Wafer-Scale Ag₂S-Based Memristive Crossbar Arrays with Ultra-Low Switching-Energies Reaching Biological Synapses

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Memristive crossbar arrays (MCAs) are crucial for energy-efficient neuromorphic computing.
• Current complementary metal-oxide-semiconductor (CMOS)-compatible MCAs consume significant energy due to slow memristor kinetics.
• Bridging the energy gap between artificial and biological synapses remains a key challenge.

Purpose of the Study:

• To develop wafer-scale memristive crossbar arrays (MCAs) with energy consumption comparable to biological synapses.
• To enable CMOS-compatible fabrication of MCAs at low temperatures for broader applications.
• To demonstrate the potential of these MCAs in deep learning tasks like image recognition.

Main Methods:

• Fabrication of Ag₂S-based MCAs using CMOS-compatible processes below 160°C.
• Optimization of Ag₂S electrolyte microstructure to enhance Ag⁺ ion migration.
• Integration of MCAs with advanced training algorithms to compensate for memristor nonidealities.

Main Results:

• Achieved wafer-scale Ag₂S-based MCAs with a record low switching threshold of approximately -0.1 V.
• Demonstrated ultra-low switching energies in the femtojoule range, comparable to biological synapses.
• Attained 92.6% accuracy in image recognition simulations using the developed MCAs.

Conclusions:

• The Ag₂S-based MCAs offer a promising pathway towards ultra-high energy-efficient neuromorphic computing.
• Low-temperature fabrication enables integration with flexible electronics.
• Advanced training algorithms effectively compensate for device nonidealities, unlocking deep learning potential.