Neuromorphic Computing:

The Future of Intelligent

Machines
A
SEMINAR REPORT
Submitted by
Sakshi Bajare
(Roll No.: 2314511254)

In partial fulfillment for the award of the degree

of
MASTER OF APPLICATIONS
In
Directorate of Online Education
Manipal University Jaipur

2023-2025
 Table of contents

Abstract .............................................. 1–2

• Brief overview of neuromorphic computing
• Key objectives and significance
• Summary of findings and potential impact

Chapter I: Introduction ........................ 4–6

• Background & Motivation
– Evolution from von Neumann to brain-inspired computing
• Problem Statement
– Limitations of traditional architectures
• Significance
– Why neuromorphic computing matters
• Scope of the Report
– Overview of topics covered (theory, hardware, applications, challenges)

Chapter II: Seminar Objectives and Methodology ........ 7-8

• Objectives of the Study
– Examination of neuromorphic algorithms, hardware designs, and applications
• Methodology
– Literature survey and data collection
– Comparative analysis with traditional computing
– Case studies/examples

Chapter III: Literature Review .................... 9-13

• Theoretical Foundations
– Biological inspiration and neural network models
– Spiking Neural Networks (SNNs), STDP, Hebbian Learning
• Hardware Implementations
– Overview of neuromorphic chips (IBM TrueNorth, Intel Loihi, SpiNNaker)
– Role of memristors and emerging materials
• Comparative Analysis
– Comparison with conventional architectures
• Current Research Trends & Challenges

Chapter IV: Industry Applications and Challenges .... 14-17

• Robotics & Automation
– Real-time control and energy efficiency
• AI & Machine Learning
– Integration for unsupervised learning and pattern recognition
• Edge Computing & IoT
– Low-power, on-device processing solutions
• Healthcare & Biomedical Applications
 – Brain–machine interfaces, prosthetics, medical imaging
• Aerospace & Defense
– Autonomous navigation, target recognition
• Challenges
– Scalability, energy consumption, programming complexity, standardization issues

Chapter V: Future Scope and Opportunities ........ 18–21

• Emerging Trends
– Integration with AI, hybrid systems, potential synergy with quantum computing
• Opportunities for Innovation
– New application areas and cross-disciplinary research
• Roadmap for Future Research
– Proposed directions and milestones
• Industry Impact
– How neuromorphic systems can redefine technology sectors

Chapter VI: Results and Discussion .............. 19–25

• Key Findings
– Summary of insights from literature and case studies
• Comparative Analysis
– Evaluating neuromorphic vs. traditional architectures
• Discussion on Implementation Challenges
– Practical issues and performance metrics
• Energy Efficiency & Performance Metrics
– Analysis based on research data and simulations

Chapter VII: Conclusion and Recommendations ....... 26–29

• Summary of the Study
– Recap of main insights and contributions
• Concluding Remarks
– Final thoughts on the state and future of neuromorphic computing
• Recommendations
– Suggestions for future research, industry adoption strategies, and policy considerations

Chapter VIII: References ......................... 30

• Comprehensive list of all cited research papers, articles, books, and online sources
Section A

Abstract
 Abstract

1. Overview of Neuromorphic Computing

Neuromorphic computing is an emerging computational paradigm inspired by the architecture
and functionality of the human brain. Unlike traditional von Neumann systems, which suffer
from the inherent separation of memory and processing units, neuromorphic systems integrate
these functions into networks of artificial neurons and synapses. This brain-inspired approach
enables highly parallel, event-driven processing that significantly reduces energy consumption
while offering real-time adaptive learning capabilities. Research by Merolla et al. (2014) on
IBM’s TrueNorth chip and subsequent developments in Intel’s Loihi architecture (Davies et al.,
2018) have demonstrated the feasibility of such designs, paving the way for systems that can
operate with both enhanced speed and efficiency.
Building on these innovations, neuromorphic systems replicate key aspects of biological neural
processing, such as synaptic plasticity and asynchronous signaling, which are essential for
learning and memory in the brain. By mimicking mechanisms like spike-timing-dependent
plasticity (STDP) and Hebbian learning, these systems can continuously adapt to changing
sensory inputs without relying on conventional training methods. The integration of memory
and processing at a microscopic level not only overcomes the bottlenecks of data transfer in
traditional architectures but also enables massive parallelism—allowing neuromorphic
hardware to process complex data streams in real time. Consequently, these attributes make
neuromorphic computing highly suitable for applications ranging from robotics and edge
computing to advanced artificial intelligence tasks, where energy efficiency and rapid
adaptation are paramount.

2. Key Objectives & Significance

This study aims to conduct a comprehensive investigation of neuromorphic computing by
examining its theoretical underpinnings, hardware implementations, and diverse industry
applications. The objectives are threefold:

• Theoretical Exploration: To review the principles of brain-inspired computing,

focusing on how SNNs and associated learning rules provide the basis for energy-
efficient, adaptive computation.

• Hardware & Implementation Analysis: To analyze the performance and design of

state-of-the-art neuromorphic chips such as IBM TrueNorth, Intel Loihi, and the
SpiNNaker system, while considering emerging materials like memristors that facilitate
analog signal processing.

• Application & Impact Assessment: To explore the integration of neuromorphic

systems in various fields—including robotics, artificial intelligence, edge computing,

1
Neuromorphic computing
 healthcare, and aerospace—assessing both their current use cases and potential future
impact.
The significance of this study lies in its potential to bridge the gap between theoretical
neuroscience and practical computing solutions. By addressing the limitations of conventional
architectures and demonstrating how neuromorphic systems can lead to more efficient,
scalable, and adaptive computing, the research offers a promising roadmap for next-generation
applications that require both high performance and energy conservation.

3. Summary of Findings & Potential Impact

Our review of recent research indicates that neuromorphic computing offers transformative
advantages over traditional processing paradigms. Key findings include:

• Energy Efficiency: Neuromorphic architectures can achieve dramatic reductions in

power consumption through event-driven processing, making them particularly suitable
for real-time and edge applications.

• Adaptive Learning: The incorporation of biologically inspired learning rules, such as

STDP, enables neuromorphic systems to perform unsupervised and adaptive learning
with minimal external intervention, which is crucial for dynamic environments.

• Scalability & Integration: Early implementations, such as those by Merolla et al.

(2014) and Davies et al. (2018), demonstrate that neuromorphic chips can be scaled to
support large networks of neurons, which, when integrated with conventional
computing elements, have the potential to enhance overall system performance.

