Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session G12: Emerging, Neuromorphic, and Computing Beyond Moore's LawRecordings Available
|
Hide Abstracts |
Sponsoring Units: FIAP Chair: Todd Brintlinger, United States Naval Research Laboratory Room: McCormick Place W-181C |
Tuesday, March 15, 2022 11:30AM - 11:42AM |
G12.00001: Structural study of active layer lithium niobium (III) oxide-LiNbO2 memristors Sara H Mohamed, Sebastian A Howard, Egor Evlyukhin, Galo J Paez Fajardo, Matthew J Wahila, Timothy M McCrone, William A Doolittle, Wei-Cheng Lee, Louis F Piper Memristors are a key element in developing a brain inspired neuromorphic computing system that offers a promising platform to overcome the von Neumann bottleneck. Recently, lithium niobite (LixNbO2) has been shown to be a promising candidate for the memristor, where changing the lithium content enables precise control of the resistive states. While previous work has demonstrated the correlation between the lithium concentration and the increase of metallic response in LixNbO2 via the density functional theory and x-ray photoelectron spectroscopy, a direct view of the structural variations from the delithiation and their effects on the resistivity switch remains unknown. Here, we used Raman spectroscopy to study the nature of LixNbO2 memristors by probing the structure as delithiation takes place. Raman spectra are measured for both pristine and chemically delithiated LixNbO2 crystals. We find two main peaks related to Li E2g and O A1g modes that can be used to track the local structural modifications related to the change in lithium concentration. The spectral signatures will provide essential insights on the Li- intercalation based switching mechanism in active layer LixNbO2 memristors. |
Tuesday, March 15, 2022 11:42AM - 11:54AM |
G12.00002: Deep physical neural networks using physics-aware training Tatsuhiro Onodera, Logan G Wright, Martin Stein, Tianyu Wang, Darren T Schachter, Zoey Hu, Peter L McMahon Deep neural networks have become ubiquitous in today’s data-driven world, but their energy requirements increasingly limit their scalability and broader use. Here, we propose the construction of deep physical neural networks that are made from layers of controllable physical systems, which can learn hierarchical representations of input data analogous to deep neural networks. To train these physical neural networks, we introduce a hybrid in situ-in silico algorithm physics-aware training. This training method has favorable scaling properties as it uses backpropagation, the de-facto training method for deep neural networks. To demonstrate the universality of our approach, we train diverse physical neural networks based on optics, mechanics, and electronics to experimentally perform audio and image classification tasks. Our approach broadens the possibility of using novel physical systems for deep learning and potentially enables them to perform machine learning faster and more energy-efficiently than conventional electronic processors. |
Tuesday, March 15, 2022 11:54AM - 12:06PM |
G12.00003: Structural, electronic, and polarization properties of YN and LaN Andrew Rowberg, Sai Mu, Michael W Swift, Chris G Van de Walle Semiconducting ScN has attracted significant scientific and technological interest; however, similar III-Nitrides such as YN and LaN are largely unexplored. To help fill this gap, we compare the structural and electronic properties of ScN, YN, and LaN. We find all three to have band gaps around 1 eV in their rocksalt phases, with only LaN having a direct band gap. ScN and YN both have higher-energy layered hexagonal structures; however, LaN prefers to adopt the polar wurtzite structure, which we show to be lower in energy than the rocksalt structure, and thus likely to be experimentally accessible. We calculate its polarization, along with its piezoelectric and ferroelectric properties. Wurtzite LaN has a smaller switching barrier than AlxSc1-xN alloys, lending it to potential use in ferroelectric devices. In addition, we show that wurtzite LaN and zincblende InP are close to lattice matched, and forming a heterostructure between them will result in a large bound charge of 1.3×1014 e/cm2, which could be used in tunnel junction devices. |
Tuesday, March 15, 2022 12:06PM - 12:18PM |
