ALD HfAlOx 저항 변화층을 이용한 신경세포 모방 소자의 가소성 특성 연구

Abstract

Neuromorphic devices which emulate information processing system of neural network with neuron and synapse in human brain, is getting new attention because of their self-learning ability and low power consumption. Especially, for use as synapse device in neuromorphic system, Memristor which is a kind of the neuromorphic devices, is considered and studied recently. Memristor has information transfer and storage abilities as synapse between neuron, replacing conventional DRAM (dynamic random access memory), SRAM (static random access memory). Furthermore, this device can has self-learning ability depending on its material selection and treatment. Many types of next-generation RAM (random access memory) such as FeRAM (ferromagnetic RAM), STT-mRAM (spin-transfer-torque magnetic RAM), PCRAM (phase-change RAM), ReRAM (resistive RAM) is researched for memristor device. Among these, ReRAM is the most attractive in terms of scalability, 3D integration, endurance, etc. The sensitivity of memristor synapse devices is changed when signal transferred repeatedly, this characteristic can implements the self-learning abilities to neuromorphic devices similar as neural network system. This characteristic is expressed as spike-time-dependence-plasticity (STDP) or synaptic plasticity. Synaptic characteristic of memristor is implemented by voltage and time tuning of pulsed signal based on resistance switching. The modulated memristor based on RRAM exhibits a characteristic that the current level gradually increases or decreases with iterative pulse voltage application, Through this, it is possible to express various signals according to the number of pulses. These behavior of current change can explain synaptic plasticity of memristors. Synaptic characteristic of memristor can be attributed to the gradual oxygen vacancy distribution in the switching layer. To induce this, a various kinds of switching layer/electrode selection or switching layer treatment are attempted. In this study, we used atomic layer deposition (ALD) based HfAlOx laminated dielectric switching layer and thin Al interlayer below top electrode to implement synaptic plasticity for memristor. HfAlOx can have excellent thermal stability and higher oxygen vacancy content through intermixing of Al2O3/HfO2 multiple layers, and Al interlayer can progress resistance switching.| 인간 뇌 신경망의 뉴런과 시냅스 정보전달 시스템을 모사하는 뉴로모픽 소자는, 자가학습능력과 낮은 에너지 소모율을 갖기 때문에 최근 들어 차세대 전자소자로 주목받고 있다. 특히, 뉴로모픽 시스템에서 시냅스 소자로써 사용되는 memristor가 최근 들어 주로 연구되고 있다. memristor는 뉴런 사이에서 정보를 전달하고 저장하는 시냅스와 같은 능력을 갖고 있으며, 이로 인해 기존에 사용되던 DRAM(dynamic random access memory), SRAM(static random access memory)과 같은 메모리 체계를 대체할 것으로 기대되고 있다. 뿐만 아니라 memristor는 소자 구성 물질이나 처리 방법에 따라 자가학습능력을 가질 수 있기 때문에 더욱 주목받고 있다. FeRAM (ferromagnetic RAM), STT-mRAM (spin-transfer-torque magnetic RAM), PCRAM (phase-change RAM), ReRAM (resistive RAM) 과 같은 많은 타입의 차세대 RAM (random access memory) 소자 들이 memristor로 이용되기 위하여 연구되고 있다. 그러나 이 중 ReRAM이 가장 미세화 가능성, 3차원 집적, 안정성 등에서 가장 효과적이므로 주로 연구되고 있다. memristor 시냅스 소자의 민감도는 전기 신호가 반복적으로 전달될수록 변화하게 되는데, 이러한 특성을 통해 인간 뇌 신경망에서와 같이 뉴로모픽 소자에 자가학습능력을 부여할 수 있다. 이러한 특성은 spike-time-dependence-plasticity (STDP) 혹은 synaptic plasticity라고 불린다. ReRAM 구조 기반 memristor 소자의 synaptic 특성은 resistance switching을 기반으로 펄스 신호의 전압과 시간 조율을 통해 구현될 수 있다. memristor 소자는 반복적인 펄스 전압 신호를 인가함에 따라 resistance switching을 통해 전류 레벨이 점진적으로 증가하거나 감소하게 되는데, 이에 따라 펄스 인가 횟수에 대한 다양한 신호 상태를 표현할 수 있다. 이러한 전류 변화 거동을 통해 memristor의 synaptic plasticity 특성을 설명하게 된다. memristor의 synaptic 특성은 switching layer 내 oxygen vacancy의 완만한 분포 상태에서 유도될 수 있다. 이 상태를 유발하기 위하여, 다양한 switching layer 및 electrode 물질 및 처리 방법이 시도되고 있다. 본 연구에서는, atomic layer deposition (ALD) 방식으로 증착된 HfAlOx laminated dielectric switching layer와 top electrode/switching layer 사이에 삽입된 얇은 Al interlayer를 사용하여 memristor의 synaptic plasticity를 구현하였다. HfAlOx는 Al2O3/HfO2 multiple layers를 통해 높은 열적 안정성과 oxygen vacancy 구성을 갖게 하고, Al interlayer는 resistance switching 기반 특성을 개선함으로써 synaptic 특성을 구현하는데 영향을 주었다.; Neuromorphic devices which emulate information processing system of neural network with neuron and synapse in human brain, is getting new attention because of their self-learning ability and low power consumption. Especially, for use as synapse device in neuromorphic system, Memristor which is a kind of the neuromorphic devices, is considered and studied recently. Memristor has information transfer and storage abilities as synapse between neuron, replacing conventional DRAM (dynamic random access memory), SRAM (static random access memory). Furthermore, this device can has self-learning ability depending on its material selection and treatment. Many types of next-generation RAM (random access memory) such as FeRAM (ferromagnetic RAM), STT-mRAM (spin-transfer-torque magnetic RAM), PCRAM (phase-change RAM), ReRAM (resistive RAM) is researched for memristor device. Among these, ReRAM is the most attractive in terms of scalability, 3D integration, endurance, etc. The sensitivity of memristor synapse devices is changed when signal transferred repeatedly, this characteristic can implements the self-learning abilities to neuromorphic devices similar as neural network system. This characteristic is expressed as spike-time-dependence-plasticity (STDP) or synaptic plasticity. Synaptic characteristic of memristor is implemented by voltage and time tuning of pulsed signal based on resistance switching. The modulated memristor based on RRAM exhibits a characteristic that the current level gradually increases or decreases with iterative pulse voltage application, Through this, it is possible to express various signals according to the number of pulses. These behavior of current change can explain synaptic plasticity of memristors. Synaptic characteristic of memristor can be attributed to the gradual oxygen vacancy distribution in the switching layer. To induce this, a various kinds of switching layer/electrode selection or switching layer treatment are attempted. In this study, we used atomic layer deposition (ALD) based HfAlOx laminated dielectric switching layer and thin Al interlayer below top electrode to implement synaptic plasticity for memristor. HfAlOx can have excellent thermal stability and higher oxygen vacancy content through intermixing of Al2O3/HfO2 multiple layers, and Al interlayer can progress resistance switching.