• Diverse Applications: The ability to process information in a parallel and energy-

efficient manner positions neuromorphic systems as viable solutions in fields ranging
from autonomous robotics and sensor networks to advanced biomedical devices and
brain-machine interfaces.
The potential impact of these findings extends beyond technical innovation. Neuromorphic
computing is poised to revolutionize how we design and implement intelligent systems,
offering a path toward more sustainable and adaptive technologies that closely mirror the
efficiency of natural neural processes. As research continues to evolve, the convergence of
neuroscience, materials science, and computer engineering is expected to further refine these
systems, fostering breakthroughs that may redefine the landscape of artificial intelligence and
high-performance computing.

2
Neuromorphic computing
 Section B

Seminar Report

3
Neuromorphic computing
 Chapter I : Introduction

1.1 Background & Motivation

Traditional computing systems, predominantly based on the von Neumann architecture,
have served as the backbone of modern digital technology. However, as applications become
more complex and data-intensive, the inherent separation between processing and memory in
these systems creates significant performance and energy efficiency bottlenecks. In contrast,
the human brain exemplifies a highly efficient and adaptive information-processing system,
where memory and computation occur simultaneously within a vast, interconnected network
of neurons and synapses. Inspired by this natural efficiency, the field of neuromorphic
computing has emerged as a promising alternative.
Early demonstrations, such as IBM’s TrueNorth chip (Merolla et al., 2014), showcased how
millions of spiking neurons could be integrated on a single chip to perform complex tasks with
remarkably low power consumption. Similarly, Intel’s Loihi processor (Davies et al., 2018) has
advanced the concept by incorporating on-chip learning capabilities that emulate biological
neural adaptation. These pioneering works have motivated researchers to explore
neuromorphic architectures as a means to overcome the limitations of conventional computing
systems, especially in applications that require real-time processing and energy efficiency.

1.2 Problem Statement

The limitations of traditional digital computing become particularly apparent when
addressing tasks that require rapid, low-energy decision-making and learning in dynamic
environments. Conventional architectures often struggle with issues such as memory
bottlenecks, high power consumption, and inefficient data transfer between separated
processing and memory units. These challenges have spurred the development of alternative
approaches that seek to mimic the brain’s inherent parallelism and adaptive capabilities.
Neuromorphic computing addresses these issues by employing spiking neural networks
(SNNs) and biologically inspired learning mechanisms like spike-timing-dependent plasticity
(STDP) and Hebbian learning (Indiveri et al., 2011). These systems are designed to process
information in an event-driven manner, significantly reducing power consumption and latency
compared to traditional systems.

1.3 Significance
The significance of neuromorphic computing extends across multiple dimensions. First,
its energy efficiency is of paramount importance in an era where power constraints and
environmental considerations are critical. By processing only the essential events rather than
continuously clocking all circuits, neuromorphic systems offer a scalable solution for
embedded and edge applications. Second, the adaptive learning capabilities inherent to

4
Neuromorphic computing
neuromorphic architectures—demonstrated by on-chip learning in processors like Loihi—
enable these systems to dynamically adjust to new data and environmental conditions without
extensive retraining. This is especially valuable in fields such as robotics, autonomous systems,
and healthcare, where responsiveness and adaptability are essential.
Furthermore, neuromorphic computing represents a convergence of neuroscience, materials
science, and computer engineering. The integration of emerging technologies such as
memristors, which emulate synaptic behavior through their ability to retain and modulate
conductance states, has broadened the scope of neuromorphic hardware (Basu et al., 2018).
Collectively, these advancements underscore the transformative potential of neuromorphic
systems—not only as a means to address current computational challenges but also as a
pathway toward developing future intelligent systems that mirror the efficiency and
adaptability of biological brains.

1.4 Scope of the Report

This report aims to provide a comprehensive examination of neuromorphic computing by
addressing both its theoretical underpinnings and practical applications. The scope of the study
includes:

• A detailed review of the theoretical foundations of neuromorphic computing,

including the biological inspiration behind spiking neural networks and learning
algorithms such as STDP.

• An analysis of hardware implementations, reviewing state-of-the-art neuromorphic

chips like IBM TrueNorth, Intel Loihi, and other emerging systems that leverage
advanced materials and device technologies.

• An exploration of industry applications, assessing how neuromorphic computing is

being applied in domains such as robotics, artificial intelligence, edge computing, and
biomedical engineering.

• A discussion of the challenges and limitations currently faced by neuromorphic

systems, such as scalability, programming complexity, and the need for standardized
development frameworks.

• Insights into the future scope and potential impact of neuromorphic computing,
including emerging trends and opportunities for further research.

5
Neuromorphic computing
By integrating perspectives from leading research papers and technological case studies, this
report seeks to not only elucidate the current state of neuromorphic computing but also to
highlight its promising future in revolutionizing next-generation computational paradigms.

6
Neuromorphic computing
 Chapter II : Seminar Objectives and Methodology

2.1 Objectives of the study

The primary objective of this study is to provide a comprehensive investigation into
neuromorphic computing by examining its theoretical foundations, hardware implementations,
and practical applications. Drawing upon established research (e.g., Merolla et al., 2014;
Davies et al., 2018; Indiveri et al., 2011; Basu et al., 2018), the study aims to achieve the
following specific objectives:

• Theoretical Exploration: To review the fundamental principles underlying

neuromorphic computing, with a particular emphasis on spiking neural networks
(SNNs) and biologically inspired learning mechanisms such as spike-timing-dependent
plasticity (STDP) and Hebbian learning. These mechanisms are critical for replicating
the adaptive behavior observed in biological neural systems.

• Hardware Analysis: To analyze the design, performance, and scalability of state-of-

the-art neuromorphic hardware. This includes a detailed examination of systems such
as IBM’s TrueNorth, Intel’s Loihi, and the SpiNNaker platform. The study also
considers emerging technologies (e.g., memristors) that have been highlighted in recent
literature as key enablers for efficient synaptic emulation.

• Application Assessment: To explore how neuromorphic computing is applied across

various industries—ranging from robotics and artificial intelligence to healthcare and
aerospace—and to assess its potential in overcoming the limitations of conventional
computing architectures. By evaluating case studies and experimental results from
published research, this study seeks to elucidate the practical advantages and challenges
of neuromorphic systems.

