High-Performance On-Chip Racetrack Resonator Based on GSST-Slot for In-Memory Computing

Area of Science:

• Photonics
• Materials Science
• Computer Engineering

Background:

• The von Neumann architecture causes significant power consumption and time delays in electronic computing due to data shuttling.
• Photonic in-memory computing using phase change materials (PCMs) offers a promising alternative for increased efficiency and reduced power usage.
• Existing PCM-based photonic computing units require improved extinction ratios and insertion loss for large-scale applications.

Purpose of the Study:

• To develop a high-performance photonic in-memory computing unit.
• To enhance the extinction ratio and reduce insertion loss in PCM-based optical computing devices.
• To demonstrate the potential of the proposed device for accurate and energy-efficient computations.

Main Methods:

• Fabrication of a 1x2 racetrack resonator using a Ge2Sb2Se4Te1 (GSST)-slot structure.
• Characterization of optical performance, including extinction ratio and insertion loss.
• Implementation of scalar multiplication operations and evaluation on the MNIST dataset for neuromorphic network applications.

Main Results:

• Achieved high extinction ratios of 30.22 dB (through port) and 29.64 dB (drop port).
• Demonstrated low insertion loss of ~0.16 dB (drop port, amorphous) and ~0.93 dB (through port, crystalline).
• Obtained a resonant wavelength tuning range of 7.13 nm and 94.6% recognition accuracy on MNIST, with 28 TOPS/W energy efficiency.

Conclusions:

• The GSST-slot racetrack resonator significantly improves extinction ratio and reduces insertion loss for photonic in-memory computing.
• The device enables accurate and energy-efficient scalar multiplication and neuromorphic computing applications.
• This work presents a viable approach for power-efficient, large-scale optical computing networks.