Optical: systems and elements – Holographic system or element – Using modulated or plural reference beams
Reexamination Certificate
2000-08-27
2002-09-03
Henry, Jon (Department: 2872)
Optical: systems and elements
Holographic system or element
Using modulated or plural reference beams
C359S022000, C359S035000
Reexamination Certificate
active
06445470
ABSTRACT:
TECHNICAL FIELD
The present invention relates to simultaneous spatial modulation of angularly multiplexed optical beams that are individually coherent but mutually incoherent, to multiplexed volume holographic recording, readout, and interconnections, and, more particularly, to photonic interconnections, photonic implementations of neural networks, optical signal processing, optical information processing and computing, optical memory, optical displays, copying of multiplexed volume holograms, and multiplexed volume holographic optical elements.
BACKGROUND ART
A wide variety of information systems applications exist that require a high density of interconnections among device or system nodes, or a high density of rapidly accessible memory, or both. These applications include, for example, neural networks, telecommunications switching systems, digital computing, optical displays, and information (including signal) processing. In many such applications, key requirements on the chosen interconnection technology include low insertion losses, high interchannel isolation (freedom from interchannel crosstalk), a high degree of potential fan-in and fan-out at each node, weighted interconnection channels, and high capacity. Comparable requirements exist for the chosen memory technology, including low latency (rapid information access), parallel information retrieval, low effective bit error rates (high signal-to-noise ratio), high density information storage, and input/output compatibility with the remainder of the system.
In order to satisfy these many and varied requirements, multiplexed volume holographic optical elements provide an attractive alternative to electronic implementations of high capacity interconnection and memory elements. In fact, the very nature of a volume holographic optical element tends to blur the distinction between a pure interconnection network on the one hand, and a pure memory subsystem on the other, as it is in many ways simultaneously well-suited to both roles. Even so, previous methods for recording information or interconnection patterns in highly multiplexed volume holographic optical elements, and for reading them out, have not proven satisfactory in terms of throughput, crosstalk, and capacity. Furthermore, they have not proven to be manufacturable, due to the fact that information from a master volume holographic optical element could not previously be efficiently transferred to or duplicated in another such element.
In forming multiplexed volume holograms, one of three approaches is typically taken: (1) sequential, which involves several temporally-sequenced (and hence incoherent) exposures of the individual components of the hologram, done by rotating or translating the hologram (or the source beam, reference beam, or object beam); (2) simultaneous and fully coherent, which involves the use of two or more mutually coherent beams, each encoded with information and serving as a reference beam for the other(s); and (3) some combination of sequential and simultaneous fully coherent.
The first approach has the major disadvantage that temporal sequencing is time-consumptive, which can be of considerable importance in applications envisioned herein, for which the number of independent interconnections that must be recorded is extremely large. Also, in many holographic recording materials, sequential exposures tend to erase previously recorded information, leading to the necessity of incorporating unwieldy programmed recording sequences in order to result in the storage of a predetermined set of interconnections.
The second approach is designed to circumvent the above sequencing difficulties, but suffers instead from the coherent recording of unwanted interference patterns (holograms) that give rise to deleterious crosstalk among the various (supposedly independent) reconstructions, as described in more detail below.
The third approach is subject both to sequential recording time delays and the necessity for programmed recording schedules, as well as to the generation of undesirable crosstalk. As such, none of the previously employed multiplexed recording techniques allows for the generation of three-dimensional, truly independent interconnections between two or more two-dimensional planar arrays within the context of a temporally efficient recording scheme.
In all of the prior art approaches to the holographic recording of a multiplexed interconnection, two primary forms of interchannel crosstalk are encountered to a greater or lesser extent. Coherent recording crosstalk arises from the simultaneous use of multiple object and reference beams, all mutually coherent with each other. The mutual coherence causes additional interconnections to be formed other than those desired. Reconstruction with independently valued inputs results in the generation of output beams that cross-couple through the undesired interconnection pathways, which compromises the independence of the desired interconnection channels.
A second, unrelated form of crosstalk arises due to beam degeneracy, which occurs whenever a single object beam is used with a set of reference beams to record a fan-in interconnection to a single output node (e.g., neuron unit in the case of the photonic implementation of neural networks). (Fan-in is the connection of multiple interconnection lines to a common output node.) This latter form of crosstalk is present even when the set of object beams is recorded sequentially.
Of at least equally serious consequence is the optical throughput loss that results from interconnection fan-in so constructed as to exhibit beam degeneracy. In many well-documented cases, this loss is severe, resulting in at least an (N−1)/N loss (or, equivalently, a 1/N throughput efficiency) for the case of an N-input, N-output interconnection system, as reported by J. W. Goodman,
Optica Acta
, Vol. 32, pages 1489-1496 (1985). This is a truly daunting loss factor for interconnection systems such as those envisioned for neural networks, which may both require and be capable of 10
5
to 10
6
inputs and outputs.
In certain types of photorefractive materials, an additional throughput loss can arise from the incoherent superposition of several gratings within the same volume of the holographic optical element, due to the reduction in the effective modulation depth of the recorded holographic fringes. This effect occurs primarily in photorefractive crystals that generate an index of refraction or absorption change in response to local gradients in the intensity distribution, but would not be expected to occur in linear photorefractive materials that generate an index of refraction or absorption change in direct proportion to the magnitude of the local intensity distribution. In a number of cases, this effect can also result in at least an (N−1)/N loss for the case of an N-input, N-output interconnection system, as reported by P. Asthana, “Volume Holographic Techniques for Highly Multiplexed Interconnection Applications”, Ph.D. Dissertation, University of Southern California (1991), available from University Microfilms, Ann Arbor, Mich.
In the prior art, few attempts have been made to address the extremely important technological problem of duplicating the contents of a fully recorded, heavily multiplexed volume holographic optical element or interconnection device, particularly in the case of neural network interconnections. For example, to the inventors' knowledge, there is no known prior technique for rapid copying of a volume hologram that is angularly multiplexed in two dimensions, other than that described in the parent and grandparent applications of the present application.
In the case of neural network interconnections, the training and/or learning sequences may be quite involved; in some cases, the training and/or learning sequences may result in a unique interconnection, and the exact learning sequence may not be reproducible in and of itself at all. In such cases, it is desirable to replicate the contents of the interconnection medium in such a manner that a
Jenkins B. Keith
Tanguay, Jr. Armand R.
Collins David W.
Henry Jon
University of Southern California
LandOfFree
Apparatus for simultaneous spatial modulation of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus for simultaneous spatial modulation of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus for simultaneous spatial modulation of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2870442