International Journal of Trend in Scientific Research and Development (IJTSRD)

Volume 9 Issue 3, May-Jun 2025 Available Online: [Link] e-ISSN: 2456 – 6470

Neuromorphic Computing
Tripti R Kulkarni, Nandan P, Monisha S. S
ECE Department, Dayananda Sagar Academy of Technology and Management, Bengaluru, Karnataka, India

INTRODUCTION How to cite this paper: Tripti R

What is Neuromorphic Computing? Kulkarni | Nandan P | Monisha S. S
Neuromorphic computing refers to neural-inspired systems designed "Neuromorphic Computing" Published
for non–von Neumann architectures that integrate principles from in International
neuroscience, machine learning, AI, hardware design, and materials Journal of Trend in
Scientific Research
science. Initially focused on analog circuits mimicking biological and Development
neurons and synapses, the field has expanded to encompass a broad (ijtsrd), ISSN: 2456-
range of hardware and software systems. Core features of 6470, Volume-9 |
neuromorphic systems include co-located memory and computation, Issue-3, June 2025, IJTSRD79910
simple communication between neurons and synapses, and local pp.97-103, URL:
learning capabilities. Many also exhibit spiking behavior, nonlinear [Link]/papers/[Link]
dynamics, high connectivity, plasticity, robustness, and the ability to
process noisy or incomplete data. These systems are typically event- Copyright © 2025 by author (s) and
driven, enabling low-power operation and emphasizing temporal International Journal of Trend in
dynamics. Their development requires interdisciplinary collaboration Scientific Research and Development
Journal. This is an
across neuroscience, computer science, engineering, and materials Open Access article
science. distributed under the
terms of the Creative Commons
Attribution License (CC BY 4.0)
([Link]

Current State of Neuromorphic Computing Research

Neuromorphic computing spans multiple disciplines, making it difficult to fully define due to diverse goals and
approaches across neuroscience, computer science, engineering, and materials science. It lies on a spectrum of
repurposable computing platforms, contrasting with synchronous von Neumann architectures through increased
parallelism and asynchrony. One branch of research focuses on accelerating deep learning by creating hardware
tailored to specific networks (e.g., CNNs) and training methods like backpropagation. These systems-often
industry-developed (e.g., Google’s TPU, Intel’s Nervana Engine)-fit the neuromorphic definition but rely heavily
on large labeled datasets, differing from other neuromorphic approaches.

Figure1: Spectrum of repurposable computing platforms [WSP: Hylton]

Figure 2: Two types of memristors that could be used in neuromorphic systems [Chua1971,
WSP:Williams].

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 97

 International Journal of Trend in Scientific Research and Development @ [Link] eISSN: 2456-6470
Advancements and Architectural Perspectives in Neuromorphic Computing
One of the most popular technologies associated with building neuromorphic systems is the memristor (also
known as ReRAM). There are two general types of memristors: nonvolatile, which is typically used to implement
synapses, and locally active, which could be used to represent a neuron or axon (Figure 2). Nonvolatile
memristors are also used to demonstrate activation functions and other logical computations. Memristors used to
implement synapses are often used in a crossbar.

Figure 3: ReRAM used as synapses in a crossbar array [WSP:Saxena].

Figure 4: A von Neumann or traditional architecture from the computer science perspective
Open Issues
Neuromorphic computing includes researchers in fields such as neuroscience, computing, computer and electrical
engineering, device physics, and materials science. The focus of the workshop was to identify the major questions
from a computing perspective of neuromorphic computing or questions that can be addressed primarily by
computational scientists, computer scientists, and mathematicians and whose solutions can benefit from the use of
high-performance computing (HPC) resources.

Figure 5: A potential neuromorphic architecture from the computer science perspective

Neuron, Synapse, and Biological Component Models in Neuromorphic Systems
Neuromorphic computing systems require careful selection of neuron and synapse models. Neuron models vary
from simple threshold-based models like McCulloch-Pitts to complex biologically accurate ones like Hodgkin-
Huxley. These models differ in biological realism, computational cost, and suitability for different applications.

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 98

 International Journal of Trend in Scientific Research and Development @ [Link] eISSN: 2456-6470

Figure 6: Example neuron models [Hodgkin1952, Izhikevich2003, Gerstner2002, McCulloch1943].

