previous holographic storage systems, the storage media are made of cheap and mass-produceable photopolymer layers instead of expensive crystals. Furthermore, the microholographic storage method can be downward compatible with today's Compact-Disk (CD)- and Digital-Versatile-Disk (DVD) systems.

The alignment-free optical module based on a 2-f optical system and the optical microconnectors is proposed. Its availability is confirmed by experiments.

The optical components of a 64 X 64 optoelectronic crossbar interconnect are described. The specifications and performance of the bulk, micro-optic, diffractive and thin- film components are discussed.

An InGaAs/InP coupled quantum well layer sequence specially designed for Mach-Zelmder interferometric space switching is presented. Each coupled quantum well consists of three 27A InGaAs strained quantum wells separated by 1 iA InP barriers. The structure shows a red shift of the absorption edge as high as 8Onm with 1OV applied bias. Using these coupled quantum wells, we realized a Mach-Zehnder interferometric space switch with low attenuation and a switching voltage of 3 .1V for 4mm long phase shifting sections. Furthermore, we realized full polarization independent switching using 0.85% tensile strained and coupled quantum wells. However, we found that the electrorefraction in this structure was not optimal since the red shift of the lowest confined level and the blue shift of higher confined levels yield opposite contributions to the electrorefraction. This compensation can be circumvented in asymmetrical coupled quantum wells resulting in a 1 0 times larger electrorefraction. Coupled quantum wells (CQW) increase the degree of freedom in material design. For instance, the value ofthe bandgap, the total CQW well width and the bias induced change in overlap between electron and hole envelope wavefunctions can be tuned. These tools can be used to optimize a material design for specific applications. In this paper we first present a MZI space switch using a CQW design with optimized Quantum Confined Stark Effect (QCSE) bandgap shift. Secondly we will introduce an improved layer design using asymmetric coupled quantum wells. These wells are optimized for large bias induced changes in the electron-hole envelope wavefunction overlap. The most important design constraints for a material for a Mach-Zehnder interferometric (MZI) space switch at 1 .55tm are waveguide transparency, physical length and polarization independence. When using QCSE tuning sections with InGaAs/InP quantum wells, the QW-thickness is limited to a maximum of 40A for preserving waveguide transparency. On the other hand, such a small QWthickness is far from ideal for an appreciable QCSE resulting in a small index ofrefraction change and thus in long switches. The bandgap shift due to the Quantum Confined Stark Effect increases rapidly with increasing quantum well width [1]. The QCSE in a CQW structure with three 27A InGaAs quantum wells and very thin 1 1A InP barriers is similar to the QCSE in a single 103A quantum well (3x27A InGaAs + 2x1 iA InP). This is due to the excellent coupling of the carrier wavefunctions between the neighboring 27A quantum wells. The inset in figure 1 shows such CQW structure with the envelope wavefunctions of electron and hole ground level. These CQW's combine a room temperature bandgap at 1 390nm, necessary for waveguide transparency at 1 550nm, with a total CQW well width of 103A required for an optimized QCSE red shift

Optical byte recognition using volume holographic correlator is presented. The storage of 256 multiplexed holograms is performed and the phase-coded byte discrimination in real- time is experimented.

A programmable optical bistable device, based on a ring fiber optic circuit in which is inserted an optical amplifier, is presented. The optical amplifier has the purpose both to obtain the bistable duration and to reset the device by means of the loop gain compression effect.

Fan—out phase gratings are micro—optical elements vhicli split an incoming beani into au arra of light beanis with equal power. whereas the diffracted beams are focused by a Fourie-transforin lens. These elements are used. for example. iii optical iiitercoiuiects. in multidetector systems. and in parallel optical processing [ I I. We present different concepts for the fabrication of lwbrid elements which combine the faii-out and focusing function. The combination of the refractive and diffractive function results in a monolithic element with miniaturized dimensions and has therefore a high potential for applications in optical microsystems. To achieve this functionality. we used two different designs and different fabrication technologies. The hybrid elements were fabricated by injection moulding. double-sided photolithographv. and direct laser writing. Single elements as well as large arrays of elements have been fabricated.

