Low-power artificial neuron networks with enhanced synaptic functionality using dual transistor and dual memristor

Area of Science:

• Neuromorphic Engineering
• Artificial Intelligence Hardware

Background:

• Artificial neurons aim to mimic biological systems for enhanced neural network computing.
• High energy consumption in artificial neuron circuits is a major challenge.
• Memristors show promise for efficient synaptic implementation in hardware.

Purpose of the Study:

• To design and implement an efficient circuit-level artificial neuron network using memristor-based synapses.
• To achieve low power utilization in artificial neurons based on a modified Morris-Lecar model.
• To address the challenges of integrating memristive circuitry into neuron hardware.

Main Methods:

• Designed a mixed CMOS memristor artificial neuron network incorporating a memristor synapse model.
• Implemented artificial neurons in standard CMOS technology with a focus on low power consumption.
• Utilized a Dual Transistor and Dual Memristor (DTDM) synapse circuit for memristor-based artificial neurons.
• Simulated the circuit using Spectre with 45 nm CMOS technology.

Main Results:

• The proposed memristor-based Morris-Lecar (ML) neuron with a DTDM synapse circuit demonstrated significantly low power consumption (12.55 pW).
• Achieved a high spiking frequency of 22.72 kHz.
• Recorded an energy efficiency of 2.13 fJ per spike.

Conclusions:

• The developed memristor-based artificial neuron network offers a viable solution for low-power, high-performance neural computing.
• The DTDM synapse circuit effectively integrates memristors for efficient artificial neuron implementation.
• This research contributes to advancing neuromorphic hardware with bio-inspired, energy-efficient designs.