Low-Voltage, CMOS-Free Synaptic Memory Based on LiTiO₂ Redox Transistors

Area of Science:

• Materials Science
• Computer Engineering
• Neuroscience

Background:

• Neuromorphic computers utilize analogue neural networks to reduce power consumption by integrating memory and logic.
• Current artificial synapses often require high write voltages and exhibit nonlinear weight updates, hindering efficient computation.

Purpose of the Study:

• To develop a novel inorganic redox transistor for artificial synapses.
• To achieve linear weight updates at low operating voltages and current densities.
• To demonstrate a sharp transistor subthreshold slope for improved neuromorphic performance.

Main Methods:

• Development of an inorganic redox transistor based on electrochemical lithium-ion insertion into LiTiO2.
• Minimization of open-circuit voltage and utilization of a low-voltage memristor selector.
• Operation in an electrochemically driven phase transformation regime.

Main Results:

• The LiTiO2 redox transistor demonstrated linear weight updates at low current densities and write voltages as low as 200 mV.
• Achieved an exceptionally sharp transistor subthreshold slope of 40 mV/decade.
• Overcame the limitations of high write voltages and nonlinear updates in conventional artificial synapses.

Conclusions:

• The developed LiTiO2 redox transistor offers a promising solution for efficient neuromorphic computing.
• Linear weight updates and low operating voltages pave the way for more power-efficient and predictable artificial neural networks.
• The device's sharp subthreshold slope enhances its potential for advanced computing applications.