- 學位論文
氧化鉿基電橋式記憶體類神經突觸特性之研究
HfOx-based Conductive Bridge Random Access Memory as Analog Neuromorphic Synapse
本文將於2025/07/22開放下載。若您希望在開放下載時收到通知，可將文章加入收藏。
摘要
由於馮·諾依曼架構的計算方式存在著弊端，使得新型態神經形態計算的發展變得非常重要。非揮發性記憶體中的電橋式記憶體由於其快速的切換速度、低功耗、可多階操作和具有互補式金屬氧化物半導體製程兼容性，成為人工電子突觸元件的熱門候選人之一。 本文提出了優化氧化鉿基電橋式記憶體的電導線性度和電阻切換穩定度的方法。首先，我們探討不同的上電極（銅和碲）對氧化鉿基電橋式記憶體的影響。在設定和重置過程中，上電極採用碲比採用銅表現出更漸變的電阻切換。其次，我們研究了MgO層對氧化鉿基電橋式記憶體的影響。插入MgO層的元件可以改善記憶體的耐久性和陡峭的電阻切換行為。在插入MgO層的這兩個元件中，Te/MgO/HfOx/TiN元件顯示出穩定的高阻態和好的線性度。第三，我們研究了改變MgO和HfOx層厚度對Te/MgO/HfOx/TiN元件的影響。論文的最後，我們研究了HfOx層在300°C退火對Te/MgO/HfOx/TiN元件的影響。此元件不僅提高了可靠性，例如耐久性和記憶能力，還改善了突觸特性中的多階特性和線性度。元件的增強及抑制行為在每次行為各包含500個脈衝數下的非線性度分別為1.53與2.79，，且元件循環的增強及抑制行為可穩定操作高達120次，總共含有120000個脈衝數。Te/MgO(1nm)/HfOx(10nm)/TiN CBRAM元件展現類比的電阻切換行為和出色的可靠性，適合用作神經形態計算系統的電子突觸元件。
並列摘要
Due to the drawbacks of the computing paradigm based on the von Neumann architecture, the development of novel neuromorphic computing has become very important. Conductive bridge random access memory (CBRAM) device in non-volatile memories is the one of the reliable candidates for artificial synaptic devices due to fast switching speeds, low power consumption, multilevel storage, and CMOS compatibility. In this thesis, we proposed methods to optimize conductance linearity and resistive switching stability in HfOx-based CBRAM device. Firstly, we investigated the effect of different top electrodes (Cu and Te) on HfOx-based CBRAM device. The Te-based CBRAM device exhibits more gradual switching than the Cu-based CBRAM device during the set and reset processes. Secondly, we investigated the effect of MgO interface layer on HfOx-based CBRAM device. By inserting an MgO layer, the endurance and abrupt switching behavior of HfOx-based CBRAM device can be improved. Among the two devices inserted into the MgO layer, the Te/MgO/HfOx/TiN device exhibits stable high resistance state (HRS) and better linearity. Thirdly, we investigated the effect of varying the thickness of the MgO and HfOx layer on Te/MgO/HfOx/TiN device. Finally, we investigated the effect of HfOx layer 300°C annealing on the Te/MgO/HfOx/TiN device. The device not only improves reliability such as endurance properties and retention characteristics, but also improves synaptic characteristics including multilevel characteristics and nonlinearity. The nonlinearities of potentiation and depression are 1.53 and 2.79 with 500 conductance states, and the device exhibites 120 training epochs with a total of 120000 pulse numbers. The Te/MgO(1nm)/HfOx(10nm)/TiN CBRAM device with analog switching behavior and excellent reliability can potentially be used as an artificial synaptic device for the neuromorphic computing system.
參考文獻
[1]P. Yao, et al. "Face classification using electronic synapses." Nat. Commun., vol. 8, no. 1, pp. 1-8, 2017.
[2]G. Hinton, et al. "Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups." IEEE Signal Process. Mag., vol. 29, no. 6, pp. 82-97, 2012.
[3]H. Hoang, S. Lee-Urban, and H. Muñoz-Avila. "Hierarchical Plan Representations for Encoding Strategic Game AI." AIIDE., pp. 63-68, 2005.
[4]M. H. Jarrahi, "Artificial intelligence and the future of work: Human-AI symbiosis in organizational decision making." Bus. Horiz., vol. 61, no. 4, pp. 577-586, 2018.
[5]G. Indiveri and S.-C. Liu, "Memory and information processing in neuromorphic systems." Proc. IEEE, vol. 103, no. 8, pp. 1379-1397, 2015.