後退火處理對氧化鋁覆蓋層效應於氧化鉿鋯鐵電記憶體之電性分析與切換特性可靠度探討
No Thumbnail Available
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
隨著科技的迅速發展，如人工智慧(Artificial Intelligence, AI)、元宇宙(Metaverse)等新興概念，運用了大量的穿戴式裝置，甚至因為全球疫情關係產生出新型態的工作方式，使得電子產品的需求大增。而作為非揮發性(Non-volatile)，擁有低功耗、高微縮性的鐵電記憶體已成為極具潛力的角色。本實驗取代傳統的鐵電材料，以二氧化鉿(HfO2)作為鐵電薄膜的基底，而為了更容易誘發鐵電特性，選擇了可摻雜比例範圍較廣的鋯(Zr)來進行元素摻雜。氧化鉿鋯(HfZrO)電容可以透過提高金屬閘極的機械或熱應力，促使氧化鉿鋯薄膜晶相轉變而產生鐵電極化現象，本實驗選擇熱穩定性較佳的氮化鉭(TaN)作為金屬閘極，並藉由閘極氮含量4%、7%、10%的調變，來觀察氧化鉿鋯電容的電性變化，且結合金屬後退火處理(Post Metal Annealing, PMA)來提升應力以增強鐵電特性。氧化鉿鋯薄膜於退火處理後，可以提升兩倍殘餘極化量，然而鋯元素在高溫下較易使氧化鉿鋯電容有漏電流的現象，因此本實驗藉由添加2 nm、3 nm的氧化鋁介電覆蓋層來減少漏電流路徑的產生。隨後本實驗亦透過耐久度測試(Endurance)、定電壓測試(Constant Voltage Stress, CVS)來觀察經過氮化鉭氮含量調變、添加氧化鋁覆蓋層後，對氧化鉿鋯電容可靠度的影響。經由實驗結果可以發現，添加2nm的氧化鋁覆蓋層能夠有效提升元件的可靠度。本實驗得到的最佳化參數，為氮化鉭閘極氮含量7%，添加2 nm的氧化鋁介電覆蓋層，且經過450oC的後退火處理後，其兩倍殘餘極化量可達36.91 C/cm2，在-3 V量測電壓下的漏電流可以維持在1.65×10-6 A。耐久度測試方面，操作電壓為±3 V時，元件可以承受至1.62×107次循環，且在106的操作次數下能保有25.73 C/cm2的兩倍殘餘極化量。而在-2.6 V~-2.8 V的定電壓測試下，一萬秒內的漏電流機制則皆維持在電荷捕捉(Electron Trap)的狀態。最後，本實驗利用鐵電記憶體遲滯與非揮發性的優勢，將收斂後最佳化的條件與無添加氧化鋁覆蓋層的對照組，進行仿神經型態應用的比較。透過非線性公式所得到的結果，有添加2 nm覆蓋層的條件可以呈現較低的非線性係數，因此更能有效模擬仿神經突觸元件的運作，在未來期望能藉由資料訓練及參數調變，更進一步應用於深度學習與人工智慧的範疇。
With the rapid development of technology, such as artificial intelligence (AI), Metaverse and other emerging concepts, the use of a large number of wearable devices, and even the new working style are generated because of the global epidemic, so that the demand for electronic products has increased. As a non-volatile memory, ferroelectric memory with low power consumption and high miniaturized has become a potential role.Instead of the traditional ferroelectric materials, hafnium dioxide (HfO2) was used as the substrate of the ferroelectric thin film in this work. In order to induce ferroelectric properties more easily, the zirconium element with a wide range of doping was selected. Hafnium zirconium oxide (HfZrO) capacitors can promote the crystal phase transformation of hafnium zirconium oxide films by increasing the mechanical or thermal stress of the metal gate to produce ferroelectric polarization. In this work, tantalum nitride (TaN) with better thermal stability was selected as the metal gate, and the electrical characteristics of the hafnium zirconium capacitors were adjusted by modulating the nitrogen content of tantalum nitride to 4%, 7%, and 10%, and post metal annealing (PMA) was carried out to increase stress to enhance the ferroelectric properties. After annealing, the hafnium zirconium oxide thin film can produce more ferroelectric phase to reach better ferroelectric polarization. However, leakage current of hafnium zirconium oxide capacitors is easier to be found at high temperature situation. Therefore, leakage current paths are reduced by adding 2 nm, 3 nm aluminum oxide capping layers in this work. Afterwards, endurance and constant voltage stress (CVS) test was used to observe the influence on the reliability of hafnium zirconium oxide capacitors after modulating the nitrogen content of tantalum nitride and adding aluminum oxide capping layers. It can be found that reliability of the device by adding a 2 nm aluminum oxide capping layer was effectively improved in the result.Hafnium zirconium oxide capacitor with 7% nitrogen content of tantalum nitride gate by adding 2 nm aluminum oxide capping layer was the optimized condition in this work. After PMA at 450oC, the remanent polarization can reach 36.91 C/cm2, and the leakage current can be maintained at 1.65×10-6 A in -3 V measurement voltage. Under the optimized condition, this device can withstand up to 1.62×107 cycles in endurance test when the operating voltage was ±3 V, and remanent polarization retained 25.73 C/cm2 at 106 cycles. In the constant voltage test of -2.6 V~-2.8 V, the leakage current mechanism within 10k seconds was maintained in the state of electron trap. Finally, with the advantage of hysteresis and non-volatility of ferroelectric memory, optimized condition and control condition which without the addition of aluminum oxide capping layer were compared for neuromorphic applications. A lower non-linear coefficient under the optimized condition was obtained by the non-linear formula in the result, so it can more effectively simulate the operation of the neuromorphic synaptic component. In the future, it is expected that it can be further applied to the fields of deep learning and artificial intelligence through data training and parameter modulating.
