Optical computer architectures for supervised learning systems

McAulay, Alastair D.

doi:10.1109/2.144400

Cited by 11 publications

(16 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This two-by-two single basic block can be used for routing electron waves or for computations in a large multistage network, just like an optical network. 11 Finally and most importantly, with three-inputs-one-output arrangements, the possibility of an even higher-order three-bit computing, such as a full adder or one-dimensional cellular automation array, can be realized as we will show here. In this work we show that a full adder can be constructed using two four-terminal rings.…”

Section: Introductionmentioning

confidence: 82%

Four-terminal quantum resistor network for electron-wave computing

Ramamurthy

2002

Phys. Rev. B

View full text Add to dashboard Cite

Interconnected ultrathin conducting wires or, equivalently, interconnected quasi-one-dimensional electron waveguides, which form a quantum resistor network, are presented here in four-terminal configurations. The transmission behaviors through such four-terminal networks are evaluated and classified. In addition, we show that such networks can be used as the basic building blocks for a possible massive wave computing machine in the future. In a network, each interconnection, a node point, is an elastic scatterer that routes the electron wave. Routing and rerouting of electron waves in a network is described in the framework of quantum transport from Landauer-Buttiker theory in the presence of multiple elastic scatterers. Transmissions through various types of four-terminal generalized clean Aharonov-Bohm rings are investigated at zero temperature. Useful logic functions are gathered based on the transmission probability to each terminal with the use of the Buttiker symmetry rule. In the generalized rings, even and odd numbers of terminals can possess some distinctly different transmission characteristics as we have shown here and earlier. Just as an even or odd number of atoms in a ring is an important quantity for classifying the transmission behavior, we show here that whether the number of terminals is an even or an odd number is just as important in understanding the physics of transmission through such a ring. Furthermore, we show that there are three basic classes of four-terminal rings and the scaling relation for each class is provided. In particular, the existence of equitransmission among all four terminals is shown here. This particular physical phenomena cannot exist in any three-terminal ring. Comparisons and discussions of transmission characteristics between three-terminal and four-terminal rings are also presented. The node-equation approach by considering the Kirchhoff current conservation law at each node point is used for this analysis. Many useful logic functions for electron-wave computing are shown here. In particular, we show that a full adder can be constructed very simply using the equitransmission property of the four-terminal ring. This is in sharp contrast with circuits based on transistor logic.

show abstract

Section: Introductionmentioning

confidence: 82%

Four-terminal quantum resistor network for electron-wave computing

Ramamurthy

2002

Phys. Rev. B

View full text Add to dashboard Cite

show abstract

“…In fact as discussed earlier and in Ref. 8, we use an image intensifier to read a high resolution thin film array because the orange luminescence is weak. We follow the image intensifier with an optically addressable liquid crystal ͑LCLV͒ to return the incoherent light to the argon wavelength to complete a loop of our system.…”

Section: All-optical Systemsmentioning

confidence: 99%

“…Consequently, the energy available is small. For this reason we follow the electron trapping material with an image intensifier and a liquid crystal television 8,23 ͑LCLV͒. In this paper we use the designation OLCLV for the combination of image intensifier and LCLV.…”

Section: Using the Et Materials For Elemental Matrix Multiplicationmentioning

confidence: 99%

“…The simplest and earliest learning algorithms for supervised learning are the perceptron and Hebbian. 1,8 Their simplicity makes them good candidates for all-optical learning systems. The Hebbian is selected for unsupervised learning in this paper ͑Ref.…”

Section: Introductionmentioning

confidence: 99%

“…We have previously used electron trapping materials for optical supervised perceptron learning. 10 Electron trapping materials 8,[11][12][13] are used in this paper because they also have strong potential for an unsupervised all-optical alarm system. The material is well suited to storing and updating the filter history because it is a natural high resolution memory material with the potential for 1 m resolution.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Optical two‐dimensional alarm system using electron trapping optical material

McAulay

Wang

1996

Opt. Eng

View full text Add to dashboard Cite

An alarm system was constructed and tested to demonstrate for the first time an all-optical system that uses electron trapping material with unsupervised learning. The alarm system provides an output when there is a deviation from a time weighted average of a 2-D input image. Classic Hebbian unsupervised learning is used to update the 2-D time weighted average weight array recursively with an all-optical loop. The fabrication of micrometer resolution nanosecond response electron trapping material is described and equations derived to show for the first time that electron trapping materials correctly multiply in parallel elements in a 2-D input image with corresponding elements in a second 2-D weight array. The alarm system equations and associated optic implementation are modified for a binary system instead of a bipolar one because of positivity of intensity. The system uses two electron trapping spatial light rebroadcasters (SLRs), one stores the 2-D weight array and the other the scalar output for synchronization and delay of the all-optical loop. The experiment uses a folded configuration to save cost by using only one image intensifier and liquid crystal light valve for both SLRs. Optical experimental results show the correct functioning of the system. © 1996 Society of Photo-Optical Instrumentation Engineers.

show abstract

Ultrafast Digital-Optical Arithmetic Using Wave-Optical Computing

Haist

2008

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Optical computer architectures for supervised learning systems

Cited by 11 publications

References 5 publications

Four-terminal quantum resistor network for electron-wave computing

Four-terminal quantum resistor network for electron-wave computing

Optical two‐dimensional alarm system using electron trapping optical material

Ultrafast Digital-Optical Arithmetic Using Wave-Optical Computing

Contact Info

Product

Resources

About