G12.00004: Mechanism of Thermally Activated Memristive Switching in Percolating Networks of 2D Semiconductor Nanosheets Vinod K Sangwan, Sonal V Rangnekar, Joohoon Kang, Jie Gu, Haihua Wang, Mark C Hersam Brain-inspired computing hardware is emerging as a promising paradigm to dramatically reduce power consumption compared to conventional digital computing limited by the von Neumann bottleneck. Nanoscale devices such as non-volatile memristors and volatile dynamical switches are being explored as underlying building blocks for neuromorphic circuits. Memristors based on two-dimensional (2D) materials enable tunable electrostatic coupling, novel bio-realistic functions, large-area, flexible, and printed neuromorphic circuits [1-4]. However, the memristive switching mechanisms in 2D nanosheet composites are poorly understood. Here, we present thermally activated memristive switching mechanisms in percolating networks of diverse solution-processed 2D semiconductors including MoS2, ReS2, WS2, and InSe [5]. The mechanism is elucidated by direct observation of channels using spatially resolved optical and chemical analyses (revealing oxygen-deficient filaments) and in situ thermal analysis (revealing local heating up to 115 °C). These devices show high switching ratios (up to 103) at low global electric fields (≈4 kV cm−1) that is explained by a thermally assisted electrical discharge that preferentially occurs at the sharp edges of 2D nanosheets. These results establish percolating networks of 2D nanosheets as an interesting materials system for high-order non-dynamical systems for neuromorphic circuits. |
Tuesday, March 15, 2022 12:18PM - 12:30PM |
G12.00005: Theory of Ge diffusion during the oxidation of Si/SiGe nano-fins Blair Tuttle, Mark Law, kevin Jones, Chappel Sharrock, Sokrates T Pantelides Recent experimental studies led to the detection of a novel, rapid diffusion of Ge down the sidewall of a Si/SiGe nano-fin structure during high-temperature oxidation. Specifically, the diffusion coefficient for Ge motion along the sidewall is up to five orders of magnitude larger than for Ge diffusion into bulk Si. This enhanced diffusion process can result in the formation of single-crystal Si nanowires embedded in a defect-free SiGe matrix. In this presentation, density-functional-theory calculations are combined with experimental observations to elucidate the atomic-level processes involved in the observed enhanced Ge motion. The calculated energy barriers for Ge motion along the oxidizing interface followed by a reintroduction into the crystalline Si, which results in the formation of SiGe, are consistent with the experimental values. |
Tuesday, March 15, 2022 12:30PM - 12:42PM |
G12.00006: Solving the Planar Potts Problem with Photonics-Inspired Clock Models Mostafa Honari Latifpour, Mohammad-Ali Miri Combinatorial optimization problems involve optimizing a cost function defined over all the combinations of a discrete set of objects. Typically, the size of the configuration space of such problems grows faster than polynomial scaling with the number of variables involved, which makes an exhaustive search for the optimal combination impractical. Thus, there is a great interest in developing methods for finding an approximation of the global optimal solution with polynomial scaling. Here, inspired by a recently proposed photonic Potts machine, we introduce a simplified nonlinear dynamical system that can faithfully model the planar Potts Hamiltonian. By employing two annealing mechanisms based on adiabatic parametric deformations of the governing Lyapunov function and harnessing chaotic dynamics, we can dynamically search for the ground state of the Potts Hamiltonian and find high-quality approximate solutions to the underlying NP-hard combinatorial problem. The simplicity and the inherently parallel nature of this system of coupled equations allow for its fast and efficient numerical integration on high-performance digital processors. |
Tuesday, March 15, 2022 12:42PM - 12:54PM |
G12.00007: Resonance for analog recurrent neural network Yurui Qu, Ming Zhou, Erfan Khoram, Nanfang Yu, Zongfu Yu Wave-based analog computing enjoys benefits of intrinsic parallelism and it can be extremely energy efficient compared to digital computing [1]. However, the transient nature of propagating waves makes it difficult to construct memory in the wave domain. Since memory is indispensable for computing in the temporal data, researchers have to resort to other means to realize the effect of memory such as optoelectronic conversion, routing through long waveguides and random internal feedback. In all these works, the memory is implicitly built into the complex structures, and physical intuition and interpretation are lacking. However, in the resonance system, we can include resonators with different lifetimes to realize both short-term and long-term memory. Here through a set of theoretical work, we propose resonance as a general form of memory to be used for complex temporal computing and advanced recurrent models such as LSTM. The findings here have broad impact and help to shape the future computing based on optical and acoustic waves. |
Tuesday, March 15, 2022 12:54PM - 1:06PM |
