Lim Chanjin, Kim Taegil, Park YoungJu, Kim Daeho, Shin ChaeHo, Ha Suji, Lin Jin-Liang, Li Yuan, Park Junwoo
Department of Chemistry, Sogang University, Seoul, 04107, Republic of Korea.
Bruker Nano Surface, Bruker Korea Co, Ltd., Seoul, 05840, Republic of Korea.
Adv Sci (Weinh). 2025 Jun;12(23):e2505016. doi: 10.1002/advs.202505016. Epub 2025 Apr 30.
This paper demonstrates the use of molecular artificial synapses in neuromorphic computing systems designed for low energy consumption. A molecular junction, based on self-assembled monolayers (SAMs) of alkanethiolates terminated with 2,2'-bipyridine complexed with cobalt chloride, exhibits synaptic behaviors with an energy consumption of 8.0 pJ µm. Conductance can be modulated simply by applying pulses in the incoherent charge transport (CT) regime. Charge injection in this regime allows molecules to overcome the low energy barrier for C─C bond rotations, resulting in conformational changes in the SAMs. The reversible potentiation/depression process of conductance achieves 90% accuracy in recognizing patterns from the Modified National Institute of Standards and Technology (MNIST) handwritten digit database. The molecular junction further exhibits both rectifying and conductance hysteresis behaviors, showing potential for use in selector-free synaptic arrays that efficiently suppress sneak currents.
本文展示了分子人工突触在为低能耗设计的神经形态计算系统中的应用。一种基于用与氯化钴络合的2,2'-联吡啶封端的链烷硫醇自组装单分子层(SAMs)的分子结,展现出能耗为8.0 pJ µm的突触行为。通过在非相干电荷传输(CT) regime中施加脉冲,电导可被简单调制。在该 regime中的电荷注入使分子能够克服C─C键旋转的低能垒,导致SAMs发生构象变化。电导的可逆增强/抑制过程在从修改后的美国国家标准与技术研究院(MNIST)手写数字数据库识别模式时达到90%的准确率。该分子结还展现出整流和电导滞后行为,显示出在有效抑制潜行电流的无选择器突触阵列中的应用潜力。