Optimum structures for an InGaAs MQW pin diode modulator operating at low voltage (< equals 5V) are explored where the criterion is maximum reflectivity change.

In this paper we present theoretical and experimental results on the nonlinear, optical, electrical, electrooptical and optoelectronic properties of hybrid GaAs/AlAs multilayer-heterostructures. These structures exhibit fast nonlinear properties and high sensitivity which can be used for high-speed information processing in microwave-photonics.

Review of our work on all optical address recognition: past demonstration and current project using state-of-the-art optical bistables and VCSELs.

One of the most intriguing aspects of Vertical-Cavity Surface-Emitting Lasers (VCSEL's) lies in their polarization behavior, which differs from the one of edge-emitting lasers due to the lack of a dominating polarization selection mechanism in the quasi cylindrically symmetric structure. A number of different explanations of this polarization behavior, and especially of the observed polarization switching, have been put forward. A first model is of a thermal nature and attributes the switching to the shift (caused by the current heating) of the material gain spectral profile relative to the cavity resonances for the two frequency-split polarization modes [1J. An alternative thermal explanation focuses on the effects of a thermally induced waveguide occurring in the VCSEL (the so-called thermal lensing effect) and the subsequent competition of the modal gains of the two orthogonal polarization modes coexisting in the slightly anisotropic laser resonator [2].

We present the fabrication of dry etched calcite polarization-selective diffractive optical elements. We demonstrate the feasibility of both submicron and multilevel elements with increased performance.

The technology of deep proton lithography in PMMA (poly methyl methacrylate) is a fabrication method for monolithic integrated refractive micro-optical elements and micro-mechanical holder structures, which allows structural depths in the order of several hundred microns[l.21. Different optical functions can he fabricated in one block and form monolithic integrated optical systems. In addition mechanical support structures and alignment features can he integrated with these optical systems. This paper will focus mainly on the technological requirements of the irradiation, development and diffusion setups. which are necessary to achieve predictable and reproducible results with deep proton lithography.

We simulate and compare optical transmission efficiencies, throughputs and interconnection lengths of free-space and POF-based guided-wave optical interconnection systems for different types of microcavity emitters.

A scheme using CO₂ laser beam to thermally fix selective layers or lumps of photorefractive holograms to integrate and miniaturize an optical 3D system into a single block of LiNbO₃ crystal is suggested. Theoretical and experimental investigations on laser local thermal fixing are given, and a stage of the modified gamma network is experimentally demonstrated.

A hybrid complex matrix multiplication demonstrator using stacked LiNbO₃ electro-optic outer-produce devices is reported, which operates in a fast and partition version of engagement array scheme with mixed negabinary digits. The system is designed for processing 16-digit complex values, and experiments verify the less than 1 bit accuracy.

Main characteristics of the SLM's based on the chalcogenide glass photoconductor--liquid crystal structure under different operating conditions and SLM applications as input and real-time holographic devices in the JTC and as integrating/threshold elements in optical neural networks are discussed.

Many interesting proposals have been made during the last two decades to realize spatial light modulators, switches and logic gates for analog and digital optical processors. However the known devices don't meet usually the complex of requirements (operating power, framing speed, resolution, etc.). It stimulates searching new optical materials, schemes, effects. In the report presented we consider two new effects that have been revealed in theoretical analysis of light diffraction by corrugated dielectric layer. Potential ways of practical application of these effects in optical processing devices are briefly discussed.

We report on the complexity, alignment, mechanical stability, and assessment of an optical neural network, with special emphasis on the liquid crystal light valve.

The figure-of-merit of a cubic nonlinear optical response (CNOR) of a substance, which is defined as is a major parameter for the applicability and competitivity of a set of nonlinear optical devices including phase conjugate mirrors, bistable elements, optical transistors etc.

We present a method with two steps based on diffraction and interference for the characterization in situ of the spatial light modulators in a real-time correlator. The operating curves are used to adapt the filter.