• Comparative Evaluation: To compare neuromorphic architectures with traditional

von Neumann systems in terms of energy efficiency, computational speed, and
adaptability, thereby establishing the contexts in which neuromorphic computing
provides a tangible benefit.

2.2 Methodology
This study employs a systematic literature review methodology, underpinned by qualitative
analysis, to synthesize findings from a range of peer-reviewed research papers and technical
reports. The methodological approach comprises the following steps:

• Literature Collection: Relevant articles, conference proceedings, and technical

documents were sourced from established databases such as IEEE Xplore, Nature, and

7
Neuromorphic computing
 Springer. Key studies (e.g., Merolla et al., 2014; Davies et al., 2018; Indiveri et al.,
2011) served as primary references.
• Thematic Analysis: Collected literature was categorized according to thematic areas—
such as theoretical foundations, hardware implementations, and industry applications.
This allowed for the identification of common trends, recurring challenges, and areas
of innovation in neuromorphic computing.

• Comparative Evaluation: The study incorporates a comparative analysis to juxtapose

neuromorphic systems with traditional computing architectures. Performance metrics,
such as energy efficiency and computational throughput, were examined based on
reported experimental data.

• Case Studies & Examples: Where applicable, case studies and experimental findings
from existing research were integrated into the analysis to provide concrete examples
of neuromorphic computing in action. These examples offer insights into both the
practical benefits and the current limitations faced by neuromorphic systems.

• Synthesis of Findings: Finally, the study synthesizes the reviewed literature to present
an integrated perspective on neuromorphic computing, highlighting both its current
state and potential future directions.

8
Neuromorphic computing
 Chapter III : Literature Review
This chapter provides an in-depth review of neuromorphic computing by synthesizing findings
from a broad spectrum of peer-reviewed research papers and technical reports. The discussion
is organized into five major subsections: Theoretical Foundations, Hardware Implementations,
Applications and Use Cases, Comparative Analysis and Performance Metrics, and Challenges
and Future Directions. Each subsection draws upon seminal works and recent advances to
present a detailed picture of the field.

3.1 Theoretical Foundations

Neuromorphic computing is fundamentally rooted in the desire to emulate the remarkable
computational efficiency of the human brain. Traditional computing models, based on the von
Neumann architecture, separate memory and processing, leading to bottlenecks that limit
performance and energy efficiency. In contrast, the brain’s structure—characterized by a
massive network of interconnected neurons and synapses—serves as an inspiration for
developing systems that process information in parallel and in an event-driven manner.
A cornerstone of neuromorphic theory is the development of Spiking Neural Networks (SNNs).
Unlike conventional neural networks that use continuous values, SNNs process information
using discrete spikes, more closely mimicking the way biological neurons communicate.
elaborates on the mechanisms of spike-timing-dependent plasticity (STDP) and Hebbian
learning, which allow these networks to adapt based on the temporal correlation of spikes. This
adaptive learning capability is critical for real-time, unsupervised learning, as it enables
neuromorphic systems to dynamically modify synaptic strengths in response to environmental
stimuli.
Further theoretical models have focused on the integration of memory and computation. By
interweaving these two traditionally separate functions, neuromorphic architectures aim to
overcome the von Neumann bottleneck, enabling simultaneous processing and storage of
information. Theoretical studies have demonstrated that such architectures can achieve a level
of computational efficiency that is orders of magnitude better than that of conventional systems,
particularly for tasks requiring parallel processing and low power consumption. In addition,
research published on arXiv has provided mathematical models and simulations that
underscore the potential of neuromorphic principles to transform computational paradigms.

3.2 Hardware Implementations

The evolution from theoretical constructs to practical hardware has been a major driver of
progress in neuromorphic computing. One of the seminal contributions in this domain is IBM’s
TrueNorth chip, as presented by Merolla et al. (2014). TrueNorth integrates over one million
programmable neurons and 256 million synapses on a single chip, achieving ultra-low power
consumption while executing complex neural computations (Nature). This work sets a

9
Neuromorphic computing
precedent for subsequent neuromorphic designs by proving that large-scale neural networks
could be embedded in hardware.
Following TrueNorth, Intel’s Loihi processor has further advanced the field by incorporating
on-chip learning mechanisms. Loihi supports real-time adaptation by implementing neural
models that can modify their synaptic weights based on incoming stimuli, a feature that mirrors
biological learning. Detailed insights into Loihi’s architecture and performance can be found
on Intel’s research portal (Intel Neuromorphic Computing). Loihi has demonstrated the ability
to process sensory information efficiently, making it suitable for applications that demand low
latency and high adaptability.

Fig : Illustration of how a conventional fully-connected neural network (left) can be mapped
onto the neurosynaptic core of IBM’s TrueNorth chip (right). In TrueNorth, neurons, axons,
and synapses are organized in a crossbar array to enable highly parallel, event-driven
processing. Each core integrates both computation and memory, allowing synaptic connections
and neuron states to be updated in real time with minimal power consumption. The PRNG
(pseudo-random number generator) module supports stochastic neural behaviors and learning
rules.
Another noteworthy system is the SpiNNaker platform, developed by the University of
Manchester. SpiNNaker is designed to simulate large-scale SNNs and is particularly well-
suited for real-time applications in robotics and cognitive modeling. The architecture of
SpiNNaker allows for the simulation of millions of neurons concurrently, and its scalability has
been highlighted in numerous technical reports available on its official website.

10
Neuromorphic computing
Fig. Block diagram of the neuromorphic hardware proposed by Yin et al. [49], illustrating a
three-layer (256-256-10) configuration with spike encoding, weight fetching, and dedicated
SRAM for weight storage, which contributes to high energy efficiency.

In addition to these digital implementations, there is growing interest in leveraging emerging

materials, such as memristors, to build analog neuromorphic systems. Memristive devices are
capable of emulating synaptic behavior by adjusting their resistance based on past activity,
thereby offering a pathway to energy-efficient, high-density neuromorphic circuits. Basu et al.
(2018) discuss the integration of memristors into neuromorphic architectures in detail, and their
findings can be accessed via RSC Publications.