Neuromorphic Computing: Biological Realism vs. Functionality
One of the key research questions in neuromorphic computing is determining the level of complexity and
biological realism necessary to achieve desired functionality. The models selected for a neuromorphic system
should align with the system's ultimate goal. If the goal is biologically realistic simulations, the models must
replicate biological systems as accurately as possible. However, the focus of this report is on the notion that the
primary aim of neuromorphic architecture should be to create computationally efficient systems, not necessarily
to mimic biological behaviour in detail.

Figure 7: Levels of abstraction in biological brains and what functionality they may allow
[WSP:Aimone].
LITERATURE REVIEW dynamics, enabling multi-state memory and leaky
[1] This paper introduces an in-memory integrate-and-fire neuron behavior. When integrated
neuromorphic computing (IMNC) chip that supports a into a spiking neural network, the system achieves up
hybrid topology of spiking and artificial neural to 90% accuracy on the MNIST dataset, showcasing
networks (S/ANNs). The chip features a ring-based potential for energy-efficient neuromorphic
architecture optimized for sparse data flows, computing. IEEE Resource Center.
achieving high energy efficiency and accuracy. [3] This research explores the integration of photonic-
Experimental results demonstrate over 95% accuracy electronic resonant tunneling diode (RTD) neurons
in tasks like voice activity detection and ECG with spiking flip-flop memory for neuromorphic
anomaly detection, with a dynamic energy computing. The proposed system utilizes spike-
consumption of 0.43 pJ per synaptic operation. encoded information processing, leveraging the high-
[Link]. speed and low-power characteristics of RTD devices.
[2] This work presents multilayer spintronic devices This approach aims to enhance the performance and
that function as both synapses and neurons in efficiency of neuromorphic systems.
neuromorphic systems. The devices exhibit discrete [4] Yang discusses the use of integrated memristor
resistance states due to magnetic domain wall networks to address the challenges of higher-

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 99

 International Journal of Trend in Scientific Research and Development @ [Link] eISSN: 2456-6470
complexity neuromorphic computing. The paper emphasizes the importance of neuromorphic
highlights the potential of memristor-based principles in achieving efficient and scalable AI
architectures to emulate complex neural behaviors systems.
and facilitate scalable, energy-efficient computing
[11] This work addresses the challenge of training
systems. The integration of memristors offers
quantized spiking neural networks (QSNNs) by
advantages in terms of density and functionality for
proposing a cosine-annealed learning rate schedule
neuromorphic applications.
combined with weight-independent adaptive moment
[5] NeuroSEE presents a neuromorphic processing estimation. The approach mitigates issues arising
framework designed for visual prostheses. By from gradient discontinuities during training, enabling
utilizing spike representation encoding and a bio- QSNNs to escape local minima and achieve near-
inspired spiking neural network model, the state-of-the-art performance on complex datasets. The
framework achieves significant energy efficiency authors provide empirical evaluations demonstrating
improvements-up to 15 times lower power the effectiveness of their method across multiple
consumption compared to conventional CNN-based [Link].
approaches. The system demonstrates high correlation
[12] This study explores the impact of dense and
with primary visual cortex responses, indicating its
sparse mapping schemes on the performance of
potential for prosthetic applications. IEEE EMBS
resistive random-access memory (RRAM)
[6] This paper explores the application of deep architectures in deep learning applications. The
learning techniques to train spiking neural networks authors present a design space exploration
(SNNs). It discusses the challenges and methodology to quantify the benefits and limitations
methodologies for adapting backpropagation and of these mapping schemes, considering factors such
other optimization strategies to the spiking domain. as power consumption, noise susceptibility, and
The study provides insights into bridging the gap network architecture. Their findings provide insights
between traditional deep learning and spiking neural into optimizing RRAM-based accelerators for various
computation. deep learning [Link].
[7] Ottati et al. examine the trade-offs between [13] This resource provides an overview of
spiking and non-spiking digital hardware neuromorphic computing, focusing on the integration
architectures for deep learning acceleration. The of memristors in hardware design to emulate neural
paper provides a comparative analysis of energy processing. It covers the transition from circuit-level
efficiency, computational complexity, and implementations to algorithmic considerations,
performance, offering guidance on selecting highlighting the potential of memristor-based systems
appropriate architectures for specific applications. in achieving energy-efficient and brain-inspired
[8] Sharma reviews the role of memristor-based computing solutions. The content is aimed at
networks in neuromorphic computing for artificial researchers and practitioners interested in the
intelligence tasks. The article discusses the interdisciplinary aspects of neuromorphic computing.
advantages of memristor devices in implementing [14] This paper investigates the analog synaptic
synaptic weights and their impact on network behaviors of carbon-based self-selective RRAM
dynamics, learning capabilities, and scalability. It also devices for in-memory supervised learning
addresses challenges such as variability and non- applications. The authors demonstrate the potential of
linearity in memristor-based systems. these devices to emulate synaptic functions, such as
[9] Li et al. propose a multi-core neuromorphic weight update and retention, which are crucial for
system aimed at enhancing the energy efficiency of neuromorphic computing systems. Their findings
deep neural network training. The system leverages contribute to the development of more efficient and
parallel processing and in-memory computing scalable memristor-based learning systems.
techniques to reduce power consumption while [15] This research presents an implementation of a
maintaining high performance. Experimental results quantized convolutional neural network (CNN) on a
demonstrate the system's effectiveness in training parallel-connected memristor crossbar array, targeting
complex models with reduced energy requirements. edge AI platforms. The authors propose a radix-5
[10] Cauwenberghs discusses the design of CNN architecture utilizing 1-bit memristors,
neuromorphic circuits tailored for large-scale achieving learning results comparable to high-
artificial intelligence applications. The paper covers precision models while reducing the area of the
aspects such as circuit architectures, scalability, and memristor crossbar array by half. Their work
integration with existing AI frameworks. It demonstrates the feasibility of deploying efficient

