Highly reliable forming-free conductive-bridge random access memory via nitrogen-doped GeSe resistive switching layer

Conductive-bridge random access memory (CBRAM) is gaining attention as a non-volatile memory device for next-generation storage-class applications. However, CBRAM cells exhibit stochastic natures during continuous bi-stable resistive switching, stemming from the randomness of high-mobility metal ions in the resistive switching layer. This randomness limits wafer-scale integration with complementary metal–oxide–semiconductor (CMOS) circuits. In this study, we fabricated a reliable forming-free CBRAM cell consisting of a Pt capping layer, a Cu active source layer, a nitrogen-doped GeSe resistive switching layer, and a W bottom electrode. We compared the continuous resistive switching loops with and without nitrogen contents in the GeSe layer, demonstrating that the nitrogen-doped GeSe CBRAM cell improved electrical variation for the forming and set voltages to below 10%. Using this nitrogen-doped GeSe-based CBRAM cell, we achieved outstanding synaptic plasticity characteristics compared to un-doped GeSe-based CBRAM cells. Finally, we designed a small-scale deep neural network trained with a hardware-based backpropagation learning rule, achieving recognition accuracy of up to 95.57% on handwritten image datasets. Our study demonstrates that the nitrogen-doped GeSe-based CBRAM cell can achieve high reliability and stable synaptic plasticity, thereby contributing to the advancement of next-generation memory technologies.