3.3 Applications & Use Cases

The practical implications of neuromorphic computing are vast, with numerous applications
spanning across different industries. In robotics, neuromorphic systems are being used to
enhance sensor processing and real-time control. For instance, SNN-based controllers allow
robots to adapt rapidly to dynamic environments, improving both precision and efficiency.
Such capabilities are critical for autonomous navigation and manipulation tasks in uncertain or
changing conditions.
Within the field of artificial intelligence, neuromorphic architectures are gaining traction as
energy-efficient alternatives for performing complex computations. Research featured in
Nature Electronics (Nature Electronics) has demonstrated that neuromorphic processors can be
used for deep learning tasks, particularly in scenarios where traditional high-power GPUs are
impractical. These systems are particularly advantageous in edge computing applications,
where processing must occur locally to reduce latency and conserve energy.
In healthcare, neuromorphic computing has promising applications in the development of
brain–machine interfaces (BMIs) and advanced biomedical signal processing. By accurately

11
Neuromorphic computing
emulating the brain’s neural dynamics, neuromorphic devices can interpret complex neural
signals for applications such as prosthetic control and neurorehabilitation. Studies have shown
that the parallel and adaptive nature of neuromorphic systems enables improved performance
in interpreting electroencephalography (EEG) and other biosignals, with significant
implications for patient care and medical diagnostics.
The Internet of Things (IoT) also stands to benefit from neuromorphic approaches. With the
proliferation of connected devices, the need for efficient, on-device processing is more critical
than ever. Neuromorphic processors, with their low-power consumption and real-time
responsiveness, are ideal for processing sensor data locally, thereby reducing the reliance on
cloud-based resources. This capability is essential for applications that require immediate
decision-making, such as environmental monitoring and smart city infrastructures.

3.4 Comparative Analysis & Performance Metrics

A major area of focus in neuromorphic research is the comparison of these systems with
conventional digital architectures. Numerous studies have quantified the benefits of
neuromorphic systems in terms of energy efficiency, latency, and scalability. For example, the
TrueNorth chip and Loihi processor have been benchmarked against traditional CPUs and
GPUs, with results consistently showing significant reductions in power consumption per
operation. Detailed performance comparisons are available in studies hosted on IEEE Xplore
and summarized in technical reports on arXiv.
Performance metrics typically include:

• Energy Consumption: Neuromorphic processors often operate at power levels that are
several orders of magnitude lower than those of conventional systems, particularly for
tasks involving continuous data streaming.

• Processing Latency: Due to their event-driven nature, neuromorphic systems can

process information with lower latency, making them ideal for real-time applications.

• Scalability: The parallel architecture of neuromorphic systems allows them to scale

more effectively than traditional sequential processors. This is particularly important
for applications that require the simulation of large neural networks.
The comparative studies indicate that while neuromorphic systems may still face challenges in
terms of general-purpose computing capabilities, they offer clear advantages in specialized
applications that demand energy efficiency and adaptive learning. These advantages are well-
documented in the literature, providing a strong rationale for further development and
integration of neuromorphic hardware.

12
Neuromorphic computing
3.5 Challenges & Future Directions
Despite the significant promise of neuromorphic computing, several challenges remain that
must be addressed to facilitate broader adoption. One of the foremost challenges is the
programming complexity inherent in designing and optimizing spiking neural networks.
Unlike conventional machine learning frameworks, neuromorphic systems require specialized
programming models that can effectively manage the temporal dynamics of spiking neurons.
The lack of standardized development tools and frameworks has been noted in numerous
reviews, such as those available on IEEE Xplore.
Another challenge is the heterogeneity of hardware architectures. With multiple platforms such
as TrueNorth, Loihi, and SpiNNaker employing different design philosophies and operating
parameters, it is difficult to establish universal performance benchmarks. This fragmentation
hinders the interoperability and comparative evaluation of neuromorphic systems.
Looking ahead, future research is expected to focus on the development of hybrid architectures
that combine the strengths of neuromorphic and traditional computing. Such systems could
leverage the energy efficiency and parallelism of neuromorphic processors for specific tasks
while relying on conventional processors for more general-purpose computing. Moreover,
advances in emerging materials—particularly the use of memristors and other nanoscale
devices—are anticipated to further enhance the performance and integration of neuromorphic
systems. Research on hybrid models is discussed in detail in recent articles on Nature and IEEE
Xplore.
Additionally, the field is exploring novel learning paradigms that could improve the
adaptability and robustness of neuromorphic systems. This includes research into unsupervised
learning techniques and biologically plausible algorithms that more accurately capture the
nuances of neural processing. The convergence of neuroscience, materials science, and
computer engineering is expected to drive these innovations, paving the way for next-
generation computing systems that more closely mimic the efficiency and adaptability of
biological brains.
In summary, while neuromorphic computing has demonstrated substantial potential through its
energy efficiency, real-time processing, and adaptive capabilities, ongoing research is essential
to overcome its current limitations. Addressing challenges such as programming complexity,
standardization, and hardware heterogeneity will be critical for transitioning neuromorphic
systems from experimental prototypes to widespread industrial applications. The insights
gathered from current research not only highlight the transformative potential of neuromorphic
computing but also provide a roadmap for future investigations and technological
advancements.

13
Neuromorphic computing
 Chapter IV : Industry Applications & Challenges
Neuromorphic computing has transitioned from a primarily theoretical concept to practical
technology with promising applications across multiple industries. This chapter explores how
neuromorphic systems are being implemented in key sectors such as robotics, artificial
intelligence, edge computing, healthcare, and aerospace, while also addressing the challenges
that impede their widespread adoption.

4.1 Robotics & Automation

Neuromorphic computing plays a pivotal role in robotics and automation by offering highly
efficient real-time processing and adaptive control. The event-driven architecture of spiking
neural networks (SNNs) enables robots to process sensory information and respond to
environmental stimuli with minimal latency. For instance, researchers have demonstrated the
use of neuromorphic controllers to manage robotic motion control and manipulation tasks in
dynamic environments. These controllers adjust to changes in real time, enabling robots to
navigate complex terrains and perform precision tasks under uncertain conditions.
A notable study by Indiveri et al. (2011) (IEEE Xplore) highlights how neuromorphic circuits
can emulate neural behavior, leading to robust adaptive control in robotic systems. The parallel
processing capabilities and low power consumption of neuromorphic chips make them
particularly suitable for mobile and autonomous robots that must operate continuously on
limited battery power.