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 100
 International Journal of Trend in Scientific Research and Development @ [Link] eISSN: 2456-6470
neural network models on memristor-based hardware
for edge computing applications.
[16],[17],[18] This paper presents FGMSVM with
one-against-one and maximum voting, using K-means
clustering for noninvasive cardiovascular disease
screening. It also explores a deep learning method for
detecting five arrhythmia types via PPG and surveys
approaches for noninvasive fetal oxygen saturation
measurement using PPG.
Fig 1.1.2: Neuron & Synapse Modeling
[19],[20] This research presents a novel method to
modify the current VCO for adjustable output voltage
levels and develops an OP-AMP circuit using 22 nm
FinFET technology with high-k gain for improved
performance.
METHEDOLOGY
Spiking Neural Network (SNN) Architecture
Neuromorphic computing systems emulate the brain
by using Spiking Neural Networks (SNNs), which
process information through discrete spikes, closely
resembling biological neural activity. This approach
enables low-power, event-driven computation and Fig 1.1.3: Event-Driven Communication
offers improved temporal dynamics compared to
traditional neural networks. Neurons activate only
when input exceeds a threshold, communicating via
timed spikes, ensuring energy is used only by active
components. A notable example is IBM’s TrueNorth
chip, which simulates over one million neurons and
256 million synapses for tasks such as visual and
pattern recognition.
Neuron and Synapse Modeling
Neuromorphic chips are composed of artificial
neurons and synapses designed to replicate key Fig 1.1.4: Hardware Implementation
biological functions, enabling complex learning,
memory, parallelism, and adaptability-fundamental to
bio-inspired intelligence. Neurons integrate incoming
spikes and generate outputs, while synapses adjust
signal strength based on learning rules such as Spike-
Timing Dependent Plasticity (STDP). These
functionalities are implemented using digital, analog,
or mixed-signal circuits. A notable example is Intel’s
Loihi chip, which supports real-time, on-chip learning
through programmable synaptic plasticity.
Fig 1.1.5: Neuromorphic Brain Fusion

