This paper reviews the progress of free-space digital optical systems demonstrated by AT&T.

Optical interconnects at the cabinet-to-cabinet, board-to-board, and multichip module-to- multichip module levels will enable future avionics systems requirements to be met by eliminating undesirable compromises associated with electrical interconnects. Fiber optics is the well established medium of choice for cabinet-to-cabinet applications, while planar polymeric interconnects are required at the backplane level. Significant advances have been made in demonstrating practical polymer interconnects compatible with existing board fabrication principles, however both waveguide loss and interfaces to optoelectronic components require further improvement before the technology is broadly applicable.

A holographic optical interconnect technique capable of massive parallel interconnect operations is introduced. Its applications in digital data processing and pattern recognition were also experimentally realized.

A second generation digital optical computer (DOC II) has been developed which utilizes a RISC based operating system as its host. This 32 bit, high performance (12.8 GByte/sec), computing platform demonstrates a number of basic principals that are inherent to parallel free space optical interconnects such as speed (up to 10¹² bit operations per second) and low power 1.2 fJ per bit). Although DOC II is a general purpose machine, special purpose applications have been developed and are currently being evaluated on the optical platform.

An approach to optical interconnect networks at the module level is presented that addresses the requirements imposed by electronic system manufacturing, such as thermal stability, low cost, and compatibility with standard electronic design, fabrication, and assembly processes. Research is presented on poled polyimide electro-optic materials with extended thermal stability, poled polyimide integrated optic switches acting as transmitters, and a demonstration of a CMOS-compatible optical interconnect.

Routine use of optical interconnections in MCM based computing systems ideally favors monolithic integration to achieve both high density and manufacturability. The central issue facing this monolithic evolutionary path is the compatibility of both III-V semiconductor growth and subsequent optoelectronic device and passive optical interconnection processing with existing and future generations of CMOS and advanced packaging technology. The influence of GaAs heteroepitaxy and device processing on submicron CMOS is the subject of an ongoing program seeking to experimentally determine compatibility conflicts and through understanding of their physical mechanisms identify directions for achieving GaAs heteroepitaxy compatibility with future CMOS generations. Following a brief review of GaAs heteroepitaxy compatibility concerns, preliminary results from the current experimental program exploring the influence of both thermally simulated and actual GaAs heteroepitaxy on commercial 0.9 micrometers (0.6 micrometers minimum channel length) CMOS are presented including parametric device modeling, interface state, and hot electron measurements of experimental test lot devices.

The development of an optical interconnected multichip module (MCM) is underway at UNC- Charlotte. The approach is to use optical interconnects within a digital multichip module for connections that are longer than a certain length (break even line length). For these connections the optical link dissipates less power than the corresponding electrical link. One of the main goals of this project is to develop a technology for optical interconnects that can be implemented with minimal modification to current devices. We are currently in the process of developing a series of MCM systems. System 1 will be a testbed that enables testing of hologram encoding techniques, alignment tolerances, optical link efficiency, and thermal properties. System 2 will use the knowledge gained in the development of system 1 to build a small functional system to demonstrate the technology.

In a previous paper, we proposed a new approach to the optical interconnection of telephone- card signals. Our approach was developed around a multimode-fiber star coupler as the optical bus, and standard optoelectronic components for transmitters and receivers. In this paper, we report on recent developments of two practical demonstrators of the above concept, namely: (1) an optical backplane for 24 X 24 I/O signals interconnection operating at 1.1 Gbit/s, with a transmitter and receiver fabricated in hybrid technology and fully connectorized to the star coupler; (2) an optical clock distribution capable of operating up to 4.5 Gbit/s with < 5% amplitude droop. Common to both units are a good S/N ratio (> 25 dB), very little time unbalance between channels (< 15 ps), and an inherent bandwidth-length product of 40 Gbit/s(DOT)m as determined by the coupler and fiber pigtails.

In this paper, we report an optical bus fabrication technology using compression-molding technique. The linear dimension of such a waveguide is well beyond that of a microlithographically defined waveguide. The interconnection patterns such as fan-ins and fan- outs can be easily defined by the mold plunger. The resolution of compression molding can be as high as 2 micrometers . Therefore, optical bus density as high as approximately 10⁴ channels/cm is producible while the linear dimension of the waveguide can be much larger than that made through conventional microlithography. Employment of optically transparent electronic packaging polymers (OTEPPs) as the system buses automatically provides process compatibility with silicon IC fabrication. All the polymer microstructure waveguide materials are either thermosets or thermoplastics. Both can be molded to a specific shape as desired.