4.2 Artificial Intelligence & Machine Learning

The intersection of neuromorphic computing and artificial intelligence (AI) is one of the most
exciting areas of research today. Traditional deep learning models, while powerful, are often
constrained by high power requirements and the need for extensive computational resources.
In contrast, neuromorphic processors offer a more energy-efficient alternative by processing
information in an event-driven manner.
Research published in Nature Electronics (Nature Electronics) illustrates that neuromorphic
architectures can support unsupervised and adaptive learning, which is particularly beneficial
for AI applications that require continuous learning from streaming data. For example, the
adaptive capabilities of Intel’s Loihi processor (Intel Neuromorphic Computing) enable it to
adjust to new patterns in real time without the need for extensive retraining. This is essential
for developing AI systems that can operate in dynamic, real-world environments where
conditions change rapidly.
Neuromorphic AI systems are finding applications in areas such as pattern recognition,
decision-making, and anomaly detection, where the ability to process information in parallel
leads to faster and more robust responses.

14
Neuromorphic computing
4.3 Edge Computing & Internet of Things (IoT)
The deployment of neuromorphic computing in edge devices and IoT systems addresses the
growing need for local, real-time data processing while minimizing energy consumption. With
billions of sensors and connected devices in operation, transferring all data to centralized cloud
servers is both inefficient and impractical. Neuromorphic processors, with their ultra-low
power consumption and high responsiveness, offer a solution by enabling on-device processing
of sensor data.
For example, studies on energy-efficient neuromorphic hardware indicate that these systems
can dramatically reduce the power consumption of IoT devices, allowing them to operate
longer on limited energy sources. By processing data locally, neuromorphic systems can
facilitate rapid decision-making and reduce latency, which is critical for applications such as
environmental monitoring, smart home systems, and industrial automation.
The inherent parallelism of neuromorphic architecture also supports the simultaneous
processing of data from multiple sensors, making them ideally suited for complex IoT scenarios
where a multitude of inputs must be integrated and analyzed in real time.

4.4 Healthcare & Biomedical Applications

In the healthcare domain, neuromorphic computing is revolutionizing the way medical data is
processed and interpreted. One of the most promising applications is in the development of
brain–machine interfaces (BMIs), where neuromorphic systems can decode neural signals with
high accuracy and low latency. This capability is essential for controlling prosthetic limbs or
for neurorehabilitation therapies that require real-time feedback.
Research has demonstrated that neuromorphic processors can effectively handle the complex,
non-linear dynamics of biomedical signals. For instance, studies have shown improvements in
the analysis of electroencephalography (EEG) signals and other bio signals by employing
neuromorphic algorithms that mimic the brain’s processing techniques. Additionally, the low
power requirements of neuromorphic chips make them suitable for implantable devices where
battery life is a critical concern.
Beyond BMIs, neuromorphic systems are being explored for applications in medical imaging,
where their ability to process large volumes of data in parallel can lead to faster image
reconstruction and enhanced diagnostic accuracy. The adaptive learning capabilities inherent
in these systems also offer potential for personalized medicine, where treatment strategies can
be dynamically adjusted based on continuous patient monitoring.

4.5 Aerospace & Defense

The aerospace and defense sectors are increasingly turning to neuromorphic computing for
applications that require robust, low-latency data processing under extreme conditions.
Autonomous navigation systems, for instance, benefit from the real-time processing

15
Neuromorphic computing
capabilities of neuromorphic architectures, which enable unmanned aerial vehicles (UAVs) and
autonomous ground vehicles to react swiftly to changing environments.
Neuromorphic systems are also being investigated for target recognition and sensor fusion in
defense applications. The ability to integrate information from multiple sources and make rapid
decisions is critical for applications such as missile guidance and surveillance systems.
Research into neuromorphic hardware, such as Intel’s Loihi and IBM’s TrueNorth, indicates
that these systems can maintain high performance even under conditions of variable sensor
input and unpredictable environmental factors (Nature).
Furthermore, the energy efficiency of neuromorphic processors is a significant advantage in
aerospace applications, where power resources are limited and the reliability of systems is
paramount.

4.6 Challenges in Industrial Applications

Despite the promising applications of neuromorphic computing, several challenges hinder its
widespread industrial adoption:

• Programming Complexity: Developing algorithms for spiking neural networks

requires specialized knowledge and frameworks that differ substantially from
conventional deep learning platforms. The complexity of programming event-driven
architecture remains a barrier to rapid development and deployment.

• Standardization of Hardware: With various neuromorphic platforms such as

TrueNorth, Loihi, and SpiNNaker employing different design philosophies,
establishing standardized benchmarks for performance and interoperability is
challenging. This fragmentation complicates efforts to integrate neuromorphic systems
into existing industrial infrastructures.

• Scalability & Integration: While neuromorphic systems excel in energy efficiency

and parallel processing, scaling these architectures to support a wide range of
applications remains an ongoing challenge. Integration with conventional computing
systems, which dominate current industrial practices, requires the development of
hybrid models that leverage the strengths of both paradigms.

• Reliability & Robustness: In critical applications such as aerospace and healthcare,

system reliability and robustness are non-negotiable. Neuromorphic hardware, being a
relatively new technology, must undergo rigorous testing to meet the high standards
required for industrial deployment.

16
Neuromorphic computing
 • Economic Viability: The cost associated with developing and manufacturing
neuromorphic chips, particularly those that incorporate emerging materials like
memristors, can be prohibitive. Ensuring that these systems are cost-effective in the
long term is essential for their adoption in commercial applications.

17
Neuromorphic computing
 Chapter V : Future Scope & Opportunities

5.1 Emerging Trends

Recent advancements in neuromorphic computing have paved the way for several emerging
trends that are likely to shape the future of this field. One of the key trends is the integration of
novel materials, such as memristors, which have demonstrated significant promise for
replicating synaptic behavior with high energy efficiency. Basu et al. (2018) provide an in‐
depth discussion on the role of memristive devices in enhancing the scalability and
performance of neuromorphic systems (RSC Publications).
Another notable trend is the convergence of neuromorphic architectures with conventional
digital processors. Studies, including those highlighted in IEEE Xplore and Nature Electronics
indicate that hybrid systems—where neuromorphic co-processors are integrated with classical
CPUs/GPUs—can capitalize on the strengths of both paradigms. This integration is particularly
promising for applications requiring both high-performance data processing and energy
efficiency. Additionally, there is growing interest in developing more sophisticated spiking
neural network (SNN) models that incorporate biologically plausible learning rules, such as
spike-timing-dependent plasticity (STDP), to further improve adaptive learning capabilities in
real-world scenarios.