With the rapid development of technology, such as artificial intelligence (AI), Metaverse and other emerging concepts, the use of a large number of wearable devices, and even the new working style are generated because of the global epidemic, so that the demand for electronic products has increased. As a non-volatile memory, ferroelectric memory with low power consumption and high miniaturized has become a potential role.Instead of the traditional ferroelectric materials, hafnium dioxide (HfO2) was used as the substrate of the ferroelectric thin film in this work. In order to induce ferroelectric properties more easily, the zirconium element with a wide range of doping was selected. Hafnium zirconium oxide (HfZrO) capacitors can promote the crystal phase transformation of hafnium zirconium oxide films by increasing the mechanical or thermal stress of the metal gate to produce ferroelectric polarization. In this work, tantalum nitride (TaN) with better thermal stability was selected as the metal gate, and the electrical characteristics of the hafnium zirconium capacitors were adjusted by modulating the nitrogen content of tantalum nitride to 4%, 7%, and 10%, and post metal annealing (PMA) was carried out to increase stress to enhance the ferroelectric properties. After annealing, the hafnium zirconium oxide thin film can produce more ferroelectric phase to reach better ferroelectric polarization. However, leakage current of hafnium zirconium oxide capacitors is easier to be found at high temperature situation. Therefore, leakage current paths are reduced by adding 2 nm, 3 nm aluminum oxide capping layers in this work. Afterwards, endurance and constant voltage stress (CVS) test was used to observe the influence on the reliability of hafnium zirconium oxide capacitors after modulating the nitrogen content of tantalum nitride and adding aluminum oxide capping layers. It can be found that reliability of the device by adding a 2 nm aluminum oxide capping layer was effectively improved in the result.Hafnium zirconium oxide capacitor with 7% nitrogen content of tantalum nitride gate by adding 2 nm aluminum oxide capping layer was the optimized condition in this work. After PMA at 450oC, the remanent polarization can reach 36.91 C/cm2, and the leakage current can be maintained at 1.65×10-6 A in -3 V measurement voltage. Under the optimized condition, this device can withstand up to 1.62×107 cycles in endurance test when the operating voltage was ±3 V, and remanent polarization retained 25.73 C/cm2 at 106 cycles. In the constant voltage test of -2.6 V~-2.8 V, the leakage current mechanism within 10k seconds was maintained in the state of electron trap. Finally, with the advantage of hysteresis and non-volatility of ferroelectric memory, optimized condition and control condition which without the addition of aluminum oxide capping layer were compared for neuromorphic applications. A lower non-linear coefficient under the optimized condition was obtained by the non-linear formula in the result, so it can more effectively simulate the operation of the neuromorphic synaptic component. In the future, it is expected that it can be further applied to the fields of deep learning and artificial intelligence through data training and parameter modulating.
Description
Keywords
鐵電記憶體, 氧化鉿鋯, 氮化鉭, 氧化鋁覆蓋層, FRAM, HfZrO, TaN, Al2O3 Capping layer