New packaging techniques for opto-electronic multichip modules (OE-MCMs), including OE substrates and optical coupling between a waveguide and a flip-chip bonded photodevice or fiber, are presented for high-speed and wide-band communication systems. The OE substrates, which offer high-density, high-speed optical and electrical interconnection, are made from low loss (0.4 dB/cm) optical polyimide waveguides fabricated on copper-polyimide electrical multilayer substrates. A total internal reflection mirror fabricated at the edge of the optical waveguide reflects the light propagating from the waveguide to a flip-chip bonded photodiode with a loss of less than 1.5 dB. The waveguides are coupled to fibers for inter-module interconnection using the self-aligning fiber guiding method.

We propose a monolithic implementation of a feed-forward neural network in AlGaAs/GaAs waveguide material with a Wannier-Stark superlattice in the core. Adjustable weights are applied with the use of voltage-controlled below-bandgap intensity modulators. Nonlinear thresholding is performed with the use of a saturable absorber component. A 2-to-1 single neuron device was fabricated by integrating two 25 dB/mm modulators and a 25 dB nonlinear switch. The device performed with an output/input range ratio of 25 dB using two 780 nm laser diodes.

Various components needed for optical backplane applications have been demonstrated from gelatin-based polymer integrated optic material. An array of waveguides have been realized with packaging density as high as 1250 channels/cm and loss of 0.1 dB/cm. A 1-to-8 Y- junction splitter and a 32 X 32 star coupler have been fabricated using photolithographic techniques. The unification of the star coupler and a modulator array for backplane optical interconnects are under investigation.

The synthesis and optical characterization of fluorinated polyimide systems with potential use in passive waveguides and electro-optic devices is reported. The effect of fluorination on optical properties such as refractive index, birefringence, and near-infrared absorbance is reviewed in terms of optical performance requirements. Synthetic methods of tuning the refractive index in order to achieve appropriate core/cladding differentials is discussed. The relation between processing parameters and refractive index for several polyimide structures also is reported. We describe the microlithographic fabrication of a multilayer polyimide rib- type waveguide that is suitable for single mode guiding. The waveguide is fabricated using photosensitive polyimide systems via negative resist imaging. A comparison of wall profiles and resolution limits afforded by the wet-chemical patterning techniques is presented. Results on channel guide coupling, propagation, and loss are described, as well as progress in producing active guides.

The packaging opto-electronic interconnects has the potential to create compact, highly dense communication networks. We present a packaged space-variant optical interconnect module using wavelength multiplexed volume holographic elements recorded in photorefractive materials. The input/output arrays (4 X 4), an illumination lenslet array, and a set of two wavelength multiplexed off-axis volume holographic lenslet arrays were integrated with a series of several glass substrates to form a free-space optical interconnect. The size of the packaged optical interconnect module was 40 mm X 24 mm X 37 mm. Reconfigurable interconnection was demonstrated on the packaged module by performing the perfect shuffle and butterfly networks, at different wavelengths. Several packaging issues, such as alignment, bonding, energy efficiency, and system scalability were studied.

Multistage interconnection networks based on the perfect shuffle topology are often suggested as candidates for large scale multiprocessor and broadband communication networks. The perfect shuffle interconnection requires global communication links that extend across the entire system and have a large number of wire crossovers. These constraints prohibit a scalable electronic implementation both within a VLSI chip and at the MCM or board levels. This paper presents the architecture of a scalable optoelectronic hardware module for building multistage interconnection networks. To achieve a scalable implementation, the design uses free-space optical interconnects for global communication links and electronic VLSI technology for local communication links and switching elements (e.g., smart pixel approach). Our approach is to engineer a network with the desired functionality, cost, and performance characteristics using generic hardware modules. In this paper, various applications are examined and their implementation using the proposed method is described.

High fan-in/fan-out, low power (1 fJ per gate), high performance computing (HPC) modules are being developed that integrate global (multidimensional) free space `smart' optical interconnects with GaAs DANE technology. This new architecture implements N<sup>4</sup> global free space optical interconnects coupled with 2-D arrays of N-bit Boolean multiplications by DeMorgan's theorem on wide word fan-ins. `Smart' interconnects provide a high speed inter- module alternative without the power requirements and cross-talk limitations of GaAs circuitry. Selected algorithms such as 64-bit addition can operate at lower power and higher speeds using global technology and GaAs logic. Thus, faster processing can be fully realized which allows for a reduction in pipeline delays.

