The continuous advancement of computing technology has led to an increasing demand for efficient data processing methods. As data processing tasks become increasingly complex, traditional von Neumann computing systems suffer from bottlenecks, limiting large-scale data handling and overall system performance. To overcome these limitations, neuromorphic computing, inspired by the structure and principles of the biological brain, has emerged as a promising alternative. By mimicking the information processing between neurons and synapses, neuromorphic computing enables rapid and energy-efficient data processing. We fabricated an RRAM array memristor and demonstrated its potential as an artificial synapse for neuromorphic computing applications. The fabricated Ni/SiN/TiN memristor successfully emulated various synaptic plasticity learning rules, including spike-timing-dependent plasticity, paired-pulse facilitation. Based on its weight update characteristics, the memristor-based pattern recognition system achieved an accuracy of 95.36 %. Furthermore, by applying a combination of voltage pulses, we replicated adaptive learning behavior in a Pavlovian conditioning experiment. Additionally, the memristor successfully mimicked key nociceptive properties, such as threshold responses, allodynia, and hyperalgesia. Finally, by tuning the reset and compliance current, we successfully implemented multi-level cell storage, and an appropriate sequence of write/erase pulses enabled binary encoding of decimal numbers from 0 to 15, demonstrating the potential for four-state edge computing applications. These results highlight the feasibility of RRAM array memristors for neuromorphic computing, artificial synapse emulation, and energy-efficient edge computing applications.

中文翻译：

---

CMOS 兼容的基于 SiN 的忆阻突触，适用于边缘计算应用

计算技术的不断进步导致对高效数据处理方法的需求不断增加。随着数据处理任务变得越来越复杂，传统的 von Neumann 计算系统面临瓶颈，限制了大规模数据处理和整体系统性能。为了克服这些限制，受生物大脑结构和原理启发的神经形态计算已成为一种很有前途的替代方案。通过模拟神经元和突触之间的信息处理，神经形态计算实现了快速、节能的数据处理。我们制造了一个 RRAM 阵列忆阻器，并展示了它作为神经形态计算应用的人工突触的潜力。制造的 Ni/SiN/TiN 忆阻器成功模拟了各种突触可塑性学习规则，包括尖峰时间依赖性可塑性、配对脉冲促进。基于其权重更新特性，基于忆阻器的模式识别系统实现了 95.36% 的准确率。此外，通过应用电压脉冲的组合，我们在巴甫洛夫条件反射实验中复制了自适应学习行为。此外，忆阻器成功地模拟了关键的伤害感受特性，例如阈值反应、异常性疼痛和痛觉过敏。最后，通过调整复位和顺从电流，我们成功实现了多级单元存储，并且适当的写入/擦除脉冲序列实现了从 0 到 15 的十进制数的二进制编码，展示了四态边缘计算应用的潜力。这些结果突出了 RRAM 阵列忆阻器用于神经形态计算、人工突触仿真和节能边缘计算应用的可行性。