G12.00008: Valleytronics beyond valley flavor Feng-Wu Chen, Bing-Chen Huang, Gin-En Hong, Yen-Ju Lin, Yu-Shu Wu In 2D hexagonal materials, two types of atomic sites (A, B) are alternately located on vertices of hexagons, leading to intriguing topological aspect in association with the binary degree of freedom (DOF) known as valley flavor1-4. Here, a useful field quantity called local valley magnetic moment (LVMM) is introduced and operationally defined by the response of an electron to local magnetic field. The study of LVMM enables insight into the local magnetic control of valley DOF. Surprisingly, for some valley states LVMM shows an oscillation with sign variation, which demonstrates the necessity to include LVMM in addition to the global, valley flavor, in the study of valleytronics. As illustrated, LVMM reflects the local difference ρA – ρB (ρA(B) = local probability on site A (B)). Thus, LVMM and the sum ρA + ρB form a comparatively complete yet insightful, real-space set of key state variables, for viewing the general valley-involved physics. A study along this line is given in the case of carrier transport. |
Tuesday, March 15, 2022 1:06PM - 1:18PM |
G12.00009: Logic-in-memory based on an atomically thin semiconductor Guilherme Migliato Marega, Yanfei Zhao, Ahmet Avsar, Zhenyu Wang, Mukesh Tripathi, Aleksandra Radenovic, Andras Kis Non-von Neumann architectures are now emerging alternatives to traditional processors for specialized applications where energy efficiency becomes a critical parameter. In particular, brain-inspired architectures show promise in efficiently targeting reconfigurable hardware and matrix-vector computations. Although this is a broad area of research that explores co-designing new systems and devices, a solid material platform that suppresses all technology’s needs has not yet been found. In this regard, two-dimensional materials are a promising class of materials due to their excellent electrical and mechanical properties as well as the emergence of new phenomena that could be exploited to create new non-von Neumann architectures. Here we demonstrate the use of MOCVD grown monolayer MoS2 as the semiconductor channel for floating-gate memories. We exploit the fabricated memories as a programmable inverter taking advantage of its precise channel’s conductance control. Next, we demonstrate a programmable NOR gate and we further propose an architecture that can produce a complete set of operations. Our results open a new path towards efficient reconfigurable hardware. |
Tuesday, March 15, 2022 1:18PM - 1:30PM |
G12.00010: Optical bistability based on the integration of a molecular nanomaterial in silicon photonics Elena Pinilla-Cienfuegos, Alba Vicente, Jorge Parra, Juan Navarro-Arenas, Pablo Sanchis-Kilders, Roger Sanchis-Gual, Ramon Torres-Cavanillas, Mónica Giménez-Marqués, Eugenio Coronado Electro-optical bistability is a functionality which can be crucial for a wide range of applications as it can enable non-volatile and ultra-low power switching performance. We investigate the integration of a molecular-based nanomaterial presenting a Spin Crossover (SCO) effect in the silicon platform for enabling optical bistability. SCO is a spin-state switching phenomenon present in some molecular compounds such as the coordination complexes of transition-metal ions in which, under certain external stimulus (variation of temperature, pressure, electric field, or light irradiation), the electronic configuration can be switched between two molecular spin states, Low Spin (LS) and High Spin (HS) states. Furthermore, the spin state change is accompanied by a change in the structural, magnetic, and optical properties, as well as in the electrical conductivity and color. These properties vary as a function of the external stimulus following a hysteretic response, recognized as one of the most promising aspects of the system since hysteresis confers bistability and thus a memory effect. Finally, the SCO material can be synthetized as nanoparticles so that it can be easily integrated in the silicon platform and have the potential to allow optical switching at room temperature. |
Tuesday, March 15, 2022 1:30PM - 1:42PM |
G12.00011: Neuromorphic architecture based on orthogonal spin current injected MTJs Venkatesh Vadde, Abhishek Sharma, Bhaskaran Muralidharan Neuromorphic computing is inspired by the human brain and can typically solve certain computational problems with high power efficiency and lesser delay. Spintronics devices can potentially provide a better hardware platform for energy-efficient neuromorphic computing than CMOS counterparts. In this work, we propose a circuit based on orthogonal spin current injected magnetic tunnel junctions (MTJs) to simultaneously perform various functions of convolutional neural network (CNN). We have developed a computational platform that incorporates HSPICE and the non-equilibrium Green's function (NEGF) approach to evaluate the performance of the proposed circuit. We show linearized switching of the MTJ using the orthogonally injected spin current. Using the linear switching region of the MTJ, we show the proposed circuit performs the simultaneous CNN functions such as rectified linear unit (ReLU) and the local max-pooling functions. Our simulations also demonstrate the robustness of the proposed circuit against thermal noise. |