Fig 1.1.1: SNN Architecture Fig 1.1.6: Neuromorphic Computing

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 101
 International Journal of Trend in Scientific Research and Development @ [Link] eISSN: 2456-6470
Fig 1.1.1: SNN Architecture and Fig 1.1.2: Neuron & energy efficiency by triggering computation only
Synapse Modeling illustrate the core components of during events. Hardware implementations using
Spiking Neural Networks (SNNs), where neurons digital, analog, or mixed-signal circuits replicate
communicate via asynchronous spikes, mimicking neuron and synapse behavior for real-time, on-chip
biological brain activity. Neurons fire when a learning. Lastly, Fig 1.1.5: Neuromorphic Brain
threshold is reached, and synapses adjust their strength Fusion and Fig 1.1.6: Neuromorphic Computing
based on learning rules like Spike-Timing Dependent highlight the integration of neuromorphic systems into
Plasticity (STDP), enabling real-time adaptation. Fig practical applications like pattern recognition and
1.1.3: Event-Driven Communication and Fig 1.1.4: decision-making, demonstrating the potential of bio-
Hardware Implementation shift focus to the event- inspired, energy-efficient computing.
driven nature of neuromorphic systems, ensuring

Fig: Neuromorphic Computing: The Future Brains of Computing

Neuromorphic computing is an innovative paradigm as memristors and specialized chips like IBM's
inspired by the structure and function of the human TrueNorth and Intel's Loihi, underscore the potential
brain, aiming to revolutionize computing by creating of neuromorphic systems to revolutionize fields
systems that mimic biological neural networks. This ranging from artificial intelligence to robotics. As
approach utilizes specialized hardware, such as interdisciplinary research continues to evolve,
spiking neural networks (SNNs), which process neuromorphic computing is poised to unlock new
information through discrete spikes, enabling energy- possibilities in cognitive computing, offering scalable
efficient, parallel, and event-driven computation. By and efficient solutions that closely mirror biological
integrating memory and processing, neuromorphic processes and pave the way for the next generation of
systems are designed to learn and adapt in real-time, intelligent systems.
offering significant advantages over traditional
REFERENCE
computing architectures in tasks that require cognitive
[1] Y. Liu et al., "Sparsity-Aware In-Memory
functions, pattern recognition, and decision-making.
Neuromorphic Computing Unit With
With applications spanning from artificial intelligence
Configurable Topology of Hybrid Spiking and
to robotics, neuromorphic computing holds the
Artificial Neural Network," IEEE Trans.
potential to deliver scalable, low-power, and highly
Circuits Syst. I, vol. 71, no. 6, pp. 2660–2673,
adaptive systems that can better emulate the brain's
Jun. 2024.
processing capabilities, ushering in the future of
intelligent, bio-inspired computing. [2] A. H. Lone et al., "Multilayer Spintronic
Devices as Synapse and Neuron for
CONCLUSION
Neuromorphic Computing Applications," IEEE
In conclusion, neuromorphic computing represents a
Resource Center, Oct. 2023.
transformative approach to computing that draws
inspiration from the brain's structure and function, [3] "Neuromorphic spiking flip-flop memory and
offering a promising alternative to traditional von spike-encoding with photonic-electronic RTD
Neumann architectures. By leveraging spiking neural neurons," IEEE Photonics Conference, Jan.
networks, event-driven communication, and bio- 2025.
inspired learning rules, neuromorphic systems enable [4] Y. Yang, "Integrated Memristor Networks for
energy-efficient, parallel, and adaptive computation, Higher-Complexity Neuromorphic
making them well-suited for tasks such as pattern Computing," IEEE Resource Center, Jun. 2023.
recognition, decision-making, and real-time learning.
The ongoing advancements in hardware design, such

