Non-Volatile Phase Modulation with Ultralow Energy Consumption Enabled by 2D Ferroelectric/TMD Heterostructures.

Area of Science:

• Materials Science
• Photonics
• Computer Engineering

Background:

• Non-volatile, low-loss phase modulation with low energy consumption is crucial for photonic in-memory computing.
• Existing photonic memory solutions often require large footprints, leading to high insertion loss and energy use.
• Novel materials and device architectures are needed to overcome these limitations.

Purpose of the Study:

• To demonstrate a compact, non-volatile phase modulator with ultra-low energy consumption for photonic in-memory computing.
• To leverage 2D ferroelectric materials for efficient optical index tuning.
• To showcase the device's potential in optical neural networks.

Main Methods:

• Fabrication of a heterostructure device comprising WS2/CIPS/graphene on a SiN microring resonator.
• Utilizing Cu+-induced polarization in CuInP2S6 (CIPS) for electrostatic tuning of WS2 refractive index.
• Characterization of switching energy, write speed, insertion loss, memory retention, and multi-level storage capabilities.

Main Results:

• Demonstrated a compact non-volatile phase modulator with ultra-low switching energy (2.5 pJ/cycle) and insertion loss (0.2 dB).
• Achieved a fast write speed of 5 V/µs and stable 8-bit multi-level memory with projected >10-year retention.
• Integrated the modulator into an optical neural network, achieving 92% accuracy on MNIST handwritten digit recognition.

Conclusions:

• The developed WS2/CIPS/graphene heterostructure offers a promising solution for high-performance photonic in-memory computing.
• The device's compact size, low energy consumption, and non-volatile multi-level memory are key advantages.
• This work establishes new avenues for developing efficient hardware-accelerated neural networks.