Tuesday, March 15, 2022 1:42PM - 1:54PM |
G12.00012: All-Photonic Artificial Neural Network Processor Via Non-linear Optics Jasvith R Basani, Mikkel Heuck, Stefan Krastanov, Dirk Englund Integrated photonic architectures have been shown to accelerate conventional computing tasks that have been deemed as bottlenecks in traditional electronics. Among these tasks, one that is fundamental to neural networks is matrix processing. We propose the use of an all-optical integrated chip to accelerate deep learning. The architecture we introduce encodes information in the complex amplitudes of frequency modes that act as neurons. Information processing by intermodulating the neurons is implemented by the nonlinear optical process of Four-Wave Mixing (FWM). The FWM among neurons and controlled pump modes allows us to implement the linear transformation within microring resonators. Our nonlinear activation function relies on pulse distortion via nonlinear interactions, followed by controlled capture of these pulses into microring resonators. The proposed design is novel in its ability to attain arbitrarily large computational speeds by increasing the power of the pump modes, within limits imposed by heating due to losses. Additionally, it provides a completely unitary weight matrix, thus opening up the prospects of reversible computing. Through simulations, we show that our design achieves performances commensurate with present-day techniques on classification benchmarks. |
Tuesday, March 15, 2022 1:54PM - 2:06PM |
G12.00013: Design of Self-Learning Machines using Hamiltonian Echo Backpropagation Victor Lopez Pastor, Florian Marquardt A self-learning machine can be defined as a physical system that can be trained on data (similar to artificial neural networks), but where the update of the internal degrees of freedom that serve as learnable parameters happens autonomously. We introduce Hamiltonian Echo Backpropagation, a general scheme to use any time-reversible Hamiltonian system as a self-learning machine. A particularly appealing feature of our scheme is that it does not require any knowledge of the internal dynamics of the Hamiltonian system, which can be operated as a black box. We show how this scheme can be applied to some promising physical platforms. |
Tuesday, March 15, 2022 2:06PM - 2:18PM |
G12.00014: Photonic kernel machines for ultrafast spectral analysis Zakari Denis, Ivan Favero, Cristiano Ciuti We present photonic kernel machines, a machine learning-inspired scheme for ultrafast spectral analysis of noisy radio-frequency signals from single-shot optical intensity measurements that harnesses fast photonic hardware to reach throughput rates ideally well above the gigahertz. We first theoretically describe some of their key underlying principles and then numerically illustrate their performance on a photonic lattice-based implementation. We apply this model both to picosecond pulsed signals, on an energy-spectral-density estimation and a shape classification tasks, and to continuous signals, on a frequency tracking task. The presented optical-computing scheme proves robust to noise while requiring minimal control on the photonic-lattice parameters, thus making it readily implementable in realistic state-of-the-art photonic platforms. |
Tuesday, March 15, 2022 2:18PM - 2:30PM |
G12.00015: Interface Modulated Pd/Nb-SrTiO3 Resistive Switching Via Hydrogen Intercalation Srinivas Vanka, Hyungki Shin, Ryan L Roemer, Bruce A Davidson, Ke Zou Resistive random-access memory (RRAM) or memristor devices are one of the most promising technologies to achieve low-cost, high endurance, low power consumption, and CMOS compatible neuromorphic hardware to address “memory wall” problem. Strontium titanate SrTiO3 (STO) is an ideal candidate, due to large bandgap, to be used as an insulator in the simple metal/insulator/metal (MIM) architecture for resistive switching. By modulating the Schottky barrier height through applied bias, it is possible to tune the resistance states of MIM devices. Thus, interface kinetics play a crucial role in determining the resistive switching mechanism for memristor devices consisting of a Schottky junction. In this work, we examine the interfacial properties of Pd/Nb-STO resistive devices with and without hydrogen intercalation. Hydrogen molecules get adsorbed, dissociate to hydrogen atoms, and split to protons at Pd/Nb-STO interface. The I-V hysteresis characteristics reveal that the intercalated protons reduce the barrier height by 100 mV compared to sample without hydrogen annealing and decrease the resistance under reverse bias condition. The decrease in the resistance (under reverse bias condition) for annealed sample is attributed to an decrease in built-in potential (from C-V measurements). These devices showed high retention time, large off/on ratio, and fast switching characteristics. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700