Neuromorphic Circuits with Spiking Astrocytes for Increased Energy Efficiency, Fault Tolerance, and Memory Capacitance

Aybars Yunusoglu, Dexter Le, Murat Isik, I. Can Dikmen, Teoman Karadag · 2025

Read Neuromorphic Circuits with Spiking Astrocytes for Increased Energy Efficiency, Fault Tolerance, and Memory Capacitance by Aybars Yunusoglu, Dexter Le, Murat Isik, I. Can Dikmen, Teoman Karadag (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-20492.txt.

Introduction

Brain-inspired chips wire up artificial neurons, but the brain's support cells, astrocytes, are usually left out. Aybars Yunusoglu and co-authors added spiking astrocyte circuits to neuromorphic networks in 2025, chasing efficiency, fault tolerance, and memory capacity.

Each astrocyte watches over a cluster of neurons and helps repair them, so the cluster survives damage that would cripple a plain network. Their design reached a fault tolerance rate above eighty-one percent, with resilience improved by about nineteen points over rivals. A routing method maps the design onto many-core hardware. The authors argue astrocytes are the missing ingredient for neuromorphic systems that must keep working under adverse conditions.

Catalog record

CREATORAybars Yunusoglu, Dexter Le, Murat Isik, I. Can Dikmen, Teoman Karadag
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-20492