We propose a technique to implement optical digital discrete correlation based on pupil control method with coherent illumination. The technique arranges the location of pixels of an input image of a digital optical computing system. Effectiveness and capability of the technique for the correlation are numerically verified.

The observing optical geometry of a vision sensor viewing objects in liquid is determined by image processing of scale images taken for calculating refraction effect.

Optical signal decomposition on the set of M orthogonal self-fractional Fourier functions, reproducing themselves under the fractional Fourier transform is considered. The image decomposition on the self-Fourier functions (M equals 4) is demonstrated.

A new method for the calculation of the Fresnel diffraction patterns through a fast fractional Fourier transform (FRT) is presented. The FRT can be efficiently calculated for any order by using the fast-Fourier transform algorithm. The resemblances between the FRT and the Fresnel integrals allow the use of that algorithm to calculate efficiently the Fresnel integral of any object, with independence of its shape.

We report on the optical setup, device characterization and performance in a pattern recognition task of a neural network with 256 neurons and optical feedback.

The complete knowledge of geometrical parameters of a periodically modulated material is of great scientific interest, because it allows the user to determine the diffracted image. The modulation profile or the thickness of a phase holographic grating are two of these parameters. They can be determined a posteriori by the measurement of the intensities distribution over different orders, with a comparison of the theoretical and measured angular or spectral selectivities. This method involves a large number of measurements. We propose a new approach, which most of the time needs only one measurement of intensity and polarization (for each of the diffracted orders). The great advantage of the method is the processing of tedious tasks by computers.

We experimentally verified a modeled prediction of optimum polarizer and analyzer configuration for LCTV phase modulation while maintaining high transmission and restricted intensity modulation depth.

Optical fiber sensor output signals are usually electronically detected and transmitted to the control unit through many communications channels. 1'he signals may be also transformed by microprocessor and transmitted through one channel with high transmission rate. Optical fiber sensor network, for example for environmental monitoring [1], may have so many sensors that such method of data transmission is not sufficient. Hence an idea of optical fiber sensor signal processing by use of an optoelectronic neural system and then transmission of such prepared data to the control unit with help an ordinary transmission link seems to be very promising. In general, there are several thousands types of optical fiber sensors with different types of output signals. Most of them. belonging to either intensity or phase sensors, have similar output signals (Fig. 1) The characteristics shown in Fig. I may be more or less linear and symmetrical with different slopes but they are almost the same for such different variables like pH, temperature proximity, strain, angular movement etc. [2]. Such output signals cannot be directly applied for neural processing without additional transformation by use for example two liquid crystal cells (Fig. 2). One of them (LC I in Fig.2) should have threshold transmission characteristics. The second one (LC II) changes transmission level according to requirements of neural weight. In our work we apply only one liquid crystal cell for each sensor with especiall designed electronic control system playing role of smart pixel in neural processing.

In this contribution we present a method for the determination of the twist angle of an arbitrary twisted nematic liquid crystal spatial light modulator.

Optical device based on the thermoplastic recording of information is intended for the continuous registration of data from radar indicator. This device can be treated as an optical processor due to capacity to produce the summarizing on the photothermoplastic storage medium optical image caused the visual subtraction of the real trajectory of a moving object.

We propose a method to implement the Pure Phase Correlation for pattern recognition in a single SLM Joint Transform Correlator working in phase only mode. Experimental results obtained with different kind of images are given.

The fabrication technologies and bonding characteristics of three VCSEL bonding techniques are compared in order to determine the more reliable and robust.

This paper focuses on the technology issues of using monolithic integration of optical interconnects between CMOS circuitry. Although this interconnect solution looks very attractive and might help to solve the so-called interconnect hott]eneck'. a lot of problems still need to he solved. The best way of reviewing all technology issues involved is doing this by means of an experimental demonstrator.

Optical cavities, including semiconductor microresonators, are predicted to support stable 2D cavity solitons. The cavity must contain a nonlinear material, but that material need not itself support solitons. The nature and properties of such cavity solitons, including optical control and manipulation, are discussed and novel array processing schemes taking advantage of their unique features are proposed.

