HfO_x /AlO_y Superlattice-Like Memristive Synapse

Area of Science:

• Materials Science
• Computer Engineering
• Nanotechnology

Background:

• Memristors offer efficient vector-matrix multiplication for neuromorphic computing.
• Controllable multi-level conductance is crucial for synaptic applications.
• Existing memristors often suffer from abrupt filament formation and rupture.

Purpose of the Study:

• To theoretically design and validate a superlattice-like (SLL) structure switching layer for multi-level memristors.
• To refine conductive filament (CF) growth for gradual conductance switching.
• To enhance memristor performance for neuromorphic computing applications.

Main Methods:

• Fabrication of Ti/(HfO_x /AlO_y )_SLL /TiN memristors.
• Characterization using transmission electron microscopy (TEM) and X-ray photoelectron spectroscopy (XPS).
• Theoretical validation through ab initio calculations and nudged elastic band (NEB) calculations.

Main Results:

• The SLL structure of HfO_x /AlO_y film was confirmed.
• Optimized SLL memristors demonstrated linear synaptic weight updates (α = 1.06).
• A convolutional neural network (CNN) using SLL memristive synapses achieved 94.95% handwritten digit recognition accuracy.

Conclusions:

• The SLL memristor design enables controllable and gradual conductance switching.
• The proposed memristor exhibits fast operation (30 ns), long retention (≥ 10⁴ s at 85 °C), scalability, and CMOS compatibility.
• This work provides a design strategy for advanced neuromorphic computing devices.