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 102
 International Journal of Trend in Scientific Research and Development @ [Link] eISSN: 2456-6470
[5] C. Wang et al., "NeuroSEE: A Neuromorphic as diabetes and cerebral infarction using
Energy-Efficient Processing Framework for photoplethysmography signals,
Visual Prostheses," IEEE J. Biomed. Health ResultsinOptics,Volume3,2021,100062,ISSN26
Inform., vol. 26, no. 8, pp. 4132–4141, Aug. 669501,[Link]
2022. 62.
[6] J. K. Eshraghian et al., "Training Spiking [17] Performance evaluation of deep learningmodels
Neural Networks Using Lessons From Deep in detection of different types of arrhythmia
Learning," Proc. IEEE, Sep. 2023. using photo plethysmography signals TR
[7] F. Ottati et al., "To Spike or Not to Spike: A Kulkarni, ND Dushyanth - International Journal
Digital Hardware Perspective on Deep of Information Technology, 2021
Learning Acceleration," IEEE J. Emerg. Sel. [18] T. Kulkarni and B. N. Krupa, "Noninvasive
Topics Circuits Syst., vol. 13, no. 4, pp. 1015– method to find oxygen saturation of the fetus
1025, Nov. 2023. using photoplethysmogram," 2016 International
[8] S. Sharma, "Memristor-Based Neuromorphic Conference on Electrical, Electronics,
Networks for AI Tasks," IEEE Circuits Syst. Communication, Computer and Optimization
Mag., vol. 25, no. 3, pp. 121–132, Aug. 2024. Techniques (ICEECCOT), Mysuru, India,
2016, pp. 40-42,
[9] M. Li et al., "Energy-Efficient Multi-Core doi:10.1109/ICEECCOT.2016.7955182
Neuromorphic System for Deep Neural
Network Training," IEEE Trans. Comput. [19] L, Srivathsava & Kulkarni, Tripti. (2011).
Imaging, vol. 10, no. 4, pp. 500–510, Dec. Novel Design of VCO with Output Peak to
2024. Peak Control. International Journal of
Instrumentation Control and Automation. 192-
[10] G. Cauwenberghs, "Neuromorphic Circuits for 195. 10.47893/IJICA.2011.1035.
Large-Scale Artificial Intelligence," IEEE J.
Solid-State Circuits, vol. 60, no. 6, pp. 1473– [20] Gowdagere, Vasu & Gururaj, Bharathi &
1481, Jun. 2024. Jatkar, Mandar & Kulkarni, Tripti & Kulkarni,
Roopa. (2023). Design of FinFET Based Op-
[11] J. K. Eshraghian et al., "Navigating Local Amp Using High-K Device 22 nm Technology.
Minima in Quantized Spiking Neural 10.3233/ATDE231033.
Networks," 2022 IEEE Artif. Intell. Circuits
Syst. Conf., Incheon, Korea, Jun. 2022. [21] J. K. Eshraghian et al., "A Leaky-Integrate-and-
Fire Neuron Model of Spontaneous Reset of
[12] C. Lammie et al., "Design Space Exploration of Thin-Film Metal-Oxide Resistive Switches," J.
Dense and Sparse Mapping Schemes for Nanoscience Nanotechnology, vol. 21, no. 3,
RRAM Architectures," 2022 IEEE Int. Symp. pp. 1920–1926, Mar. 2021.
Circuits Syst., Austin, TX, USA, May 2022.
[22] S. You et al., "Low-Power Wireless Sensor
[13] "Neuromorphic Computing with Memristors: Network Using Fine-Grain Control of Sensor
From Circuits to Algorithms," IEEE Resource Module Power Mode," Sensors, vol. 21, no. 9,
Center, Jan. 2025. p. e3198, Jan. 2021.
[14] Y. C. Chen et al., "Analog Synaptic Behaviors [23] X. Wang et al., "High Density Memristor-
in Carbon-Based Self-Selective RRAM for In- CMOS Ternary Logic Family," IEEE Trans.
Memory Supervised Learning," 2021 IEEE 71st Circuits Syst. I, vol. 68, no. 1, pp. 264–274,
Electron. Components Technol. Conf., pp. Jan. 2021.
1645–1651, 2021.
[24] G. Cauwenberghs, "Moore with Less: Ultra-
[15] J. Lee et al., "Quantized Convolutional Neural Low Energy Neuromorphic Circuits and
Network Implementation on a Parallel- Systems for Large-Scale Distributed AI," IEEE
Connected Memristor Crossbar Array for Edge Resource Center, May 2024.
AI Platforms," J. Nanoscience Nanotechnology,
vol. 21, no. 3, pp. 1854–1861, Mar. 2021. [25] M. Isik et al., "NeuroVM: Dynamic
Neuromorphic Hardware Virtualization," IEEE
[16] (Kulkarni, 2021)Tripti R Kulkarni, N.D. High Performance Embedded Computing
Dushyant, Early and noninvasive screening of (HPEC) Proceedings, Sep. 2024.
common cardio vascular related diseases such

@ IJTSRD | Unique Paper ID – IJTSRD79910 | Volume – 9 | Issue – 3 | May-Jun 2025 Page 103