STT-MRAM을 위한 MgO 터널 베리어 기반 자기터널접합의 열화 특성에 대한 연구

Abstract

스핀 전달을 이용한 자기메모리 STT-MRAM은 차세대 메모리의 대표주자로서, 비휘발성, 빠른 동작 속도, 무한대의 재기록성, 높은 집적도, 낮은 소비 전력, 일반적인CMOS 공정의 호환이 가능한 장점을 가지고 있다. STT-MRAM은 자기터널접합 (MTJ) 소자를 기억소자로 사용하는데 강자성체 물질 사이에 얇은 절연막을 삽입하고 하나의 강자성체내의 자기 방향을 고정해 놓고 (고정층) 전류 주입 방향에 따라 다른 하나의 강자성체내의 자기 방향을 조절할 수 있도록 하여 (자유층) 자기방향의 평행, 반평행 상태에 따라 나타나는 저항을 데이터로 저장한다. 이때 평형, 반평행 상태에 따른 저항값의 변화를 터널자기저항비 (TMR) 이라고 한다. 현재 MTJ의 터널 베리어는 결정화된 MgO를 사용하고 있으며, 그 이유는 (001) 방향으로 성장된 MgO 터널 베리어의 경우 기존에 사용되던 AlO 터널 베리어보다 큰 TMR 효과를 확인할 수 있었기 때문이다. 그러나MgO 베리어의 경우 ~1nm 정도의 매우 얇은 두께로 제작 되어야 하기 때문에 추후 동작시의 전압/전류에 따른 barrier 막의 열화 및 고장 특성이 신뢰성에 큰 문제를 발생시킬 것이다. 최근에는 많은 연구를 통해 MgO 기반의 MTJ의 열화 및 고장특성에 대한 연구 결과들이 발표되고 있다. 본 논문의 연구 주제는 MgO 기반의MTJ의 신뢰성 및 열화 특성 연구이다. 챕터 1에서는 기본적인 MRAM과 MTJ의 소개를 진행하였고, 챕터2에서는 기존 CMOS 에서의 얇은 산화막의 신뢰성에 대해 설명하였다. 챕터 3에서는 소자 제작 방법 및 다양한 실험 분석 방법을 중점적으로 설명하였다. 챕터 4에서는 MgO 기반의 MTJ의 신뢰성 모델링을 진행하였다. MgO 기반의 MTJ의 열화특성 연구를 통해, 기존의 Time Dependent Dielectric Breakdown (TDDB) Model을 이용하여 MgO 기반의 MTJ 에서의 적합한 TDDB Model을 제안하였다. 챕터 5에서는 제작된 샘플에 전기적인 스트레스를 인가하여 측정된 TDDB 결과와 MgO parameter를 이용한 TDDB Model Curve를 비교하였다. 온도 변화에 따른 1/E Model에 self-heating을 적용한 Model curve가 실험결과와 가장 적합한 결과를 확보할 수 있었다. 또한 Mg layer를 통한 MgO 계면 상태 변화에 따른 TDDB 특성을 확보한 결과, MgO 계면이 계면 상태가 나쁠수록 1/E Model에, 계면 상태가 좋을 수록 Power-law Model에 더 적합한 결과를 확인할 수 있었다. 종합적으로 온도/계면 상태에 따라 열화 및 고장 특성이 달라지며, 함께 TDDB Model 을 차별화해서 적용해야 할 것으로 보인다. 챕터 6에서는 불량 MTJ를 효율적으로 스크린 할 수 있는 방법에 대해 연구를 진행하였다. Constant voltage stress (CVS)를 통해 열화/고장 특성 변화와 Interval voltage stress (IVS)를 통해 trap/detrap 현상을 확인하였다. 각각의 특성 확인을 통해 열화 및 고장 특성이 베리어의 두께 및 온도 변화에 의존함을 확인할 수 있었으며 이는 베리어 내의 trap 및 결함이 증가하기 때문으로 추정된다. 또한 베리어 두께, 온도, stress 변화에 따라 일정한 비율로 변화하는 Time to breakdown (TBD) 확인 할 수 있었으며, 이 결과를 통해 비정상적인 MTJ를 효율적으로 스크린 할 수 있는 방법을 제시하였다. 챕터 7에서는 1) Mg layer의 유/무에 따른 MgO 계면 상태 2) 버퍼층 의 표면 거칠기 변화에 따른 열화 특성을 확인하였다. Mg layer의 유무 및 바이어스 방향에 따라 열화 특성이 달라짐을 확인하였다. 이 결과를 통해 Mg layer 가 MgO의 결정성을 좋게 만들어 MgO/Mg 계면상의 trap을 감소 시키기 때문이다. 또한 버퍼층 의 표면 거칠기가 나쁠수록 열화 특성이 나빠짐을 확인하였다. 버퍼층의 표면 거칠기가 MgO 터널 베리어의 두께/결정성/거칠기에 영향을 미쳐 열화 및 고장특성에 영향을 미치는 것이 TDDB 분석을 통해 확인 할 수 있다. 이와 같은 연구결과는 앞으로의STT-MRAM 연구 및 응용에 있어 많은 도움을 줄 것으로 기대된다. |Spin-transfer torque magnetoresistive random access memory (STT-MRAM) is an excellent candidate for next-generation memory because of its non-volatility, high-speed operation, endurance, high density, low power consumption, and compatibility with the standard complementary metal oxide semiconductor (CMOS) process. STT-MRAM consists of a thin tunnel oxide barrier between two ferromagnetic layers, called a magnetic tunnel junction (MTJ). The stored data in a MTJ depends on the magnetization vector of the ferromagnetic layers. The magnetization vector in one ferromagnetic layer is fixed (called pinned layer) and that in the other ferromagnetic layer is variable (called free layers). The magnetization vector between the free and pinned layers has a parallel or antiparallel state. The MTJ resistance of the anti-parallel state is larger than the resistance of the parallel state. The difference in conductance according to the magnetization vector is called the tunnel magneto resistance (TMR) ratio. Crystalline MgO has been considered the best tunnel oxide barrier for MTJ applications, because a MgO barrier with (001) growth enhance in the TMR effect over conventional amorphous alumina-based MTJs. However, because the MgO barrier is extremely thin, approximately 1 nm, a degradation of the barrier film by voltage stress is predicted to be a severe problem for reliability. Therefore, an understanding of the degradation mechanism is important to create reliable STT-MRAMs. Recently, much research has been performed to clarify the degradation and breakdown mechanism in thin MgO-based MTJs, and many experimental results related to MgO barrier degradation have been reported. The objective of this thesis is to study the Magnetic Tunnel Junction reliability and to analyze the degradation characteristics. Chapter 1 introduces the fundamental MRAM. In chapter 2, the thin film oxide reliability is portrayed and