The spatial-frequency analysis of moving objects usually involves freeze-frame imaging followed by post-detection digital processing. We present principles that allow real- time non-imaging methods for measuring the spatial frequency content of an incoherently-illuminated moving object. A theoretical basis is provided by the Van Cittert-Zernike theorem, which states that the mutual intensity of the light from a planar object is governed by the spatial Fourier transform of the optical intensity distribution associated with the object. The objective of systems described in this paper is to convert the mutual intensity of the wave field incident on a plane into useful signal information. Basic concepts are presented and some practical considerations discussed.

The near field intensity pattern of a light pulse diffracted by an ultrasonic pulse under Bragg conditions is an exact copy of that ultrasonic pulse.

A Gbyte/s-class optical-interconnection subsystem for a parallel computer was developed and it operated stably in a testbed system of the parallel processing machine RWC-1 (Real World Computer-1). Although it consisted of MMF parallel optical modules, data was transmitted over 1 km because of the deskew operation of a one-chip transmitter and receiver LSIs. Random packets were transmitted without error over 17.5 hours, which correspond to a BER of 10^-15.

Over the years, the distances at which optical interconnect technology is considered competitive with copper has been steadily decreasing. For instance, initial implementation of fiber optic communication was in long distance telecommunications. Eventually fiber optic links began to appear in local area network backbones over distance scales from 100 meters to 2km. Most recently, there has been an acceptance that fiber optic interconnects are competitive inside servers or switches, at distances on the order of 10 meters. Therefore one might ask at what level optical interconnects will be applied in the future. Will they be cost effective inside a box (1 meter?). Could they even provide an interconnect function between chips on a board or MCM (1-10 cm)? As interconnect distances become progressively shorter, the cost limitations become more and more severe.

InGaAs/CMOS smart pixel arrays have been developed for free- space optically connected processing systems. A parallel optoelectronic crossbar switch demonstrator targets 1 Tbit/s aggregate input to a single chip.

We have built an optical interconnection subsystem to interconnect the nodes of a massively parallel computer, and have achieved highly reliable 48-bit synchronized parallel- optical-data transmission (600-MB/s throughput).

We present the design and analysis of a 3D OptoElectronic Stacked Processor System. This system is aimed at combining stacked VLSI chips with free-space optoelectronic Inputs and Outputs for optimal power*volume/throughput balance. Challenges in such systems include finding appropriate thermal extraction schemes, packaging and alignment of the multiple stacks, and appropriate micro-optical and optoelectronic component design.

Technologies for area optical interconnect between CMOS IC's are reviewed. Emphasis is on hybrid and monolithic integration methods of optoelectronic devices with CMOS IC's as well as on suitable optical pathways.

It is the ideal of a computer designer to have a huge, yet very fast memory connected to a uniprocessor core. But in reality, these two requirements, fast and huge, are not reconcilable. For this reason, a memory hierarchy was introduced that consists of very fast and small memory close to the processor core (the registers) but slower and larger memory further away from the processor (Figure 1). When data is needed, it is fetched from the slower memory into faster memory, from which it can be quickly accessed.

Free-space optical interconnections within a single chip can slow of halt the growth in the number of metal levels required as gate count increases. Quantifying these benefits leads to interesting conclusions and questions.

Photonic smart pixel arrays are being developed for use in high density input/output applications. Arrays containing 1024 smart pixels are integrated on a single chip which occupies 1.6 cm X 1.6 cm of area. A smart pixel cell is made up of a vertical cavity surface emitting laser (VCSEL), a detector, an analog receiver, a laser driver, and digital logic circuitry. Additional circuitry can be embedded in the smart pixel cell to provide increased functionality, such as digital logic operations and memory functions. A diffractive optical interconnect element can be added to the smart pixel array for optimization of a free space optical interconnect system. The oxide-confined GaAs/AlGaAs VCSELs and the GaAs/AlGaAs detectors will be flip-chip bonded to the surface of the CMOS substrate. The surface of the CMOS substrate will be processed after circuit fabrication in preparation for flip-chip bonding. This processing will make the pads of the CMOS compatible with the contacts of the VCSEL and detector surfaces. The integrated smart pixel combines low power consumption and high speed operation in an optoelectronic module for the use in high speed optical interconnects.

