Oscillatory Neural Network with Tunable Frequency for Brain-Inspired Neuromorphic Computing.

Area of Science:

• Neuromorphic Engineering
• Solid-State Devices
• Integrated Circuits

Background:

• Neuromorphic computing aims to mimic the brain's structure and function for efficient computation.
• Existing neuromorphic hardware often faces challenges in scalability and energy efficiency.
• Oscillatory neural networks (ONNs) offer a promising paradigm for brain-inspired computing.

Purpose of the Study:

• To introduce a novel silicon transistor-based oscillator with frequency tunability (SOFT).
• To demonstrate the potential of SOFT devices for implementing neuromorphic computing tasks.
• To enable low-cost, high-density ONNs through standard fabrication processes.

Main Methods:

• Fabrication of homologous single transistor-based oscillators (1T-O) and resistors (1T-R) using complementary metal-oxide-semiconductor (CMOS) technology.
• Integration of 1T-O and 1T-R devices on a single wafer, leveraging their structural identity.
• Demonstration of template matching using coupled SOFT devices and temporal signal classification via first-harmonic injection locking (FHIL) with multiple SOFTs.

Main Results:

• Successful implementation of frequency tunability in a silicon transistor-based oscillator (SOFT).
• Demonstration of template matching and temporal signal classification using SOFT devices.
• Achieved high-density integration and potential for energy-efficient, scalable ONNs.

Conclusions:

• SOFT devices offer a viable solution for building energy-efficient and scalable neuromorphic computing systems.
• The use of standard CMOS fabrication makes SOFT technology cost-effective and readily implementable.
• This work paves the way for advanced brain-inspired computing architectures.