5.2 Opportunities for Innovation

The unique attributes of neuromorphic computing—namely its event-driven processing,
inherent parallelism, and low energy consumption—open up a broad array of opportunities for
innovation across several sectors. In robotics and autonomous systems, for instance, the ability
of neuromorphic processors to execute real-time adaptive control can lead to the development
of more agile and responsive machines. Research by Indiveri et al. (2011) underscores how
neuromorphic circuits can enable robust sensory integration and motion control, which are
critical for complex robotic applications.
In the realm of artificial intelligence, neuromorphic computing offers the potential for a
paradigm shift by supporting unsupervised learning and rapid adaptation. This is particularly
significant for edge computing scenarios, where devices must operate under strict power
constraints while processing data locally. Intel’s Loihi processor (Intel Neuromorphic
Computing) exemplifies how on-chip learning can be harnessed to build systems that
continuously adapt to new information without the need for extensive retraining.
Furthermore, neuromorphic systems are poised to revolutionize healthcare applications. Their
capability to process complex neural signals with minimal energy expenditure makes them
excellent candidates for brain–machine interfaces (BMIs) and real-time medical diagnostics.
The adaptive and parallel nature of these systems may lead to breakthroughs in personalized
medicine, where continuous monitoring and rapid decision-making are essential.

18
Neuromorphic computing
5.3 Roadmap for Future Research
To fully realize the potential of neuromorphic computing, several research directions and
milestones must be pursued:

• Standardization & Benchmarking:

There is a pressing need for developing standardized programming frameworks and
performance benchmarks. Current literature, including analyses available on arXiv,
emphasizes the importance of unified metrics for energy consumption, latency, and
scalability. Establishing common standards would facilitate better comparison among
various neuromorphic platforms (e.g., TrueNorth, Loihi, SpiNNaker) and accelerate
their integration into commercial applications.

• Hybrid Architectures:
Future research should explore hybrid models that combine the efficiency of
neuromorphic processors with the versatility of conventional architectures. Such
systems could, for example, offload energy-intensive tasks to neuromorphic co-
processors while maintaining general-purpose computing capabilities through
traditional hardware. This approach is anticipated to yield systems that are both highly
adaptive and broadly applicable.

• Interdisciplinary Collaborations:
Advancing neuromorphic computing requires the convergence of insights from
neuroscience, materials science, and computer engineering. Collaborative research
efforts can drive innovations in emerging materials, improve the design of
neuromorphic circuits, and refine biologically inspired algorithms. Cross-disciplinary
initiatives are crucial for addressing current limitations and pushing the boundaries of
what neuromorphic systems can achieve.

• Development of Robust Software Tools:

Creating user-friendly development environments and simulation tools is essential for
broadening the adoption of neuromorphic computing. Enhanced software support
would lower the barrier to entry for researchers and developers, enabling them to
design, test, and deploy neuromorphic algorithms more efficiently.

19
Neuromorphic computing
5.4 Industry Impact
The transformative potential of neuromorphic computing is expected to have far-reaching
implications across various industries. In robotics, the integration of neuromorphic controllers
can lead to more efficient, real-time processing systems that improve the safety and
effectiveness of autonomous machines. In the field of artificial intelligence, energy-efficient
neuromorphic processors offer a pathway to deploy intelligent systems in power-constrained
environments, such as mobile devices and IoT networks.
In healthcare, the ability to emulate brain-like processing could revolutionize the design of
prosthetics, BMIs, and diagnostic tools by providing rapid, adaptive responses to biological
signals. Aerospace and defense applications are also poised to benefit; neuromorphic systems
can enhance autonomous navigation, sensor fusion, and target recognition capabilities while
operating under stringent energy budgets.
Ultimately, as neuromorphic computing matures, its integration into hybrid computing
environments is expected to redefine how industries approach complex computational
challenges. By bridging the gap between biological efficiency and digital precision,
neuromorphic technologies promise to drive innovation and unlock new applications that were
previously unattainable with traditional architectures.

Fig. This graph illustrates the exponential expansion of the neuromorphic computing market,
projecting an increase from $1.44 billion in 2024 to $1.81 billion in 2025, corresponding to a
compound annual growth rate (CAGR) of 25.7%. The upward trend is driven by significant
advancements in artificial intelligence, the rise of cognitive computing applications, enhanced

20
Neuromorphic computing
real-time processing capabilities, and increased industry investment and collaborative
initiatives. This visual reinforces the report’s discussion on market trends and underscores the
economic momentum behind neuromorphic computing.

21
Neuromorphic computing
 Chapter VI : Results & Discussion

6.1 Key Findings

Recent investigations into neuromorphic computing have yielded a number of significant
findings that are consistently supported by the literature:

• Ultra-Low Power Consumption:

Research by Merolla et al. (2014) demonstrated that IBM’s TrueNorth chip integrates
over one million spiking neurons and 256 million synapses while consuming only a few
watts of power. This ultra-low power operation contrasts sharply with traditional digital
processors, which require substantially more energy for comparable tasks.

• Adaptive, Real-Time Learning:

The development of the Intel Loihi processor, as detailed by Davies et al. (2018), has
confirmed that on-chip, event-driven learning is achievable in hardware. Loihi’s
implementation of spiking neural networks (SNNs) enables systems to adjust synaptic
weights dynamically in response to input—mirroring the adaptive learning observed in
biological neural networks.

• Scalability Through Parallel Processing:

Neuromorphic systems leverage massively parallel architectures, allowing them to
handle numerous simultaneous operations. This characteristic is essential for tasks such
as complex sensory processing and real-time decision-making. Studies have shown that
the distributed processing nature of neuromorphic hardware scales efficiently as
network complexity increases.

• Energy-Performance Trade-Off:
Comparative studies indicate that neuromorphic processors can achieve orders of
magnitude lower energy per operation compared to conventional CPUs or GPUs. These
energy savings are particularly evident in applications that benefit from event-driven
computation, such as sensory data processing and autonomous control.

22
Neuromorphic computing
6.2 Comparative Analysis
Comparative analyses in the literature reveal several important performance metrics:

• Energy Efficiency:
Benchmarks reported in studies (e.g., see summaries on IEEE Xplore and arXiv)
consistently show that neuromorphic systems, like TrueNorth and Loihi, operate at
significantly lower power levels. Quantitative comparisons demonstrate that the energy
cost per neural spike is much lower than the energy used per operation in traditional
architectures.

• Latency & Throughput:

Due to their asynchronous, event-driven nature, neuromorphic systems exhibit reduced
processing latency. This translates into faster response times in real-time applications.
Additionally, the parallelism inherent in these architectures supports high throughput
when processing multiple data streams concurrently.

