Tailoring transient-amorphous states: towards fast and power-efficient phase-change memory and neuromorphic computing.
Tailoring transient-amorphous states: towards fast and power-efficient phase-change memory and neuromorphic computing.
复制标题
DOI：
10.1002/adma.201402696
复制
发表时间：
2014-11-26
期刊：
影响因子：
--
通讯作者：
Elliott SR
中科院分区：
文献类型：
--
作者：
Lee TH;Loke D;Huang KJ;Wang WJ;Elliott SR
DOI：10.1002/adma. 201402696至关重要，特别是对于PCM技术的神经形态计算应用，例如电子突触[4，5]或生物启发的算术计算设备[6]，其中许多电脉冲以复杂的方式用于诱导相变。然而，很少有系统的研究，这种相互作用的物理基础尚未报道。我们在这里报告的新方法，以克服这些限制，通过剪裁瞬态非晶（TA）状态的PC材料。以良好编程的方式施加多个电激励脉冲不仅能够开发快速、低功率和高效（并行写入）形式的PCM，而且还提供了甚至实现类似生物学的神经形态功能的机会，这可以受益于TA状态的动态控制。我们首先证明了TA-状态的时间演化的刺激脉冲的应用程序。采用了三种不同的刺激脉冲，每个刺激脉冲具有不同的幅度（V St）或长度（d St），同时允许低电压/高电压（LO-HI）刺激和ACT脉冲之间的无偏时间段（d TS）。每个刺激脉冲代表图1a中电压宽度（VW）图中的不同区域，并且对TA状态产生不同影响，如图1 B所示。在任何情况下，由刺激脉冲产生的TA相总是显示出比熔融淬火非晶（a-）状态（60 ns;参见图S2）更短的最小ACT长度（完全结晶的最小ACT脉冲长度），这意味着暴露于刺激脉冲总是引起更快的结晶（图1c）。此外，刺激脉冲越强（和越宽），最小ACT长度变得越短。更有趣的是，当两个LO-HI脉冲之间存在一个无偏周期时，所有的TA态都表现出自发跃迁TA-状态可以由形成成核动力学理论基础的簇大小分布表示，并且刺激脉冲时ACT时间的缩短可以通过它们的演变来描述。在微观尺度上，TA态可以通过它们在无序网络结构中的中程有序度来描述，该无序网络结构在激发时局部地和暂时地波动。基于成核动力学理论的数值计算[14-16]表明，响应于温度变化，熔融淬火的非晶相nam的团簇数量在有限的时间内在升高的温度下逐渐演变为稳态分布nss。[17]这种时间依赖性是生长过程中热激活过程的结果。
DOI: 10.1002/adma. 201402696 crucial, especially for neuromorphic-computing applications of PCM technology, such as electronic synapses,[4, 5] or bio-inspired arithmetic-computing devices,[6] in which many electric pulses are used in a complicated way to induce phase transitions. However, few systematic researches regarding the physics underlying this interaction have been reported yet. We report here new methodologies to overcome these limitations by tailoring transient amorphous (TA) states of PC materials. Applying multiple electrical-excitation pulses in a well-programmed manner not only enables the development of a fast, low-power and efficient (parallel-writing) form of PCM, but also provides an opportunity even for achieving biologylike neuromorphic functionalities, which can benefit from a dynamic control of TA-states. We first demonstrate the temporal evolution of TA-states upon the application of a stimulus pulse. Three different stimulus pulses, each having a different amplitude (V St) or length (d St), were employed, while allowing an unbiased time period (d TS) between the low-voltage/high-voltage (LO-HI) stimulus and ACT pulses. Each stimulus pulse represents a different region in the voltage-width (VW) diagram in Figure 1a, and causes a different influence on the TA states, as shown in Figure 1 b. In any case, the TA-phases generated from the stimulus pulse always show shorter minimum ACT lengths (the minimum ACT pulse length for full crystallization) than that for the melt-quenched amorphous (a-) state (60 ns; see Figure S2), which means that exposure to the stimulus pulse always gives rise to faster crystallisation (Figure 1c). In addition, the stronger (and wider) is the stimulus pulse, the shorter the minimum ACT length becomes. A more interesting observation is that, when there exists an unbiased period between two LO-HI pulses, all the TA-states show a spontaneous transition (or relaxation) to other TA-states, each of which presents, progressively, a longer minimum ACT length, up to a characteristic time (∼ 1 µs).The TA-states can be represented by their cluster-size distribution that form the basis of kinetic theory of nucleation, and the shortening of the ACT time upon a stimulus pulse may be described by their evolution. On the microscopic scale, the TA-states may be described by their degree of medium-range order in the disordered-network structure that fluctuates locally and temporally upon excitations. Numerical computations [14–16] based on the kinetic theory of nucleation have shown that, in response to a temperature change, the cluster population of the melt-quenched amorphous phase nam evolves gradually to the steady-state distribution nss at an elevated temperature in a finite amount of time.[17] This time dependence is a consequence of the thermally-activated process during the growth of