Self-Rectifying Memristors Based on Dimensionally Graded Halide Perovskites

Area of Science:

• Materials Science
• Electronics Engineering
• Computer Science

Background:

• Neuromorphic in-memory computing addresses the data deluge in AI and smart electronics.
• Memristor crossbar arrays are key hardware for in-memory computing.
• Halide perovskites offer potential for memristors due to their properties, but sneak-path issues hinder performance.

Purpose of the Study:

• To develop self-rectifying halide memristors to overcome sneak-path challenges in memristor crossbar arrays.
• To create a dimensionally graded perovskite structure for improved memristor functionality.
• To enhance the efficiency and accuracy of neuromorphic in-memory computing.

Main Methods:

• Fabrication of a 2D to 3D dimensionally graded halide perovskite memristor.
• Engineering a heterojunction using specific 2D spacer cations and methylammonium lead iodide.
• Characterization of device rectification, endurance, linearity, and performance in a large crossbar array.

Main Results:

• Achieved a high rectification ratio greater than 10^3.
• Demonstrated robust synaptic characteristics with endurance exceeding 4 × 10^4 pulses and high linearity in weight updates.
• Successfully suppressed sneak currents, enabling a 140 × 140 crossbar array with 93% accuracy in image classification despite write noise.

Conclusions:

• The novel dimensionally graded perovskite memristor effectively solves the sneak-path problem.
• This self-rectifying memristor technology is suitable for large-scale neuromorphic computing arrays.
• The approach significantly advances the potential of halide perovskites in AI hardware.