Design of optoelectronic activation, local memory and weighting circuits for Compact Integrated Optoelectronic Neural (COIN) Co-processor
Author(s)Tadele, Wegene Haile
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science.
MetadataShow full item record
The Compact Integrated Optoelectronic Neural (COIN) Co-processor, a prototype of artificial neural network implemented in hybrid optics and optoelectronic hardware, aims to implement a multi-layer neural network algorithm by performing parallel and efficient neural computations. In this thesis, we design and implement optoelectronic thresholding (activation), weighting and memory circuits for the COIN processor. The first version involved the design of fixed thresholding and weighting functions. The second version incorporated a local capacitive memory element as well as variable weighting schemes. The third version introduces an additional flexibility for variable thresholding by changing the bias voltages of control transistors. A 9x9 array of proof of concept printed circuit board (PCB) with an area of 4.5x 4.5 in² and total power consumption of 1.37W was designed and tested for version-I optoelectronic neuron architecture. A spice simulation was performed for the last two versions for integrated circuit (IC) implementation. The work developed in this thesis provides some guidance on the design of optoelectronic neural activation function for the realization of the embodiment of the fully integrated COIN Co-processor to be built in the future.
Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2015.Cataloged from PDF version of thesis. Pages 7 and 8 are missing.Includes bibliographical references (pages 99-103).
DepartmentMassachusetts Institute of Technology. Department of Electrical Engineering and Computer Science.; Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Massachusetts Institute of Technology
Electrical Engineering and Computer Science.