Recontigurable computing architectures are gaining popularity as replacements for general purpose architectures in many high performance applications. Reconfigurable systems can take advantage of deep computational pipelines, perform concurrent execution and are inherently data flow in nature. Many applications can exploit these systems, such as genomic sequence scanning. Fast Fourier Transform, text searching. and computer vision. Current research efforts are applying reconfigurable computing to perform automatic target recognition, real-time image processing, and hardware implementation of neural networks. However, these architectures suffer from a trade off between slow reconfiguration times versus low logic gate densit'v when used to support large computations. This problem is due to the fact that configuration memory typically resides off-chip and reconfiguration is performed serially. Recent approaches4 solve this problem by adding an on-chip configuration cache that provides faster reconfiguration at the cost of die area. That is, the area overhead of the configuration cache gives a low total logic gate density for the architecture. These disadvantages limit the performance, and therefore the applicability of current reconfigurable systems. In this paper. a reconfigurable processor architecture is proposed that overcomes the limitations discussed above by using high bandwidth optical channels. The optical channels allow fast parallel loading of the reconfiguration control word as well as the migration of the configuration cache off-chip. The migration of configuration cache allows better utilization of the die area for reconfigurable processing elements. Further, it is possible to implement the optical detectors directly in silicon, hich does not require significant alteration of the fabrication processes. These advantages make the optically reconfigurable architecture competitive for high performance applications.

Highly interconnected multiprocessor systems are now performance limited by the backplane interconnection bottleneck associated with planar interconnection technologies. Smart pixel throughput capabilities are projected to exceed I Thitls/cm2 [1] and offer the promise of overcoming the bottlenecks of planar technologies for many types of interconnection-limited multiprocessor problems. Systems that use smart pixel-based free space optical interconnects (FSOI) provide two general dense interconnection capabilities: intelligent parallel data transfer and intelligent parallel data interchange. Optical imaging provides a high throughput approach to linking smart pixel planes for data transfer. In this case the high 110 density of smart pixels may provide a power consumption and size advantage over electronics [2,3]. For data interchange, FSOI provides the additional ability to perform the data partitioning and interleaving useful in space variant link interconnection patterns like the perfect shuffle (PS) [41,which are inherently difficult to implement in planar interconnection technologies. Such patterns are characterized by high BSBW [51. In multi-processor architecture design, there is a direct trade-off between minimum BSBW and latency in a network. It is therefore generally desirable to implement networks with the largest minimum BSBW that can be practically achieved to solve a given problem. The ability of optical elements to interconnect large arrays in space-variant patterns, without crosstalk in the medium, suggests that FSOI techniques are particularly promising for problems with high BSBW. For problems with greater than 1 ThitJsec BSBW (i.e., greater than the capabilities of a single chip) free space optical interconnects have a marked advantage [6,71. Therefore, globally interconnected multi-chip smart pixel based architectures have the potential to reap the full benefits of FSOI. This paper describes the experimental demonstration of a smart pixel based optical interconnection prototype currently being developed under the Free-space Accelerator for Switching Terabit tworks (FAST-Net) project, sponsored by the U.S. Defense Advanced Research Projects Agency. The prototype system incorporates 2-D arrays of monolithically integrated high-bandwidth vertical cavity surface emitting lasers (VCSELs) and photodetectors (PDs). A key aspect of the FAST-Net concept is that all smart pixels are distributed across a single multi-chip plane. This plane is connected to itself via an optical system that consists of an array of matched lenses (one for each smart pixel chip position) and a mirror. The optical interconnect system implements a global point-to-point shuffle pattern. The interleaved 2-D arrays of VCSELs and PDs in the prototype are arranged on a clustered self-similar grid pattern with a closest element pitch of 100 tm. The circular VCSEL elements have a diameter of 10 pm and the square PDs have an active region that is 50 jim wide. These arrays are packaged and mounted on printed circuit boards along with CMOS driver, receiver, and FPGA controller chips. Micro-positioning mounts are used to effect alignment that is consistent with current MCM chip placement accuracy. Shuffled optical data links between the multiple ICs have been demonstrated in preliminary evaluation of this system. These results suggest that a multi-Terabit optically interconnected MCM module is feasible.

