Antiferroelectric polarization enabling physical activation in CuBiP₂Se₆ for medical image processing

Area of Science:

• Materials Science
• Neuromorphic Computing
• Solid State Physics

Background:

• Antiferroelectric materials offer tunable polarization dynamics and layered van der Waals structures.
• These properties are promising for hardware-efficient neuromorphic computing and multifunctional integration.

Purpose of the Study:

• To demonstrate an antiferroelectric polarization-driven diode for simultaneous physical activation and computing-in-memory.
• To construct an in-sensor computing system for medical image classification.
• To develop a tunable activation circuit for adaptable neuromorphic systems.

Main Methods:

• Fabrication of an antiferroelectric heterojunction device.
• Integration of devices into an in-sensor computing system.
• Implementation and testing of a hardware-based tunable activation circuit.

Main Results:

• Demonstrated an antiferroelectric diode with an extended linear operating region.
• Achieved over 95% accuracy in medical image classification using the in-sensor computing system.
• Attained comparable accuracy and training loss to ideal activation functions with the hardware-based function.

Conclusions:

• Established a dual-functional antiferroelectric heterojunction for perception-computation integration.
• Highlighted potential for optically triggered, compact, low-power neuromorphic systems.
• Showcased applicability in medical image processing.