High-Linearity Ta₂O₅ Memristor and Its Application in Gaussian Convolution Image Denoising

Area of Science:

• Materials Science
• Computer Engineering
• Image Processing

Background:

• Gaussian filtering in image processing involves computationally intensive convolution operations.
• Traditional methods using Gaussian matrices strain system memory due to extensive multiplications and additions.

Purpose of the Study:

• To develop a hardware-based solution for efficient Gaussian filtering, reducing computational load.
• To explore the application of memristor devices in image convolution operations.

Main Methods:

• A W/Ta2O5/Ag memristor was fabricated and subsequently modified with a ZnO interlayer.
• The Ta2O5/ZnO heterostructure's linear pulse response was leveraged for conductance modulation.
• A 5x5 memristor array was assembled to act as a convolution kernel for Gaussian noise removal.

Main Results:

• The W/Ta2O5/ZnO/Ag bilayer memristor demonstrated improved linearity in pulse response.
• Memristor array-based denoising achieved results comparable to Gaussian matrix convolution, with an average loss of less than 5%.
• The memristor array effectively performed Gaussian noise removal in image processing.

Conclusions:

• Memristor devices offer a promising approach to mitigate computational overhead in convolution operations.
• The developed memristor array shows significant potential for image processing tasks, including applications in convolutional neural networks (CNNs).