For the packaging of integrated InP devices we present two approaches, which rely on a passive fiber alignment. In one case the fibers are hold in V-grooves etched directly into the InP, alternatively, V-grooves are etched into Si and the InP chip is assembled by a self- aligning flip chip process.

A multistage optical nonguided switching network has been proposed with orthogonally polarized data and address information. This network is unique in that the data information remains in optical form throughout the network (i.e., it is never converted into electrical information). Since the data are never regenerated throughout the network, the system has two main characteristics: (1) the bandwidth of the data is not restricted by electrical circuit considerations; and (2) the optical interconnections from one stage of the network to the next must be highly efficient. The second effect is of interest here. An interconnection that is 90% efficient between stages results in approximately a 57% attenuation of the data through a nine stage network (eight routing interconnections). After examining several methods for achieving this efficiency, a novel system layout is introduced.

In this paper we focus on the design of optically interconnected MCMs for gigabit ATM switching networks. Our approach is to design a generic hardware module that can be used to implement ATM switches with application-specific functionality, cost, and performance requirements. The module design is partitioned on the MCM such that it can be built using VLSI chips interconnected with holographic free-space optical interconnects. Holographic optical interconnects are also used for inter-MCM communication. A comparison of our approach with electrical MCM, all-optical, and guided-wave implementations of ATM switches is presented. A detailed review of the technology used in our design and various switch architectures that can be achieved can be found in references 1 and 2, respectively.

Optical fiber busses supporting multiple concurrently usable channels through the use of wavelength division multiplexing can meet the demanding interconnection requirements of high performance multiprocessors. Using a simple analytical model, we establish why global channel access protocols for the WDM channels are inadequate for a WDM optical fiber bus. We propose a simple WDM bus that supports a small number of channels and uses readily available receivers and transmitters with fixed tunings. A simple channel allocation and arbitration strategy allows up to 100 processors to be connected using a single WDM bus with only 5 WDM channels. Simulated performance shows that the bus does not become a bottleneck even when 100 processors are connected. The WDM channels also allow cache coherence and processor synchronization to be easily supported, making the bus feature- compatible with the IEEE Futurebus+ and SCI standards.

This paper explores the use of optical spanning bus interconnections using optical fiber links and wavelength division multiplexing (WDM) with statically assigned channels for implementing distributed shared memory multiprocessors (DSMM). The WDM optical links allow processor synchronization and coherent caches -- two very important requirements for a DSMM -- to be implemented efficiently. Simultaneous broadcasts possible on several channels along a WDM photonic link allow barrier synchronizations to be fast and scalable. The large bandwidth capability of WDM optical links and their ability for accommodating simultaneously -- active channels permits a write-update based cache coherence protocol to be implemented. Our proposed cache coherence protocol is sensitive to the relatively longer propagation delays along optic fiber links and possible mismatch in speeds between the electronic and photonic components. The protocol is hierarchical, reflecting the hierarchy in the interconnection, making it easily scalable with system size. In addition, the proposed protocol performs opportunistic request combining, a pleasant side-effect of using optical links.

To eliminate the intrinsic problem associated with electrical interconnects, processor-to- processor interconnects based on fiber optics are under development. The major bottleneck of fiber-based optical interconnects is their point-to-point characteristic which seriously limits their interconnectivity. As a result of the low interconnectivity, multiplexing/demultiplexing (e.g., 1:1024 and 1024:1) technology was employed to minimize optical channels. This approach significantly reduces the bandwidth of the data transfer rate. Moreover, it is incompatible with most of the IEEE standard buses (such as FASTBUS, VMEbus, and Futurebus). The high density highly distributed channel waveguide array presented in this paper is the only feasible solution to provide a lithographically defined optical interconnection network with full compatibility of IEEE standard and special purpose high performance bus systems.

A unique architecture has been developed for a compact programmable optoelectronic logic array. The system is comprised of two arrays of symmetric self electro-optic effect devices (S- SEED) and a 3 dimensional integrated guided-wave crossbar interconnection network. Presently emerging technologies would allow densities of a few hundred thousand S-SEEDs, connected in a wide sense nonblocking manner, per cubic inch. Such arbitrary connectivity, in effect a `virtual 3-D hologram,' would allow execution of complex logic functions.

