Oxide interface-based polymorphic electronic devices for neuromorphic computing

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computer Engineering

Background:

• Conventional AI hardware faces limitations in scaling, memory-computation separation, and energy efficiency.
• Emerging materials show promise but struggle with scalability, reproducibility, and compatibility.

Purpose of the Study:

• To demonstrate polymorphic electronic devices with programmable functionalities.
• To overcome limitations of conventional AI hardware using oxide heterostructures.

Main Methods:

• Utilized LaAlO3/SrTiO3 heterostructures with quasi-two-dimensional electron gas.
• Employed lateral gates to manipulate electronic properties, enabling polymorphic device functions.
• Integrated devices into circuits for digit recognition and logic operations.

Main Results:

• Demonstrated programmable transistor, memristor, and memcapacitor functionalities in a single device.
• Achieved digit recognition using transistor and memcapacitor circuits for reservoir computing.
• Performed logic operations with in-situ storage and reconfigurable synaptic logic using transistor and memristor circuits.

Conclusions:

• Developed a scalable, silicon-compatible, and energy-efficient platform for polymorphic and neuromorphic computing.
• Oxide-based monolithic integrated circuits offer a pathway to advanced AI hardware.
• The demonstrated devices support versatile applications from digit recognition to complex decision-making tasks.