A Memristive Multilayer Cellular Neural Network With Applications to Image Processing.

Area of Science:

• Artificial Intelligence
• Neuroscience
• Electrical Engineering

Background:

• Memristors are explored for compact synaptic function implementation in neural networks.
• Cellular Neural Networks (CNNs) are suitable for massively parallel analog processing.

Purpose of the Study:

• To present a novel memristive multilayer CNN (Mm-CNN) model.
• To analyze the performance and applications of the proposed Mm-CNN.
• To leverage memristor crossbar circuits for efficient synaptic weighting.

Main Methods:

• Designed a memristive multilayer CNN (Mm-CNN) architecture.
• Utilized memristor crossbar circuits to implement signed synaptic weights.
• Developed efficient Mm-CNN cell circuits for complex weighted summation.
• Conducted simulations for performance analysis in image processing applications.

Main Results:

• The Mm-CNN achieves compact and linear synaptic weighting using pairs of memristors.
• Efficient execution of complex weighted summation is demonstrated.
• Simulations showcase the model's effectiveness in image processing tasks.

Conclusions:

• The proposed Mm-CNN offers significant advantages including compactness, nonvolatility, versatility, and programmability.
• This novel architecture is well-suited for advanced neural network applications, particularly in image processing.