• Scalability:
The distributed architecture of neuromorphic systems allows them to maintain
performance as network sizes increase. Unlike conventional processors, where adding
more cores may introduce significant overhead, neuromorphic chips are designed to
scale with minimal incremental power consumption.

Fig. Comparison of the von Neumann architecture with the neuromorphic architecture.
These two architectures have some fundamental differences when it comes to operation,
organization, programming, communication, and timing, as depicted here

23
Neuromorphic computing
6.3 Discussion on Implementation Challenges
While the advantages of neuromorphic computing are clear, several challenges continue to
impede their broader adoption:

• Programming Complexity:
Developing and optimizing spiking neural networks requires a fundamentally different
programming approach than conventional deep learning frameworks. Current
ecosystems lack the extensive toolsets available for platforms like TensorFlow or
PyTorch, making it more challenging for developers to design and debug neuromorphic
applications.

• Lack of Standardization:
The diversity of neuromorphic platforms—such as TrueNorth, Loihi, and SpiNNaker—
each with their unique architectures and design philosophies, makes it difficult to
establish universal benchmarks and programming standards. This fragmentation can
hinder interoperability and comparative performance assessments.

• Integration with Existing Systems:

In industrial settings, integrating neuromorphic processors with conventional digital
systems remains a significant hurdle. Hybrid architectures, where neuromorphic co-
processors are coupled with traditional CPUs/GPUs, have been proposed; however,
efficient data exchange and task allocation between these different paradigms require
further research and development.

• Reliability & Robustness:

Ensuring that neuromorphic systems perform reliably under the varied and often
unpredictable conditions found in real-world applications (such as in healthcare or
aerospace) is critical. While laboratory experiments have demonstrated promising
results, the robustness of these systems under continuous, long-term operation still
requires thorough validation.

6.4 Energy Efficiency & Performance Metrics

Quantitative assessments of neuromorphic systems highlight their superior energy efficiency
and performance characteristics:

• Energy per Operation:

Measurements indicate that the energy required for a single neural spike in systems like
TrueNorth and Loihi is significantly lower than the energy per operation in
conventional processors. This metric is particularly relevant for battery-powered or
remote devices where energy resources are constrained.

24
Neuromorphic computing
 • Parallel Processing Benefits:
The asynchronous, event-driven processing of neuromorphic architectures minimizes
idle cycles and reduces overall latency. This efficiency is critical for tasks requiring
rapid decision-making, such as in autonomous navigation or real-time sensor fusion.

• Consistent Throughput Under Load:

Even as the complexity of the neural network increases, neuromorphic systems
maintain a high degree of parallelism. This allows for effective scaling when processing
large volumes of data, as demonstrated in various benchmarking studies published on
platforms like IEEE Xplore

6.5 Summary & Conclusions

In summary, the results discussed in this chapter underscore the transformative potential of
neuromorphic computing:
• Neuromorphic systems achieve remarkable energy efficiency and low latency through
their event-driven, parallel processing architecture.
• Comparative evaluations highlight that neuromorphic hardware, such as TrueNorth and
Loihi, significantly outperforms conventional digital systems in terms of energy
consumption and scalability.
• Despite these advantages, several challenges—including programming complexity, a
lack of standardization, integration issues, and the need for rigorous reliability testing—
remain to be addressed before neuromorphic computing can be widely adopted in
industry.
• The emerging trend toward hybrid architectures offers a promising pathway to leverage
the strengths of both neuromorphic and conventional systems, potentially leading to
new applications and enhanced performance in energy-constrained environments.

25
Neuromorphic computing
 Chapter VII : Conclusion & Recommendations

7.1 Summary of Key Points

The investigation into neuromorphic computing has illuminated several critical advantages and
challenges that shape its potential for transforming next-generation computing systems:
• Energy Efficiency and Ultra-Low Power Consumption:
Neuromorphic architectures such as IBM’s TrueNorth chip have demonstrated the
ability to integrate over one million spiking neurons and hundreds of millions of
synapses while consuming only a few watts of power (Merolla et al., 2014). This
exceptional energy efficiency is a stark contrast to the high power demands of
traditional digital processors. Intel’s Loihi processor further substantiates these findings
by showcasing adaptive, on-chip learning capabilities that maintain low energy usage
even under dynamic conditions (Davies et al., 2018).

• Adaptive and Real-Time Learning Capabilities:

The use of spiking neural networks (SNNs) and learning algorithms such as spike-
timing-dependent plasticity (STDP) (detailed in Indiveri et al., 2011) enables
neuromorphic systems to adapt in real time to varying input conditions. This adaptive
behavior is particularly valuable in applications requiring rapid response and
continuous learning, such as autonomous robotics, edge computing, and biomedical
signal processing.

• Scalability and Parallel Processing:

Neuromorphic computing systems inherently support massively parallel processing.
Their distributed architecture allows them to handle multiple data streams
simultaneously without incurring significant increases in power consumption or
latency. This scalability is critical for complex tasks such as multi-sensor integration
and large-scale neural simulations.

• Energy-Performance Trade-Offs:
Quantitative analyses reveal that neuromorphic systems offer substantial improvements
in energy per operation and processing throughput compared to conventional
architectures. These trade-offs are essential for applications where energy constraints
are paramount, such as in battery-powered IoT devices and implantable biomedical
devices.

• Challenges in Standardization and Integration:

Despite these compelling advantages, significant challenges persist. The programming
complexity, lack of standardized development frameworks, and difficulties in
integrating neuromorphic processors with existing digital systems remain major

26
Neuromorphic computing
 hurdles. These challenges impede the transition from laboratory prototypes to
commercial, industrial-scale applications.

7.2 Concluding Remarks

The transformative potential of neuromorphic computing lies in its ability to emulate the
adaptive and energy-efficient processing of the human brain. The research reviewed in this
report confirms that neuromorphic architectures provide a promising alternative to traditional
von Neumann systems. Key studies—such as Merolla et al. (2014) on TrueNorth and Davies
et al. (2018) on Loihi—demonstrate that these systems can significantly reduce power
consumption while offering robust real-time learning and processing capabilities. The inherent
parallelism of neuromorphic systems makes them especially well-suited for complex, data-
intensive tasks where traditional processors struggle to keep pace.
However, the path to widespread industrial adoption is not without obstacles. The diversity of
hardware platforms, each with its own unique design and operating characteristics, creates a
fragmented ecosystem that complicates standardization and interoperability. Furthermore, the
specialized programming models required for spiking neural networks pose a steep learning
curve for developers accustomed to conventional software frameworks.
These challenges underscore the need for a focused research agenda that addresses both the
technological and practical aspects of neuromorphic computing. Bridging the gap between
theoretical innovation and commercial viability will require concerted efforts in developing
standardized tools, hybrid architecture, and robust integration strategies.