described in CMOS. In chapter 3, the main focus is on device fabrication methods and the various experimental analysis methods. Chapter 4 investigates the reliability modeling of MgO-based MTJs. By studying the degradation of MgO-based MTJs, an optimum Time Dependent Dielectric Breakdown (TDDB) model of MgO-based MTJs was proposed. In chapter 5, from a comparison of the TDDB slopes between the observed data and theoretical estimation, we found that the 1/E model’s consideration of a temperature increase due to self-heating agreed with our experimental TDDB results the 25~125 oC temperature range. On the other hand, the results of the TDDB according to a difference in state of the MgO interface using a Mg insertion layer are as follows: 1) the state of the MgO interface with Mg layer at anode is well matched 1/E Model, 2) the state of the MgO interface without Mg layer at anode is well matched power-law model. This means that the TDDB model should be differentiated from the state of interface on MgO-based MTJs. In the result of the temperature/interface state dependence on TDDB characteristics with MgO-based MTJs, the 1/E and Power-law model’s consideration of self-heating and direct tunneling were useful for understanding the breakdown of MgO. Chapter 6 investigates the effective screen method for failure MTJ. I compared the resistance change during a Constant Voltage Stress (CVS) test and confirmed a trap/detrap phenomenon during the Interval Voltage Stress (IVS) for different barrier thicknesses and temperatures. The resistance drift representing degradation and the time to breakdown (TBD), representing the breakdown characteristics, were better for thick barriers and lower temperatures than, for thin barriers and higher temperatures. The results suggest that the breakdown and degradation due to trap generation strongly depend on both the barrier thickness and the temperature. Furthermore, as the TBD varies at steady rates with changing barrier thickness, temperature, and electric field, I assume that MTJs with an abnormally thin layer of MgO can be screened effectively based on the predicted TBD. In chapter 7, the degradation characteristics of MgO-based MTJs were investigated based on: 1) the effect of MgO surface roughness with/without a Mg insertion layer, and 2) surface roughness of the buffer layer. The MTJ with Mg at the bottom electrode showed an extremely small resistance change for the CVS test and a reduced trap/detrap phenomenon for the IVS test under a negative bias condition. This is understood to be caused by less trapping of tunneling electrons at the MgO/Mg interface due to better crystallinity. MTJs with very rough surface buffer layers showed increased resistance drift and degraded characteristics. I suggest that this resulted from reduced MgO thickness on the MTJs with high surface roughness on the buffer layer, which was estimated by the TDDB and analytic approach. As this thesis suggests, the analysis of degradation characteristics of MgO-based MTJs under various conditions, can be applied to advanced research of, and applications using, STT-MRAM.; Spin-transfer torque magnetoresistive random access memory (STT-MRAM) is an excellent candidate for next-generation memory because of its non-volatility, high-speed operation, endurance, high density, low power consumption, and compatibility with the standard complementary metal oxide semiconductor (CMOS) process. STT-MRAM consists of a thin tunnel oxide barrier between two ferromagnetic layers, called a magnetic tunnel junction (MTJ). The stored data in a MTJ depends on the magnetization vector of the ferromagnetic layers. The magnetization vector in one ferromagnetic layer is fixed (called pinned layer) and that in the other ferromagnetic layer is variable (called free layers). The magnetization vector between the free and pinned layers has a parallel or antiparallel state. The MTJ resistance of the anti-parallel state is larger than the resistance of the parallel state. The difference in conductance according to the magnetization vector is called