Optimizing Reservoir Separability in Liquid State Machines for Spatio-Temporal Classification in Neuromorphic Hardware

Journal of Low Power Electronics and Applications, Год журнала: 2025, Номер 15(1), С. 4 - 4

Опубликована: Янв. 24, 2025

In this paper, we propose an optimization approach using Particle Swarm Optimization (PSO) to enhance reservoir separability in Liquid State Machines (LSMs) for spatio-temporal classification neuromorphic systems. By leveraging PSO, our method fine-tunes parameters, neuron dynamics, and connectivity patterns, maximizing while aligning with the resource constraints typical of hardware. This was validated both software (NEST) on hardware (SpiNNaker), demonstrating notable results terms accuracy low energy consumption when SpiNNaker. Specifically, addresses two problems: Frequency Recognition (FR) five classes Pattern (PR) four, eight, twelve classes. For instance, Mono-objective running NEST, accuracies ranged from 81.09% 95.52% across benchmarks under study. The Multi-objective outperformed approach, delivering ranging 90.23% 98.77%, its superior scalability LSM implementations. On SpiNNaker platform, mono-objective achieved 86.20% 97.70% same benchmarks, further improving accuracies, 94.42% 99.52%. These show that, addition slight improvements, hardware-based implementations offer efficiency a lower execution time. example, operates at around 1–5 watts per chip, traditional systems can require 50–100 similar tasks, highlighting significant savings underscore effectiveness PSO-optimized LSMs resource-limited platforms, showcasing improved performance advantages energy-efficient processing.

Язык: Английский

---

Neuromorphic Hardware for Artificial Sensory Systems: A Review

Journal of Electronic Materials, Год журнала: 2025, Номер unknown

Опубликована: Фев. 10, 2025

Abstract Senses are crucial for an organism’s survival, and there have been numerous efforts to artificially replicate sensory perception elicit desired responses specific stimuli. Recent research is increasingly focused on developing artificial nervous systems based the unsupervised learning capabilities of neural networks (ANNs) using unstructured data. However, future ANNs, which require precise sensing in complex environments, must be capable processing a large number signals real time, ideally from continuous domains. This need massive data driving evolution hardware systems, leading development devices specifically designed (ASSs) at level. To address this challenge, sensor not only detect target substances but also enable computational functions by utilizing their inherent material properties. Research neuromorphic sensors advancing towards integration with next-generation effectively addressing scenarios we aim identify. review offers perspectives human-like computing these challenges. It examines progress implementing five representative senses device level, explores methods integrating them into ASS, provides comprehensive overview potential applications. In particular, emphasize approaches cognitively utilize discussed as neurons synapses, enabling inputs. We offer nerve future.

Язык: Английский

---

Piezo-strain induced nonvolatile control of magnetic skyrmion nonlinear dynamics for artificial synapse device applications

Applied Physics Letters, Год журнала: 2025, Номер 126(8)

Опубликована: Фев. 24, 2025

High-performance artificial synaptic devices that emulate the functions of biological synapses are crucial for advancing energy-efficient brain-inspired computing systems. Current studies predominantly focus on memristive devices, which achieve through nonvolatile electric current-assisted carrier modulation. However, these methods often suffer from excessive energy consumption. Here, a type low-energy-consumption synapse based strain-mediated electric-field control magnetic skyrmion's radius is demonstrated, where consumption 10 fJ per state and non-volatility achieved by local ferroelectric domain switching under bipolar fields. The proposed skyrmion-based device can replicate essential behaviors, including long-term potentiation (LTP), depression (LTD), paired-pulse facilitation, depression, spiking-time-dependent plasticity, aligning it closely with system. weight change non-linearity emulated modulating precisely engineering applied pulses. Simulation using Modified National Institute Standards Technology database reveals pattern recognition rate decreases exponentially increasing LTP/LTD non-linearity, quantifying effect rate. This work underscores potential single as groundbreaking approach developing high density low-energy devices.

Язык: Английский

---

Field‐Induced Phase Transitions in Cuprate Superconductors for Cryogenic in‐Memory Computing

Small, Год журнала: 2025, Номер unknown

Опубликована: Март 3, 2025

Abstract Energy‐efficient cryogenic memory systems play a critical role in wide spectrum of applications focused on ultra‐energy‐efficient information and communication technologies, such as quantum computing or superconducting electronics. Neuromorphic systems, known for their superior energy efficiency, have emerged promising approach in‐memory computing. Specifically, strongly correlated oxides that exhibit Mott metal‐insulator transitions through field‐induced oxygen movement are great interest analog neuromorphic Yet, optimizing performance at low temperatures may prove challenging due to reliance ionic motion. In this study, structures composed YBa 2 Cu 3 O 7 − x (YBCO) combined with ferromagnetic (LSMO) investigated obtain non‐volatile multilevel memristive switching effects high temperatures. This research reveals the presence two competing mechanisms, which attributed vacancies electric carriers within these structures. It is determined phase transition induced by holes primary factor influencing dynamics Additionally, physics‐based compact model proposed accurately replicates experimental findings provides tool circuit‐level design.

Язык: Английский

---

Editorial: The intersection of machine learning and physical sciences: Insights from the 2024 Nobel Prizes

APL Machine Learning, Год журнала: 2025, Номер 3(1)

Опубликована: Март 1, 2025

Язык: Английский

---

RRAMulator: An efficient FPGA-based emulator for RRAM crossbar with device variability and energy consumption evaluation

Microelectronics Reliability, Год журнала: 2025, Номер 168, С. 115630 - 115630

Опубликована: Март 12, 2025

Язык: Английский

---

Introduction to neuromorphic functions of memristors: The inductive nature of synapse potentiation

Journal of Applied Physics, Год журнала: 2025, Номер 137(11)

Опубликована: Март 17, 2025

Memristors are key elements for building synapses and neurons in advanced neuromorphic computation. made with a wide range of material technologies, but they share some basic functionalities to reproduce biological functions such as synapse plasticity dynamic information processing. Here, we explain the memristors, show that main memristor can be obtained combination ordinary two-contact circuit elements: inductors, capacitors, resistors, rectifiers. The measured IV characteristics yield clockwise counterclockwise loops, which like those from memristors. inductor is responsible set resistive switching, while capacitor produces reset cycle. By combining inductive capacitive properties gating variables represented by diodes, construct full potentiation depression responses against applied trains voltage pulses different polarities. These results facilitate identifying central dynamical characteristic required investigation synaptic

Язык: Английский

---

Neuromorphic algorithms for brain implants: a review

Frontiers in Neuroscience, Год журнала: 2025, Номер 19

Опубликована: Апрель 11, 2025

Neuromorphic computing technologies are about to change modern computing, yet most work thus far has emphasized hardware development. This review focuses on the latest progress in algorithmic advances specifically for potential use brain implants. We discuss current algorithms and emerging neurocomputational models that, when implemented neuromorphic hardware, could match or surpass traditional methods efficiency. Our aim is inspire creation deployment of that not only enhance computational performance implants but also serve broader fields like medical diagnostics robotics inspiring next generations neural

Язык: Английский

---

The growing memristor industry

Nature, Год журнала: 2025, Номер 640(8059), С. 613 - 622

Опубликована: Апрель 16, 2025

Язык: Английский

---

Chaotic recurrent neural networks for brain modelling: A review

Neural Networks, Год журнала: 2024, Номер 184, С. 107079 - 107079

Опубликована: Дек. 27, 2024

Язык: Английский

---