27
Neuromorphic computing
7.3 Recommendations for Future Research
Based on the insights gathered from the literature, several key recommendations can be made
to advance neuromorphic computing research:
• Development of Standardized Programming Frameworks:
Create unified, user-friendly software ecosystems and simulation tools specifically
tailored for neuromorphic systems. By establishing common programming standards
and benchmarks—similar to those available for conventional deep learning (e.g.,
TensorFlow, PyTorch)—researchers can reduce the complexity of designing and
deploying spiking neural networks.

• Exploration of Hybrid Architectures:

Investigate models that seamlessly integrate neuromorphic co-processors with
traditional CPUs/GPUs. Hybrid systems could leverage the energy efficiency and
adaptive learning of neuromorphic processors for specialized tasks while relying on
conventional hardware for general-purpose computing. Research should focus on
developing efficient data exchange protocols and task allocation strategies to ensure
harmonious operation between these paradigms.

• Enhanced Material Research & Device Integration:

Continue exploring advanced materials—such as memristors—that can more
accurately emulate synaptic behaviors. The work of Basu et al. (2018) highlights the
potential of these devices to further reduce energy consumption and improve the density
of neuromorphic circuits. Investment in material science research will be critical for the
next generation of neuromorphic hardware.

• Interdisciplinary Collaboration:
Foster collaborations between neuroscientists, materials scientists, and computer
engineers. Such interdisciplinary initiatives can lead to more biologically plausible
models and innovative hardware designs, driving the field toward systems that better
mimic the adaptive and robust nature of the human brain.

• Robust Testing & Benchmarking:

Implement extensive testing protocols to evaluate the reliability and robustness of
neuromorphic systems under real-world conditions. Future research should aim to
develop standardized benchmarks for energy efficiency, latency, and scalability,
enabling more meaningful comparisons across different platforms.

28
Neuromorphic computing
7.4 Recommendations for Industrial Adoption
For industries considering the integration of neuromorphic computing, the following strategies
are recommended:
• Pilot Projects & Incremental Integration:
Start with pilot projects in controlled environments to assess the feasibility of
neuromorphic systems in specific applications. These pilots can serve as testbeds for
hybrid architectures, allowing companies to gradually integrate neuromorphic
processors alongside traditional systems before full-scale deployment.

• Investment in Skill Development & Training:

Develop targeted training programs to equip engineers and developers with the
necessary skills to work with neuromorphic systems. Bridging the knowledge gap will
be essential for the successful design, programming, and maintenance of these systems
in an industrial setting.

• Strategic Partnerships with Research Institutions:

Establish partnerships between industry and academic or research institutions to
accelerate the commercialization of neuromorphic technologies. Collaborative projects
can provide access to cutting-edge research, shared technological resources, and
innovative methodologies that benefit both sectors.

• Focus on Niche Applications Initially:

Identify niche applications where neuromorphic computing’s advantages—such as
ultra-low power consumption and real-time processing—offer clear benefits. Early
success in these targeted domains can build momentum and provide a proof-of-concept
for broader applications across the industry.

• Long-Term Investment in Infrastructure:

Prepare for long-term investments that include upgrading existing digital
infrastructures to accommodate hybrid computing environments. As neuromorphic
hardware evolves, industries will need to adapt their operational frameworks to fully
exploit the benefits of these advanced systems.

29
Neuromorphic computing
 Chapter VIII : References

1. Merolla, P. A., Arthur, J. V., Alvarez-Icaza, R., Cassidy, A. S., Sawada, J.,
Akopyan, F., ... & Modha, D. S. (2014). A million spiking‐neuron integrated
circuit with a scalable communication network and interface. Science, 345(6197),
668–673.
2. Davies, M., Srinivasa, N., Lin, T. H., Chinya, G., Cao, Y., Choday, S. H., ... &
Wang, H. (2018). Loihi: A neuromorphic manycore processor with on-chip
learning. IEEE Micro, 38(1), 82–99.
3. Indiveri, G., & Liu, S. C. (2015). Memory and information processing in
neuromorphic systems. Proceedings of the IEEE, 103(8), 1379–1397.
4. Basu, A., et al. (2018). Emerging memristive materials for neuromorphic
computing: A review. Advanced Materials.
5. ArXiv Preprint on Neuromorphic Benchmarking. (2020). Neuromorphic
Computing: Models and Applications. arXiv:2007.12253.
6. Nature Electronics. (2021). Neuromorphic Computing: A Paradigm Shift. Nature
Electronics.
7. Furber, S. (2016). SpiNNaker: A Spiking Neural Network Architecture.
[Information available on the SpiNNaker Project Website].
8. Benjamin, B. V., et al. (2014). Neurogrid: A Mixed-Analog-Digital Multichip
System for Large-Scale Neural Simulations. Proceedings of the IEEE, 102(5),
699–716.
9. Qiao, N., et al. (2015). A reconfigurable on-line learning spiking neuromorphic
processor comprising 256 neurons and 128K synapses. Frontiers in Neuroscience,
9, 141.
10. Chicca, E., & Indiveri, G. (2014). Neuromorphic Electronic Circuits for Building
Autonomous Cognitive Systems. Proceedings of the IEEE, 102(9), 1367–1388.
11. Schuman, C. D., Potok, T. E., Patton, R. M., Birdwell, J. D., Dean, M. E., &
Plank, J. S. (2017). A Survey of Neuromorphic Computing and Neural Networks
in Hardware. arXiv preprint arXiv:1705.06963.
12. Roy, K., Jaiswal, A., & Panda, P. (2019). Towards spike-based machine
intelligence with neuromorphic computing. Nature, 575(7784), 607–617.
13. Furber, S., et al. (2014). A Million Cores for the Human Brain? Frontiers in
Neuroscience, 8, 141.
14. Indiveri, G., et al. (2011). Neuromorphic silicon neuron circuits. Frontiers in
Neuroscience, 5, 73.
15. Schuman, C. D., et al. (2019). Neuromorphic Computing: From Materials to
Systems Architecture.

30
Neuromorphic computing