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Neuromorphic Circuits with Spiking Astrocytes for Increased Energy Efficiency, Fault Tolerance, and Memory Capacitance Aybars Yunusoglu, Dexter Le, Murat Isik, I. Can Dikmen, Teoman Karadag Abstract In the rapidly advancing field of neuromorphic computing, integrating biologically-inspired models like the Leaky Integrate-and-Fire Astrocyte (LIFA) into spiking neural networks (SNNs) enhances system robustness and performance. This paper introduces the LIFA model in SNNs, addressing energy efficiency, memory management, routing mechanisms, and fault tolerance. Our core architecture consists of neurons, synapses, and astrocyte circuits, with each astrocyte supporting multiple neurons for self-repair. This clustered model improves fault tolerance and operational efficiency, especially under adverse conditions. We developed a routing methodology to map the LIFA model onto a fault-tolerant, many-core design, optimizing network functionality and efficiency. Our model features a fault tolerance rate of 81.10% and a resilience improvement rate of 18.90%, significantly surpassing other implementations. The results validate our approach in memory management, highlighting its potential as a robust solution for advanced neuromorphic computing applications. The integration of astrocytes represents a significant advancement, setting the stage for more resilient and adaptable neuromorphic systems. I Introduction The field of neuromorphic computing is undergoing a transformative phase, driven by the integration of biologically-inspired components. This paper introduces a groundbreaking approach that integrates the Leaky Integrate-and-Fire Astrocyte (LIFA) model into Spiking Neural Networks (SNNs). SNNs are inspired by brain dynamics known for their capabilities in energy-efficient procesing and biologically plausible learning algorithms [ 1 , 2 , 3 , 4 ] . However, SNNs remain vulnerable to faults that can impair their efficiency. Astrocytes play a critical role in regulating neuronal activity and synaptic transmission; contributing to the resilience of the biological networks [ 5 , 6 , 7 , 8 ] . This integration of astrocytic mechanisms into SNNs offers dynamic adjustments for fault tolerance [ 9 , 10 , 11 ] . Despite performance improvements with SNNs, technology scaling exhibits challenges such as increased power densities along with faults in neuron and synapse circuits. These challenges impact the SNN model’s performance capabilities and overall robustness [ 12 , 13 , 14 ] . The LIFA model is rooted in dynamic interplay between neurons and astrocytes. As a result, the LIFA model enhances neural processing by bolstering computational strength and efficiency [ 15 ] . Additionally, LIFA model integration introduces novel computational capabilities for neuromorphic computing [ 16 , 17 , 18 , 19 ] . In order to achieve energy efficiency, memory measurement, efficient routing, and fault tolerance, we designed a methodology around four key pillars. LIFA enables accurate emulation of brain functions and marks a breakthrough in neuromorphic computing by shifting beyond neuron-centric approaches. We propose four components of a fault-tolerant neuromorphic computing system: • Alternative Reduce Regime: Optimizes energy consumption through single neuron component switching, demonstrating a commitment to sustainable and efficient computing. • Memory Measurement and Management: Explores synaptic dynamics inspired by Hopfield networks to optimize memory storage and retrieval, achieving superior memory efficiency with fewer connections. • Innovative Routing and Fault Tolerance: Implements robust routing mechanisms with fault-tolerant strategies, ensuring network integrity and resilience with minimal component usage. • LIFA Model Implementation: Involves adapting astrocytic and neuronal interactions from a theoretical model to a fully functional computational model, offering computational advantages and insights. We set new benchmarks in energy efficiency, memory management, routing strategies, and fault tolerance. Our methodology, evaluated using various deep learning models, demonstrates efficacy in area and power efficiency while providing robust fault tolerance. II Leaky Integrate-and-Fire Astrocyte (LIFA) Model To understand the astrocytic dynamics in neuromorphic computing, we integrate the LIFA model into our system. The LIFA model is motivated by the principle that the astrocytic Calcium Ions (Ca 2+ ) elevate beyond a threshold; triggering the release of neuroactive molecules such as glutamate and ATP. The released molecules promote postsynaptic neural activity and known as gliotransmitters. Biophysical arguments support that subthreshold Ca 2+ dynamics set the rate of gliotransmitter-mediated postsynaptic depolarizations. The relevant time constants are \tau_{N} for neuronal activity ( v_{N} ), \tau_{G} for astrocytic Ca 2+ activity ( v_{G} ), and \tau_{p} for gliotransmitter dynamics. The ODEs (Ordinary Differential Equations) are: \displaystyle\tau_{N}\frac{dv_{N}}{dt} \displaystyle=-v_{N}+I_{N}(t) (1) \displaystyle\tau_{G}\frac{dv_{G}}{dt} \displaystyle=-v_{G}+I_{G}(t) (2) \displaystyle\tau_{p}\frac{dg}{dt} \displaystyle=-g+G(1-g)r_{G}(t) (3) where I_{N}(t) and I_{G}(t) are the synaptic inputs to neurons and glia, and r_{G}(t) is the gliotransmitter release. Synaptic weights ( w ) are proportional to postsynaptic activation ( q ), i.e. w=uq , and astrocytes contribute to postsynaptic activation by Qg , so w=u(q_{0}+Qg) , where q_{0} is the baseline postsynaptic activation. Incorporating these aspects of the LIFA model into our framework aims to achieve a more biologically accurate and efficient simulation of astrocyte-neuronal interactions. This enhances the realism of our model and provides a robust foundation for advanced computational strategies. (a) Original network. (b) LIFA-modulated network. (c) Operation of LIFA. Fig. 1: Inserting LIFA in a neural network. Fig. 2: Self-repair mechanism of an astrocyte . Figure 1a shows a neural network before astrocyte modulation. Figure 1b shows the network after integrating astrocyte modulation, demonstrating structural and functional changes due to astrocyte integration. Figure 1c shows LIFA’s operation between synaptic site and post neuron activities. Figure 2 illustrates the network’s stages: normal operation, stress, recovery, and potential failure. The black traces show the network’s threshold voltage V_{th} under normal conditions, while the red traces show the threshold voltage under stress. Stress refers to conditions that push the network beyond typical operational parameters. Recovery denotes the network’s ability to return to normal operation after stress, and v_{\text{spk}} refers to the spike voltage that triggers neuron firing, while v_{\text{idle}} denotes the idle voltage. The x-axis represents time, and the y-axis, labeled \Delta V_{th} , represents the change in threshold voltage, showing the network’s stress response and recovery capability over time. III Proposed Design Methodology We used Python to execute implementations on the CPU and GPU. The study leveraged NVIDIA’s GeForce RTX 3060 GPU and Intel’s Core i9 12900H CPU for efficient execution. The LIFA model begins with astrocytes receiving external stimuli, such as optogenetic Ca 2+ uncaging and electrical stimulation of synaptic afferents, which are crucial for initiating astrocytic responses. Astrocytes integrate these stimuli, leading to Ca 2+ signals essential for gliotransmitter release. This integration phase is influenced by the number, duration, and strength of stimuli. Fig. 3: Schematic representation of the LIFA model. Figure 3 illustrates modeling, clustering, error tolerance analysis, fault-tolerant hardware implementation, and fault simulation. Astrocytic mechanisms are incorporated into the LIFA model to improve network resilience and error recovery. The baseline design includes training with data, clustering for pattern recognition, error tolerance analysis, and fault injection testing (ARES). III-A Memory Measurement and Management This section describes the methodology to assess and optimize memory dynamics within SNNs, based on astrocytic activity’s influence on synaptic plasticity and Hopfield network principles. Astrocytic mechanisms introduced by the LIFA model alter synaptic behavior fundamentally, influencing synaptic plasticity, memory formation, and retrieval. Astrocytic modulation evaluates the memory capacity and efficiency of SNNs, inspired by Hopfield networks known for robust pattern recognition capabilities. Our methodology optimizes the number of synaptic connections, using fewer connections to maintain or enhance memory functionality. We analyze memory design capacitance within the SNN, calculating updates due to astrocyte-mediated synaptic modulation. This

Read the complete text