Neuromorphic computing, which mimics the brain's efficient processing capability, is a promising solution to the energy challenges of artificial intelligence and edge computing. Floating-gate memories (FGMs) are key building blocks for such systems, but their development has been hindered due to limited dynamic range, state instability, and conductance noise.
In a study published in Nature Communications, a research team led by Prof. WANG Zhenxing from National Center for Nanoscience and Technology, China (NCNST) demonstrated an 11-bit 2D MoS2-based FGM that exhibits a large on/off ratio, a high stability and a low noise level. The team employed bismuth as contact electrodes to eliminate the Schottky barrier, significantly boosting the on-state current to 100 μA and the on/off ratio to 108, while reducing current noise by 3 times.
The researchers also introduced a dual-pulse state editing method to stabilize memory states. This approach applies a tuning voltage pulse after programming or erasing to release trapped charges in the dielectric, minimizing conductance decay. Furthermore, the device operates in a gate-injection mode, which localizes charge injection away from the channel, preserving low noise levels even after 10⁵ operation cycles and at 85 °C.
As a result, the team realized up to 2249 clearly distinguishable conductance levels which surpass most charge-trap memories and rivaling top-performing resistive random-access memories (RRAMs), with a switching speed of 230 ns. The device also exhibited long-term retention over 104 seconds and robust endurance over 105 program/erase cycles. Theoretical analysis suggests that the bit capacity can be further increased to 17 bits by reducing interfacial defect density.
This study sets a new performance benchmark for FGMs and highlights the potential of 2D materials in high-precision, energy-efficient neuromorphic computing systems.
Contact: WANG Zhenxing
National Center for Nanoscience and Technology (NCNST)
E-mail: wangzx@nanoctr.cn