Seven common interconnection architectures are evaluated for use as guided-wave interconnection networks. Comparisons are based on characteristics such as the number of switching elements, number of stages or path length, number of layers, number of waveguide crossings, connectivity and routing algorithms. Investigation of the active splitter/active combiner, passive splitter/active combiner, crossbar, n-stage, buddy type multistage interconnection network (MIN), duobanyan and Benes architectures reveals a trade space which is both complicated and previously not well explored. Even among the topologically equivalent MINs such as the baseline, reverse baseline, regular SW banyan with S equals F equals 2, indirect binary n-cube, modified data manipulator, omega or flip networks, the number of waveguide crossings varies while other properties remain constant. Analysis of this larger set of factors reveals more complex trade offs, and presents these architectures in a new perspective.

A new optical RAM-bus (ORAM-bus) memory has been proposed. This memory has three- dimensional memory structure with the vertical optical interconnection. Data are vertically transferred in this memory by the optical coupling sense amplifier which acts as a cache memory and also as an amplifier to amplify the signal read-out from a memory cell. It was confirmed in the computer simulation that a very high data transfer speed of 64 Gbit/s can be achieved in the ORAM-bus memory with the configuration of 256 Kbits X 4 layers. The new parallel computer system with such ORAM-bus memories is also proposed.

We report a new coding technique for the implementation of an electro-optic analog-to-digital (A/D) converter. This coding and A/D conversion can be implemented with an array of tunable, high finesse, channel waveguide Fabry-Perot etalons. Preconfigured dc biases have been shown, on a proton exchanged device on x-cut LiNbO₃, to shift the quantized etalon transmission peaks with respect to the applied rf signal voltage necessary for converting analog voltages to digital optical transmissions. Some design considerations for the implementation of the device are discussed.

Surface operating metal-semiconductor-metal (MSM) modulators and photodetectors with monolithically integrated Fabry-Perot resonators are presented. Because of the planar design and the process compatibility with MESFETs, these devices are well suited for integration with electronic circuits and arrayed configurations in optical interconnects. At a switching voltage of 10 V the modulator's switching contrast exceeds 60:1 at an insertion loss of 3 dB. Up to 20 GHz a flat frequency response for photodetection and modulation is derived.

The monolithic and hybrid integration of vertical-cavity surface-emitting lasers to phototransistors, heterojunction bipolar transistors, and field-effect transistors is presented. The integrated devices or `microlaser smart pixels' exhibit a high level of performance. For example, Boolean logic functions, high gain and speed are demonstrated. These `microlaser smart pixels' integrated with micro-optics have numerous applications.

A very low cw threshold current of 2.5 mA ( 25 degree(s)C) and 8.0 mA ( 80 degree(s)C) with high reliability has been realized in the all-MOCVD grown BH lasers on p-InP substrates. A strained MQW active layer of 1.3 micrometers wavelength and the precise carrier confinement buried structure by MOCVD is employed for the BH lasers. The excellent potential of long lifetime of the all-MOCVD grown laser has also been confirmed. After the high temperature and the high current (100 degree(s)C, 200 mA) aging test, no significant degradation is observed which is comparable with the well-established LPE grown lasers. The BH laser is also operating stably over 3700 hrs under the APC condition of 50 degree(s)C, 10 mW. Finally, an extremely uniform 10-element all-MOCVD grown LD array is demonstrated, which has the threshold current uniformity of 2.4 +/- 0.1 mA ( 25 degree(s)C) and 9.2 +/- 0.2 mA ( 80 degree(s)C). The growth mechanism in the MOCVD is also described.

Epitaxial liftoff has emerged as a viable technique to integrate GaAs with silicon. The technique relies on the separation of a thin epi-GaAs film from its substrate followed by direct bonding of the thin film to a silicon substrate. The silicon substrate has to meet certain planarity and smoothness conditions in order to obtain high quality bonding. Unfortunately, processed silicon IC chips do not satisfy these conditions. In this paper, we report on the results of two different planarization techniques, plasma etch back and chemical mechanical polishing, to integrate GaAs LEDs with silicon circuits using epitaxial liftoff. A 4 by 8 array of GaAs LEDs have been integrated with silicon driver circuits using plasma etch back. We also have lifted off areas as large as 500 mm² and bonded them on five inch device wafers by chemical mechanical polishing.

In this paper we describe the design and testing of a high speed photodetector that is fully integrated into a CMOS chip, with a response time of 2.6 ns and an operating speed of 112 MHz. The photodetector is to be used in an optically interconnected multi-chip module.