the tunnel magneto resistance (TMR) ratio. Crystalline MgO has been considered the best tunnel oxide barrier for MTJ applications, because a MgO barrier with (001) growth enhance in the TMR effect over conventional amorphous alumina-based MTJs. However, because the MgO barrier is extremely thin, approximately 1 nm, a degradation of the barrier film by voltage stress is predicted to be a severe problem for reliability. Therefore, an understanding of the degradation mechanism is important to create reliable STT-MRAMs. Recently, much research has been performed to clarify the degradation and breakdown mechanism in thin MgO-based MTJs, and many experimental results related to MgO barrier degradation have been reported. The objective of this thesis is to study the Magnetic Tunnel Junction reliability and to analyze the degradation characteristics. Chapter 1 introduces the fundamental MRAM. In chapter 2, the thin film oxide reliability is portrayed and described in CMOS. In chapter 3, the main focus is on device fabrication methods and the various experimental analysis methods. Chapter 4 investigates the reliability modeling of MgO-based MTJs. By studying the degradation of MgO-based MTJs, an optimum Time Dependent Dielectric Breakdown (TDDB) model of MgO-based MTJs was proposed. In chapter 5, from a comparison of the TDDB slopes between the observed data and theoretical estimation, we found that the 1/E model’s consideration of a temperature increase due to self-heating agreed with our experimental TDDB results the 25~125 oC temperature range. On the other hand, the results of the TDDB according to a difference in state of the MgO interface using a Mg insertion layer are as follows: 1) the state of the MgO interface with Mg layer at anode is well matched 1/E Model, 2) the state of the MgO interface without Mg layer at anode is well matched power-law model. This means that the TDDB model should be differentiated from the state of interface on MgO-based MTJs. In the result of the temperature/interface state dependence on TDDB characteristics with MgO-based MTJs, the 1/E and Power-law model’s consideration of self-heating and direct tunneling were useful for understanding the breakdown of MgO. Chapter 6 investigates the effective screen method for failure MTJ. I compared the resistance change during a Constant Voltage Stress (CVS) test and confirmed a trap/detrap phenomenon during the Interval Voltage Stress (IVS) for different barrier thicknesses and temperatures. The resistance drift representing degradation and the time to breakdown (TBD), representing the breakdown characteristics, were better for thick barriers and lower temperatures than, for thin barriers and higher temperatures. The results suggest that the breakdown and degradation due to trap generation strongly depend on both the barrier thickness and the temperature. Furthermore, as the TBD varies at steady rates with changing barrier thickness, temperature, and electric field, I assume that MTJs with an abnormally thin layer of MgO can be screened effectively based on the predicted TBD. In chapter 7, the degradation characteristics of MgO-based MTJs were investigated based on: 1) the effect of MgO surface roughness with/without a Mg insertion layer, and 2) surface roughness of the buffer layer. The MTJ with Mg at the bottom electrode showed an extremely small resistance change for the CVS test and a reduced trap/detrap phenomenon for the IVS test under a negative bias condition. This is understood to be caused by less trapping of tunneling electrons at the MgO/Mg interface due to better crystallinity. MTJs with very rough surface buffer layers showed increased resistance drift and degraded characteristics. I suggest that this resulted from reduced MgO thickness on the MTJs with high surface roughness on the buffer layer, which was estimated by the TDDB and analytic approach. As this thesis suggests, the analysis of degradation characteristics of MgO-based MTJs under various conditions, can be applied to advanced research of, and applications using, STT-MRAM.