Two-dimensional fully ferroelectric-gated hybrid computing-in-memory hardware for high-precision and energy-efficient...

Area of Science:

• Materials Science
• Computer Engineering
• Artificial Intelligence

Background:

• The von Neumann bottleneck limits computing performance and energy efficiency.
• Integrating digital and analog computing in memory (CIM) is crucial for advanced AI but faces integration challenges.
• Novel hardware architectures are needed to overcome limitations in current AI processing.

Purpose of the Study:

• To propose and demonstrate a novel complementary metal-oxide semiconductor (CMOS)-compatible ferroelectric hybrid CIM platform.
• To integrate digital logic and analog computation using two-dimensional (2D) materials for enhanced AI.
• To assess the performance and efficiency of the developed CIM system for AI applications.

Main Methods:

• Fabrication of a ferroelectric hybrid CIM platform using solution-processable 2D molybdenum disulfide (MoS2) atomic-thin channels.
• Integration of Boolean logic and triggers for digital processing and multistage cell arrays for analog computation.
• Wafer-scale manufacturing achieving high yield (96.36%) and characterization of device performance (on/off ratio, endurance, retention, variations).

Main Results:

• Achieved high performance metrics: on/off ratio >10^7, endurance >10^12, retention >10 years, and low variations (~0.3%/~0.5%).
• Developed a compact 2D hybrid CIM system for dynamic tracking tasks.
• Demonstrated a high accuracy of 99.8% and a 263-fold improvement in power efficiency compared to graphics processing units (GPUs).

Conclusions:

• The proposed 2D ferroelectric-gated hybrid CIM platform offers a viable solution for overcoming the von Neumann bottleneck.
• This technology enables high-precision, energy-efficient AI processing through monolithic integration of digital and analog computing.
• The versatile CIM blocks show significant potential for various AI tasks, paving the way for next-generation intelligent systems.