The properties of vertical-cavity surface-emitting lasers (VCSELs) and VCSEL-based optical switches using MOCVD-grown epitaxial material are discussed and summarized. Also discussed are some of the factors that limit their performance.

We describe a dense and flexible all optical multi-channel communication system for high speed computer interconnects. The system can provide 10 Gb/s for each individual node with a total system capacity to 250 Gb/s using currently available technologies. The system capacity can be scaled to 1 Tb/s using optical amplifiers with a broader bandwidth and higher modulations. The system is based on the multi-beam (heterodyne) modulator (MBM) recently demonstrated in our laboratory and other current technologies in tunable laser arrays and acousto-optical tunable filter (AOTF). Each MBM automatically forms a high frequency microwave sub-carrier multiplexing (SCM) with sub-carrier frequency to tens of GHz. A MBM with sub-carriers at 17 and 21 GHz has already been demonstrated and can be scaled to higher frequencies by using a higher frequency detector. Each SCM group may consist of up to 10 one-Gb/s channels and occupies only 1 nm spectral width. Therefore we can form a conventional WDM with 25 divisions within the bandwidth of commercially available optical amplifiers.

A simple etalon-type structure was considered as the basis for a light modulator. The distinction is made between thermally induced refractive index changes and the carrier injection mechanism. Experimental results are presented to demonstrate IR light modulation in a silicon etalon structure. The modulation is controlled by both current injection and visible light changing silicon's index. This modulator was used for coherent discrimination of low level cw laser illumination on a high irradiance incoherent background.

In this paper, we proposed an all-optical crossbar switch based on a successfully demonstrated optically activated modulator (OAM). This modulator was based on a GaAlAs/GaAs channel waveguide and waveguide array. A schematic of the demonstration is shown in Figure 1. A 5 micrometers activation window was employed to input an approximately mW HeNe 632.8 nm light which, in turn, modulates the 1.3 micrometers guided light. Modulation depths from 33% to 85% have been observed on the various devices tested. The size of the activation window is much smaller than for a linear electro-optic device (approximately mm to approximately cm) and a high power laser was not used in this demonstration. Consequently, this constitutes proof that a much more practical crossbar based on GaAs/GaAlAs channel waveguides can be built. The device architecture for the all-optical low-threshold 10 X 10 crossbar switch is shown in Fig. 2 where semiconductor laser diode arrays with wavelength shorter than the band gap of GaAs semiconductor are employed to optically switch the propagation directions and thus receiving ports. Refractive index modulation as high as 10^-2 was achieved. Therefore, up to 100% on/off switching can be achieved within an activation length of 20 micrometers .

Optical board-to-board interconnects for 1 - 100 cm distances are presented. A thick light- guiding plate (typically 5 - 10 mm) is used for confinement and transport of the optical signals with low loss (-0.1 dB/cm), highly efficient holographic optical elements in dichromated gelatin are used for coupling (-0.5 dB) and binary phase gratings for broadcasting of signals. Several experiments were performed: (1) point-to-point interconnects operating at a data rate of 650 Mbit/s per channel, (2) a parallel interconnect with high packing density up to 1000 channels/cm², (3) an optical permutation stage for data permutation networks, and (4) an optical star coupler for signal broadcasting within light- guiding plates. Holographic optical elements in dichromated gelatin have shown excellent diffraction efficiency and low level of straylight at wavelengths ranging from 488 nm to 1.5 micrometers .

Abstract not available.

Results of e-beam-written electro-optic (EO) organic polymeric integrated optical (IO) channel waveguide devices developed for onboard satellite applications are reported. Processing for both strip-loaded and ridge guide structures was developed to fabricate several types of devices, such as linear and curved waveguides, optical power splitters, combiners, and phase modulators, on a 3-in.-diameter silicon wafer. A large number of polished, butt-coupled devices were tested for optical loss and EO modulation at 830- and 1,310-nm wavelengths. These devices are highly reproducible, capable of very high frequency modulation (currently at 18 GHz) with a small rf drive (0 dBm), and can be made reasonably long if required. For the optical phase modulation measurements, both subcarrier modulation with phase-sensitive detection by a lock-in amplifier, and direct detection of the rf-modulated optical carrier in a lightwave analyzer, were used. The short- and long-term stability of the corona-poled EO polymer films investigated by second harmonic generation were found to be good. IO channel waveguide development considerations leading to above performance characteristics also are presented.