Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 54064 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Khavasi, Amin
- Abstract:
- Analog computing has emerged as a promising candidate for Neural networks' implementation due to its high Interconnectivity, high bandwidth, parallel computing, high-speed processing, and low power consumption. Artificial Neural Networks have a wide range of applications; however, the Implementation of complicated Neural Networks on traditional computers would encounter two fundamental obstacles: limited processing speed and non-optimal energy consumption. This thesis's primary focus is on designing and simulating a whole-passive planar Optical neural network(ONN) based on silicon photonics technology. Firstly, the concept of ONN is studied using some lately proposed work. The device is designed based on Metaline layers consisted of slot arrays on silicon on insulator substrate(SOI). After determining the device structure, an appropriate electromagnetic numerical model has been chosen and used to simulate and train the structure. Also, To demonstrate the capability of our ONN, we benchmarked its performance on handwritten digits classifcation, which achieved an accuracy of 88.8 percent that is comparable to the state of the art. Then, for verifcation of our results, A reduced-size ONN structure is designed. The reduced design is simulated for handwritten digit images which are selected randomly from the test dataset. The simulation results shows 91% matching with analytical results.Secondly, we continue to Modify the electromagnetic model to prepare the device to perform complex matrix multiplication. by developing the proposed ONN; it is possible to achieve parallel processing and on-chip data analysis, all based on silicon photonics. Besides, the optical neural network can do these operations with low power consumption and rapidly. Additionally, the proposed structure is compact. Thus it can find widespread application in critical power situations such as mobile devices.
- Keywords:
- Optical Neural Networks ; Optical Computing ; Metasurfaces ; Analog Computing ; Matrix Multiplication ; Silicon Photonics
Digital Object List-
محتواي کتاب - view
Bookmark