Apr 26, 2024
9:00am - 9:30am
Room 343, Level 3, Summit
Marshall Tellekamp1,Olivia Schneble1,2,Christopher Muzzillo1,Mirzo Mirzokarimov1,Michelle Smeaton1,Jeramy Zimmerman2
National Renewable Energy Laboratory1,Colorado School of Mines2
Marshall Tellekamp1,Olivia Schneble1,2,Christopher Muzzillo1,Mirzo Mirzokarimov1,Michelle Smeaton1,Jeramy Zimmerman2
National Renewable Energy Laboratory1,Colorado School of Mines2
As global information consumption continues to grow, the energy required to support computations is consuming a significant portion of the global energy supply. Data centers alone will consume 4.5% of all energy by 2025,<sup>1</sup> with computing demands doubling every 3–4 years.<sup>2</sup> Neuromorphic computing promises an energy-efficient solution based on the naturally efficient mechanisms that drive human thought and memory; it was recently calculated to outperform other technologies, including quantum computing, by at least 10,000x in terms of energy per solution.<sup>3</sup> Spiking neural networks (SNNs) are extremely efficient, encoding information based on the time between voltage spikes defined by multiple <i>local</i> inputs. The dynamic nature of the spiking processes can be simulated in conventional semiconductors at the cost of energy efficiency,<sup>4,5</sup> however a more attractive solution utilizes materials that natively produce such dynamic behavior.<br/>Materials with an insulator-metal transition (IMT) are ideal for accessing nonlinear transport properties that could be useful in SNNs. In addition, IMTs can be highly sensitive to structure, defects, and environmental factors. Coupling fundamental mechanisms to stimuli such as electric field, electrochemical gating, and optical stimuli is an attractive way of encoding adaptive behavior that is a function of multiple input variables. Recently, IMT materials such as VO<sub>2</sub> and NbO<sub>2</sub> have been used to replicate the Hodgkin-Huxley action potential in pull-up/pull-down neuristor circuitry.<sup>6–8</sup> The IMT of NbO<sub>2</sub> is at 1080 K, too high for energy-efficient use, while the IMT of VO<sub>2</sub> is at 330 K (with doping up to 370 K)<sup>9</sup>, meaning active cooling is required for operation with silicon CMOS, which typically operate around 400 K. Meanwhile, RNOs have a tunable IMT from 100–600 K, with known sensitivity to many inputs.<sup>10</sup> Therefore, it is attractive to improve our understanding of RNO synthesis, control the IMT, and leverage it for neuromorphic computing applications.<br/>In this invited talk, I will overview our efforts to controllably synthesize RNO compounds (R = Gd, Eu, Sm, Nd, La) and leverage them for neuromorphic computing applications. I will first discuss synthesis of high-quality heteroepitaxial nickelate layers and bilayer stacks by RF magnetron sputtering, including successful stabilization of EuNiO<sub>3</sub> by RF sputtering which we have not seen reported elsewhere. I will also focus on the connection between uncontrollable synthetic parameters such as target aging and resulting film properties. In addition to RF sputtering, I will discuss synthesis of heteroepitaxial layers and bilayer stacks by pulsed laser deposition (PLD) while connecting deposition parameters to measured properties. In particular, I will present EuNiO<sub>3</sub>/LaNiO<sub>3</sub> bilayers with a residual resistivity ratio ~ 7 orders of magnitude.<br/>Finally, I will conclude with a discussion of our efforts to fabricate devices using nickelate bilayers, including a discussion of challenges and possible approaches to creating vertical devices for scalability. I will discuss our recent results of electrically-driven IMTs in NdNiO<sub>3</sub> and NdNiO<sub>3</sub>/LaNiO<sub>3</sub> bilayers which show strong negative differential resistance (NDR) – the key electrical characteristic leveraged in neuristor circuits. <br/> <br/>1. Liu et al. <i>Global Energy Interconnection</i> <b>3</b>, 272–282 (2020).<br/>2. Masanet et al. <i>Nat Electron</i> <b>3</b>, 409–418 (2020).<br/>4. Merolla et al. <i>Science</i> <b>345</b>, 668–673 (2014).<br/>5. Merolla et al. <i>2011 IEEE Custom Integrated Circuits Conference (CICC)</i> 1–4 (2011).<br/>6. Lin et al. <i>Nat Electron</i> <b>3</b>, 225–232 (2020).<br/>7. Pickett et al. <i>Nature Materials</i> <b>12</b>, 114–117 (2013).<br/>8. Kumar et al. <i>Nature</i> <b>585</b>, 518–523 (2020).<br/>9. Schofield et al. <i>Chemical Communications</i> <b>58</b>, 6586–6589 (2022).<br/>10. Catalano et al. <i>Rep. Prog. Phys.</i> <b>81</b>, 046501 (2018).