The generation and application of information is rapidly evolving from text and graphics based to multimedia based, and it will shortly continue to evolve to virtual reality. The evolution between these stages introduces dramatic increases in the amount of data associated with the applications. For example, where text-based meeting notes have given way to emailed copies of vugraphs, future meeting documentation may require storing and communicating an entire collaborative virtual reality session. Even in the near term, the need to store, search for, and edit large numbers of images and digital video clips will drive data storage requirements forward in home, office, and network arenas, as shown in Fre 1 .

With present electronic equipment it is difficult to perform the temporal analysis of optical signals in the picosecond range. One usually resorts to sampling techniques that do not apply to single-event signals. Pattern recognition is an alternative way to perform signal analysis. This technique is well known in the context of the spatial analysis of two-dimension scenes. It relies on the comparison of an input object with a series of filter patterns, that have been stored in a bank. It is restricted to the recognition of one pattern among those that are contained in the bank but it applies to single events and it can be very fast since it relies on the cross correlation processing capabilities of non-linear optical materials. We adapt this procedure to the temporal domain.

Photorefractive materials are of high interest for read-write holographic data storage. One of the main problems that makes the practical implementation of these memories difficult is the erasure of stored information during read-out. Several solutions like thermal fixing, electrical fixing, and two-step recording have been proposed for this problem'3, but they need some special requirements like heating the crystal, using high electric fields, or high light intensities.

SLMs' finite contrast causes noise which bounds the number of storable holograms. We present a method to reduce this noise and increase the storage capacity.

Optical signal transmission benefits, among other features, from its huge potential of multiplexing. Besides WDM or time multiplexing a new kind ofmultiplexing is possible: directional multiplexing. Measurements are presented showing that in short fiber transmission lines the angle 9 between the symmetry axis of a fiber and the principle propagation direction of a beam is conserved over distances in the order of 1 meter. So, provided a beam is coupled into a fiber under a certain off-axis angle 3 it will couple out of the fiber with the same off-axis angle. This conservation behavior can be used for coding different channels for multiplexed transmission and can provide a high bandwidth interconnection between chips or boards, i.e. within computers in a single fiber. Based on the experimental set-up used in this work an estimation on the degree ofmultiplexing will be given. Additionally, suitable components for multiplexing and de-multiplexing have to be designed. That is, on the one hand different signal channels have to be coded into different angles before being coupled into the fiber. On the other hand the directional spectrum of the outcoming beams has to mapped to different positions on a detector plane. Particular attention is paid to the de-multiplexing component: A specially designed diffractive optical element (DOE) serves as a mapping device between directional spectrum and spatial positions on the detector plane.

Inthe field of optical interconnection research, the topic perhaps closest to widespread applications is the realization of optical board to board and chip to chip interconnections [1]. The inherent limitations of electronic interconnects regarding bandwidth and electromagnetic interference are overcome by waveguide, fiber, or free space optics. With free space optics additionally the high interconnect density ofoptical imaging can be employed. One ofthe basic problems is the addressing of emitted signals to the intended detector. The concept of folding a free-space optical system into a thick transparent planeparallel substrate, was proposed by Jahns and Huang in 1989 [2]. Planar optics, also known as substrate-mode optics, fully employs the three dimensional nature of light propagation and also the fabrication methods known from integrated circuit manufacturing can be adopted. In addition to that the advantage ofthese technologies are compact packaging and simple alignment ofthe optical and electronical elements. The problem involved in folding optical light paths for imaging of extended data fields is the demand for good off-axis imaging properties ofthe optical elements. Besides diffractive solutions i.e. DOEs or CGHs [3] refractive optics offers the possibility for wavelength multiplexing since the wavelength dependence is only due to material dispersion. For off axis imaging in a folded zig-zag-path the optical system also has to be corrected by special astigmatic components.