Linear and Symmetric Artificial Synapses Driven by Hydrogen Bonding for Accurate and Reliable Neuromorphic Computing

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic computing aims to overcome the von Neumann bottleneck by mimicking brain functions.
• Artificial synapses are key components, but halide perovskites suffer from instability and non-linear characteristics.
• Existing artificial synapses exhibit stochastic ion migration and thermal instability, impairing learning and inference.

Purpose of the Study:

• To develop a stabilization strategy for CsPbI3 artificial synapses.
• To improve the linearity, symmetry, and reliability of artificial synapses for neuromorphic computing.
• To demonstrate the potential of stabilized artificial synapses in large-scale image classification.

Main Methods:

• Interface engineering using polyvinyl alcohol (PVA) and hydrogen-bonding.
• Density functional theory (DFT) calculations and experimental characterization.
• Integration of stabilized artificial synapses into a neural network for image classification.

Main Results:

• PVA forms stable O-H···I- bonds, promoting vertical lattice ordering and directional ion migration.
• Achieved highly linear and symmetric conductance modulation (αp = 0.004, αd = 0.020).
• Demonstrated an eight-fold reduction in interfacial trap density and high-temperature retention (>10^4 s).
• Neural network integration achieved image classification accuracy within 1.62% of the theoretical limit.

Conclusions:

• The PVA-based interface engineering strategy effectively stabilizes CsPbI3 artificial synapses.
• This approach overcomes limitations of stochastic ion migration and thermal instability.
• The stabilized artificial synapses show significant potential for edge AI, autonomous systems, and cognitive modeling.