Editorial: Focus issue on topological solitons for neuromorphic systems

Topological solitons are particle-like textures that arise in ordered systems, leading to novel physical phenomena and functional properties. Recently, topological solitons have emerged as promising nanoscale objects that have the potential to revolutionize Unconventional Computing. Because of their unique inherent properties, such solitons perfectly match the requirements of innovative computing paradigms beyond traditional von Neumann approaches. These properties include scalability, stability, stochastic dynamics, nonlinear responses, short-term memory

Topological solitons are particle-like textures that arise in ordered systems, leading to novel physical phenomena and functional properties.Recently, topological solitons have emerged as promising nanoscale objects that have the potential to revolutionize Unconventional Computing.Because of their unique inherent properties, such solitons perfectly match the requirements of innovative computing paradigms beyond traditional von Neumann approaches.These properties include scalability, stability, stochastic dynamics, nonlinear responses, short-term memory, and more.In magnetic and electric materials, solitons manifest themselves as localized, particle-like excitations that cannot be removed by a continuous local deformation from the ordered spin texture or the arrangement of electric dipoles, respectively.Intriguing examples are magnetic skyrmions and ferroelectric domain walls, which have already been studied intensively as information carriers for next-generation spintronics and nanoelectronics applications.Operating in frequency ranges from MHz to THz, their dynamics exhibit low input-driven behavior and remain resilient to external perturbations and temperature variations.Furthermore, their innate properties facilitate the seamless integration of many Unconventional Computing concepts, catalyzing a burgeoning field of novel neuromorphic devices and breakthrough methodologies.
This special issue introduces pioneering concepts around topological solitons in solid state systems and provides a broad perspective on their diverse applications.It delves into various intriguing manifestations of topological solitons, including domain walls, skyrmions, and biskyrmions, as well as their corresponding lattice structures.Of particular interest are the inherent memristive properties of such solitons, which offer great potential for unconventional computational paradigms as they can readily be modified by electrical means.In this issue, it is shown how different advances in the field can overcome the limitations of current technology, highlighting both the challenges and opportunities that lie ahead.
Ribeiro de Assis et al present a groundbreaking biskyrmion-based artificial neuron [1], which allows for emulating the essential properties of biological neurons, such as leakage, integration, firing, and refractory periods.This innovation is a cornerstone in the development of spiking neural networks, aiming to replicate the efficiency of the human brain by encoding information within temporal spike intervals.While software models are promising, they lack scalability and energy efficiency.In contrast, hardware-based approaches offer improved scalability, energy conservation, and spatial efficiency, although it remains challenging to replicate behaviors such as leaky-integrate-and-fire and refractoriness.This article demonstrates that the inherent properties of biskyrmions, which can be converted to skyrmions, successfully display these behaviors at an energy cost of about 1 pJ per spike, operating in the GHz range.Notably, the design of this device eliminates the need for complex nanostructuring, simplifying fabrication compared to previous methods.
The intrinsic dynamics of domain walls, which are characterized by low input requirements and particle-like behavior, are exploited by Yadav et al for memristive applications [2].Their approach enables reliable hardware-based weight adjustment in artificial neural networks.The authors exploit domain wall pinning at the thin film interface to improve the predictability and reliability of current-driven domain wall motion.This enhancement ensures the achievement of high linearity and symmetry in the device's synaptic weight updating process, which is critical for achieving long-term potentiation and long-term depression.These features are essential for efficient on-chip learning in neural networks.In addition, the authors provide an estimation of the power consumption in these synaptic devices and project their scalability with respect to on-chip learning within corresponding crossbar arrays.
Neumayer et al exploit the role of domain walls in the electric polarization switching to demonstrate a nanoscale tunable and polarization dependent behavior of the ferroelectric Sn 2 P 2 S 6 at room temperature [3].Their experimental results show that the initial polarization orientation has a significant impact on the ferroelectric switching properties and is strongly influenced by the nucleation, growth and oscillation of domain walls.In particular, they emphasize how this mechanism facilitates efficient capacitive switching, a step toward achieving the desired regime of memcapacitance.
A review of recent studies that anticipate and elucidate fundamental properties essential for Neuromorphic Computing in three prototypical dipolar materials is given by Prosandeev et al [4].The authors highlight the remarkable ability of ferroelectrics to manifest diverse and exotic polar configurations-from striped dipole patterns to vortices, bubbles, and skyrmions-under different electrical and mechanical boundary conditions.Using effective Hamiltonian methods, they shed light on relaxor ferroelectrics and antiferroelectric systems, revealing phenomena such as action potentials, integration, and the existence of multiple states with different polarization responses when subjected to THz electrical pulses.These distinctive properties pave the way for a comprehensive and innovative toolkit that is poised to revolutionize the design of polar-based neuromorphic systems.
Hu et al provide an overarching review of advances in computing systems based on magnetic topological solitons, such as magnetic domain walls and skyrmions, and highlight their importance in logical and neuromorphic computing [5].The authors carefully describe the innate properties of these solitons-non-volatility, scalability, rich physical interactions, and capacity for nonlinear behavior-and demonstrate their potential use in various concepts aimed at replicating the distinctive features of Unconventional Computing paradigms.Through a comprehensive review, they elucidate characteristic properties, spotlight most recent proposals, address challenges, and underscore how these approaches overcome existing technological barriers.
Razumnaya et al propose a ferroelectric logic device consisting of multiple nanodots sandwiched between two electrodes and encased in dielectric material [6], underscoring the central role of topological switching.They emphasize the importance of topology in delineating different pathways for polarization configuration switching within ferroelectric nanodots, revealing multi-branched hysteresis loops that enable specific interconnectivity patterns.In addition, the authors highlight the potential to exploit the multi-state properties of the proposed device alongside the negative capacitance effect observed in a similar charge-controlled architecture of ferroelectric capacitors.This strategy promises to significantly reduce the power consumption during multivalue logic operations.The developed multilevel ferroelectric devices pave the way for a transformative implementation-a topologically driven realization of discrete synaptic states within neuromorphic computing paradigms.
An insightful review that explores the realm of neuromorphic capabilities inherent in ferroelectric domain walls is provided by Sharma and Seidel [7].The review offers a forward-looking perspective on upcoming developments and applications in energy-efficient, adaptive, brain-inspired electronics and computing.The authors describe the intricate properties and dynamics of ferroelectric domain walls, with a particular emphasis on their application in memristors.They highlight the remarkable potential of ferroelectric domain walls to mimic synapses, enabling the hardware realization of neural networks.In addition, the authors illustrate how this technology can overcome existing hurdles in the field, address the resulting challenges, and present opportunities that are poised to shape future developments.
Finally, a compelling exploration of helical, spiral, stripy phase, or other one-dimensional soliton lattices is proposed by Bechler and Masell [8], introducing the concept of 'helitronics' with potential implications for Unconventional Computing.These states, prevalent in magnets, naturally exist as stray-field-free ground states, giving them high stability and minimal power requirements.Using micromagnetic simulations, the authors demonstrate the feasibility of all-electric binary and non-binary helitronic memory cells.By exploiting the inherent properties and current-driven dynamics, they also demonstrate the implementation of artificial synapses, revealing the promising prospects of using helitronics for Unconventional Computing paradigms.