Sb-contacted MoS₂ flash memory for analogue in-memory searches.

Area of Science:

• Materials Science
• Computer Engineering
• Artificial Intelligence

Background:

• Traditional hardware faces bottlenecks due to slow data transfer between memory and processing, hindering AI and edge device performance.
• Existing content-addressable memories, while promising, are limited by silicon transistor performance.

Purpose of the Study:

• To introduce a novel analogue content-addressable memory solution.
• To overcome the limitations of current memory technologies for AI applications.

Main Methods:

• Utilized atomically thin two-dimensional (2D) MoS2 flash memories with semimetal antimony contacts.
• Implemented an 8x16 analogue content-addressable memory array with 256 MoS2 flash memory devices.
• Demonstrated analogue Hamming distance computing for k-nearest neighbour classification.

Main Results:

• Achieved high read-out current (60 μA μm-1) and ON/OFF ratios (>10^9) in 2D MoS2 flash memories.
• Attained very low energy consumption (under 0.1 fJ per search per cell) and latency (36 ps) during in-memory search operations.
• Showcased high accuracy, energy efficiency, and low latency for machine learning tasks.

Conclusions:

• The developed 2D MoS2 flash memory offers a transformative solution for in-memory computing.
• This technology has the potential to enable more efficient and scalable computing for intelligent edge devices.
• Highlights the significant role of 2D materials in advancing hardware capabilities for AI.