Dual-Gated Memtransistor Crossbar Array for Neuromorphic Computing with Sneak-Current Mitigation

Dual-Gated Memtransistor Crossbar Array for Neuromorphic Computing with Sneak-Current Mitigation

SHORT DESCRIPTION

A dual-gated memtransistor crossbar array that electronically controls learning rates and independently manages read and write operations for enhanced neuromorphic computing.

• Mark Hersam*, PhD
  • Chair and Walter P. Murphy Professor of Materials Science and Engineering, Northwestern University McCormick School of Engineering
  • Director, Materials Research Science and Engineering (MRSEC)
• Hong-Sub Lee
• Vinod Sangwan

NU Tech ID 2019-168

IP STATUS

Issued U.S. Patent 12,183,811

DEVELOPMENT STAGE

TRL-3 - Experimental Proof-of-Concept: Laboratory tests have confirmed the core functionality of the dual-gated design in simulated neuromorphic environments.

BACKGROUND

Conventional computer architectures based on CMOS face challenges with high power consumption and fixed weight update rules in neuromorphic systems. Existing memristive devices struggle with sneak current interference and lack dynamic control. These limitations create a need for advanced, scalable devices that can meet the demands of next-generation artificial neural networks.

(a) Schematic illustration of a dual-gated MoS2 memtransistor crossbar array with global silicon back gate and local top gate electrodes. (b) Diagram of the dual-gated memtransistor crossbar array. (c) False-colored scanning electron microscopy (SEM) image of a 10 x 9 crossbar array. (d) Zoomed-in SEM image of the active area indicated by the red rectangle in (c). (e) Histogram of the switching ratio of 34 MoS2 memtransistors.

ABSTRACT

This invention introduces a dual-gated memtransistor crossbar array designed for neuromorphic computing. It electronically controls learning rates and separates read and write operations. The integration of dual gates mitigates sneak current crosstalk, offering improved performance over single-gated designs. Laboratory experiments validate the core function and demonstrate enhanced device behavior in neuromorphic applications.

MARKET OPPORTUNITY

The global neuromorphic computing market was valued at approximately $5.23 billion in 2023 and is projected to reach $28.77 billion by 2032, expanding at a powerful CAGR of 20.86% (Source: Zion Market Research, 2024). This growth is primarily fueled by the "energy crisis" in artificial intelligence, where traditional CMOS-based GPU and TPU architectures are reaching their thermal and efficiency limits when running massive neural networks.

A critical sub-sector within this space is the memristor market, which is expected to witness explosive growth from $275.4 million in 2024 to over $7.76 billion by 2032, at a staggering CAGR of 52.3% (Source: Fortune Business Insights, 2025). This technology represents the missing link for next-generation AI, as memristors act as artificial synapses capable of both storing memory and processing data in a single location.

APPLICATIONS

• Neuromorphic computing: Implements hardware-based neural networks for energy-efficient data processing.
• Crossbar hardware accelerators: Delivers high-density memory arrays for rapid AI computations.
• Edge computing classifiers: Supports fast decision-making in compact, low-power devices.
• Nonvolatile memory: Retains synaptic weights to enable continuous learning processes.

ADVANTAGES

• Electrically controlled learning rates: Allows dynamic tuning for optimized performance.
• Sneak current mitigation: Reduces interference in high-density crossbar arrays.
• Independent read/write operations: Enhances reliability and device performance.
• Scalable architecture: Supports high-density integration in neuromorphic systems.

PUBLICATIONS

• Mark Hersam et al., Dual-Gated MoS2 Memtransistor Crossbar Array, Advanced Functional Materials, Volume 30, Issue 45