Power-Efficient Multisensory Reservoir Computing Based on Zr-Doped HfO₂ Memcapacitive Synapse Arrays.

Area of Science:

• Materials Science
• Neuromorphic Engineering
• Computer Science

Background:

• Reservoir computing requires efficient hardware for temporal processing.
• Capacitive reservoirs offer potential power efficiency but remain underexplored.
• Existing resistive memristors have limitations in power consumption and performance.

Purpose of the Study:

• To develop a power-efficient oxide-based memcapacitive synapse (OMC) for reservoir computing.
• To investigate the potential of Zr-doped HfO2 (HZO) for capacitive reservoir applications.
• To demonstrate a multisensory processing system with high accuracy and low power consumption.

Main Methods:

• Fabrication of Zr-doped HfO2 (HZO) memcapacitive devices.
• Characterization of device nonlinearity and state richness for reservoir computing.
• Implementation and testing of the OMC-based reservoir computing system on benchmark tasks.
• Demonstration of a touchless user interface for virtual shopping.

Main Results:

• The OMC device exhibits nonlinearity and state richness suitable for reservoir computing.
• Achieved ultralow power consumption (≈113.4 fJ per spike), outperforming resistive reservoirs.
• Demonstrated high accuracy (>94%) in recognizing acoustic, electrophysiological, and mechanical data.
• Successfully implemented a proof-of-concept touchless user interface.

Conclusions:

• Zr-doped HfO2 memcapacitive synapses are effective for power-efficient reservoir computing.
• The developed system offers superior temporal processing and multisensory recognition capabilities.
• This work paves the way for advanced, low-power human-machine interfaces and machine learning platforms.