Symmetric n/p Schottky Barrier Modulation for Precision-Configurable Neural Network

Area of Science:

• Neuromorphic computing
• Solid-state device physics

Background:

• Conventional neuristors face limitations in balancing computational precision and energy efficiency for edge devices.
• Fixed precision in traditional devices necessitates trade-offs between accuracy and power consumption.

Purpose of the Study:

• To develop a single neuristor device capable of both high-efficiency nonlinear logic and high-precision linear operations for edge computing.
• To overcome the precision-efficiency trade-off inherent in current neuromorphic hardware.

Main Methods:

• Proposed a Schottky barrier neuristor with a global bottom gate to modulate the Schottky barrier, ensuring linear gate voltage-transconductance relationship.
• Utilized electrostatic doping and image force effects for uniform n-/p-type modulation and enhanced driving capability.
• Simulated an accelerator architecture based on the proposed neuristor.

Main Results:

• The Schottky barrier neuristor successfully integrates linear and nonlinear functionalities in one device.
• Achieved a 98.3% accuracy and 1359.62 TOPS/W energy efficiency in simulations.
• Demonstrated a significant reduction in device count (one-fifth) for nonlinear functions compared to silicon-based solutions.

Conclusions:

• The proposed neuristor design enables reconfigurable digital-analogue computing units for advanced edge AI applications.
• This innovation paves the way for more efficient and precise neuromorphic systems.
• The device offers a promising solution for next-generation low-power, high-performance edge computing.