Integrated artificial neurons from metal halide perovskites

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Hardware neural networks offer significant energy efficiency over conventional computers.
• Artificial neurons are crucial for neural networks, with memristive devices showing promise due to size and stochasticity.
• Existing memristive artificial neuron demonstrations are limited.

Purpose of the Study:

• To demonstrate a fully on-chip artificial neuron using halide perovskite semiconductors.
• To investigate the energy efficiency and functionality of the memristive artificial neuron.
• To assess the potential for integrating these neurons into energy-efficient neural networks.

Main Methods:

• Fabrication of an on-chip artificial neuron using microscale electrodes and a halide perovskite semiconductor active layer.
• Integration of a halide perovskite memristive device in series with a capacitor.
• Simulation of neuron populations to analyze stochastic firing and sub-threshold input detection.

Main Results:

• The device exhibited stochastic leaky integrate-and-fire behavior.
• Achieved low energy consumption of 20 to 60 picojoules per spike, surpassing biological neurons.
• Demonstrated detection of sub-threshold inputs through stochastic firing patterns.

Conclusions:

• The halide perovskite memristive device represents a viable and energy-efficient artificial neuron.
• The demonstrated neuron's stochasticity aids in detecting weak signals, crucial for neural computation.
• This technology is readily integrable with existing artificial synapses for building efficient neural networks.