Design of Neural Networks by Using Multiplexed Optical Neurons

Please enable javascript in your browser.

Jalili, Sajjad | 2021

1255 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 54648 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Mehrany, Khashayar; Memarian, Mohammad
Abstract:
Electronic processors have encountered Moore’s limitations and they can not meet expectations following computer science advances in new algorithms such as neural networks. These limitations have brought research to focus on light as a high-capacity carrier of information. Light’s degrees of freedom can be exploited whenever the structure has devices able to carry, keep, manipulate and detect each DoF's information but this exploitation may itself lead to either a large footprint or more power consumption. Moreover, there are some scientific and technological difficulties in matching different computational algorithms to these DoF’s requirements. As a result, one must consider all advantages and limitations of each degree of freedom to design a suitable structure. This thesis designs a neural network by using multiplexed optical neurons and seeks a more efficient structure by multiplexing data on different degrees of freedom of light such as angular momentum, polarization, spatial frequency, or wavelength
Keywords:
Neural Networks ; Optical Computing ; Nonreciprocal Time Variant Structures ; Micro-ring Resonator ; Optical Multiplexing