Volatile ZrO₂ Antiferroelectric Tunnel Junctions for Rapid, Energy-Efficient Physical Reservoir Computing

Area of Science:

• Materials Science
• Condensed Matter Physics
• Nonlinear Dynamics

Background:

• Physical reservoir computing demands nonlinear response, fading memory, and diverse transient states.
• Conventional nonvolatile memories often require explicit resets or long relaxation times, hindering efficient computing.

Purpose of the Study:

• To develop novel antiferroelectric tunnel junctions (AFTJs) for physical reservoir computing.
• To leverage the intrinsic properties of ZrO2-based AFTJs for reset-free fading memory and enhanced nonlinear dynamics.

Main Methods:

• Fabrication of ZrO2-based two-terminal antiferroelectric tunnel junctions (AFTJs) with an amorphous In-Ga-Zn oxide interlayer.
• Stoichiometric control to identify the optimal 2:2:1-ZrO2 AFTJ composition.
• Experimental calibration of a circuit-level reservoir model for temporal information processing assessment.

Main Results:

• The optimal 2:2:1-ZrO2 AFTJ exhibited an I_on/I_off ratio of ~890, a peak nonlinearity factor of ~48.4, and a paired-pulse facilitation index of 1.79.
• Achieved 90.4% accuracy in Modified National Institute of Standards and Technology classification with reduced spatiotemporal dimensionality.
• Demonstrated low latency (~2 µs) and energy consumption (<480 pJ per operation) for a 40,000 µm² device.

Conclusions:

• ZrO2-based AFTJs offer intrinsic self-relaxation and reset-free fading memory, suitable for physical reservoir computing.
• The developed AFTJs enable efficient temporal information processing, including waveform classification and time series forecasting.
• Projected significant improvements in latency and energy consumption with device area scaling.