6th International Conference on Engineering Technology and its Applications 2023- (6thIICETA2023)

Neuromorphic Computing: Unraveling the Physical

Properties of Brain-Inspired Systems
1st G. Mohammed 2nd Waleed H. Madhloom 3rd Mustafa abdulsattar jebur
Ahl Al Bayt University Altoosi University College Al-Zahraa University for Women
Karbala, Iraq Najaf, Iraq Karbala, Iraq
[email protected] [email protected] [email protected]

4th Mohammed Brayyich 5th Myasar Mundher adnan 6th Waleed Hameed
2023 6th International Conference on Engineering Technology and its Applications (IICETA) | 979-8-3503-0333-9/23/$31.00 ©2023 IEEE | DOI: 10.1109/IICETA57613.2023.10351270

National University of Science and The Islamic university Medical Technical College
Technology Najaf, Iraq Al-Farahidi University
Dhi Qar, Iraq [email protected] Baghdad, Iraq
[email protected] [email protected]

Abstract— Neuromorphic computing, which aims to replicate

the information processing capabilities of the human brain, has I. INTRODUCTION
gained significant attention as a promising paradigm for The demand for computing complexity and massive data
developing intelligent computing systems. The physical processing may soon be restricted by the localised on-chip
properties and characteristics of neuromorphic systems play a power dissipation of existing microchips [1]. However,
critical role in their performance and functionality. In this
there is little question that the market growth for
paper, we provide an in-depth exploration of the physical
information functioning will lead to an unsustainable and
properties of brain-inspired systems, unraveling the key
aspects of neuromorphic computing from a physical
uncontrollable rise in the world's energy consumption. As a
perspective. We discuss the underlying physical principles of result, Moore's law would've been retarded, requiring the
spiking neural networks, including their operation principles study of new systems, design, unusual material, devices, and
and types. We also review synaptic plasticity mechanisms and activities, all in the framework of computers [2,3]. Some of
their implementations in hardware. Furthermore, we analyze the brain-like abilities that typical Display adapter artificial
various hardware implementations of neuromorphic nerve cells have the ability to provide include recognition
computing, ranging from traditional CMOS-based systems, identity, and improved sensing [3-5]. Because of
technologies to emerging technologies such as memristors,
insurmountable problems created by their materials
spintronics, and photonics. We discuss the advantages and
properties, traditional transistors offer limited hope for the
limitations of different hardware technologies for
neuromorphic computing, and highlight their potential in
future.
realizing brain-inspired functionalities. Moreover, we review Quantum substances and technology are expected to serve
applications of neuromorphic computing in diverse domains, as a surface, succeeding computational framework to
including artificial intelligence, robotics, image and speech construct a system that resembles a real brain. More basic
processing, pattern recognition, and sensor networks. We research into every aspect of the problem would be
provide examples and case studies showcasing the potential of extremely helpful at this point inside the area. This Special
brain-inspired systems in real-world scenarios. Additionally, Topics edition is devoted to assessing the current state of the
we discuss the challenges and limitations of neuromorphic
field and presenting novel, potentially game-changing ideas
computing, such as power consumption, scalability, and
[4,6,7] qualities.
hardware variability, and highlight ongoing research efforts to
address these challenges. Finally, we outline future directions
To better comprehend the characteristics of quantum
and research opportunities in neuromorphic computing, components, the effect of deficiencies, and their meeting the
including potential breakthroughs and emerging trends. expectations on gadgets and systems, a wide range of
Overall, this paper provides a comprehensive overview of the methods will be used, including ab initio numerical
physical properties of brain-inspired systems in the context of simulations combined with trying to cut synthesis as well as
neuromorphic computing, shedding light on the current state nanosized structural, electronics, magnetic, as well as
of research, advancements, and potential applications in this optical categorisation [8]. A complete, quantitative,
exciting field.
transdisciplinary analysis of the pertinent materials
Keywords— computing, neuro science, neuromorphic,
subjected to challenging (electricity, thermally, magnetism,
field, applications.
and mechanical) stresses is essential. A material's
architecture and how it impacts its physicochemical
characteristics may now be precisely controlled and

979-8-3503-0333-9/23/$31.00 ©2023 IEEE 654

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on April 23,2024 at 17:39:43 UTC from IEEE Xplore. Restrictions apply.
 6th International Conference on Engineering Technology and its Applications 2023- (6thIICETA2023)
determined thanks to advancements in modern synthesis and mechanisms, such as long-term potentiation (LTP) and long-
characterization technology [9]. term depression (LTD), and their implementations in
hardware, including both traditional CMOS-based
technologies and emerging technologies such as memristors,
spintronics, and photonics.

We then analyze various hardware implementations of

neuromorphic computing, discussing the advantages and
limitations of different technologies. We cover traditional
CMOS-based implementations, which are widely used in
neuromorphic systems, as well as emerging technologies
that offer unique features for brain-inspired computing. We
highlight how different hardware technologies can be
leveraged to realize brain-like functionalities and enable
efficient and scalable neuromorphic computing systems.

Fig 1: Biological neuron and its interconnection Next, we review applications of neuromorphic computing in
diverse domains, such as artificial intelligence, robotics,
New design concepts that go beyond conventional image and speech processing, pattern recognition, and
semiconductors based on the charges or even the spins of sensor networks. We provide examples and case studies
the electrons are being created thanks to a complete grasp of showcasing the potential of brain-inspired systems in real-
the material' characteristics. For example, Fundamental world scenarios, highlighting the advantages of
mechanics provides a means of simulating brain-inspired neuromorphic computing in solving complex problems and
technologies [10]. So although "neuronal" gadgets acquire advancing technological capabilities.
"metal phases" and adjust the conduction via metallic
filament growth, synaptic electronics encode a memory Furthermore, we discuss the challenges and limitations of
store by moving and altering the concentrations of faults neuromorphic computing, such as power consumption,
(e.g., oxygen vacancies). In particular, memristive processes scalability, and hardware variability. We highlight ongoing
and light-sensitive oxides are combining to produce research efforts to address these challenges, including novel
innovative design concepts for memories, networks, and hardware designs, algorithms, and system-level
central nervous system [11]. optimizations. We also outline future directions and research
Neuromorphic computing is an interdisciplinary field of opportunities in neuromorphic computing, including
research that aims to replicate the information processing potential breakthroughs and emerging trends that can shape
capabilities of the human brain in computing systems. the field in the coming years.
Inspired by the complex and dynamic nature of the brain,
neuromorphic systems utilize physical properties and In conclusion, this paper provides an in-depth overview of
characteristics to emulate brain-like functionalities, such as the physical properties of brain-inspired systems in the
pattern recognition, learning, and adaptation. Understanding context of neuromorphic computing. By unraveling the
the physical aspects of neuromorphic systems is crucial for physical aspects of neuromorphic systems, we aim to
their design, implementation, and optimization. provide insights into the current state of research,
In this paper, we provide an in-depth exploration of the advancements, and potential applications of neuromorphic
physical properties of brain-inspired systems in the context computing. This paper serves as a comprehensive resource
of neuromorphic computing. We start by discussing the for researchers, students, and practitioners interested in
underlying physical principles of spiking neural networks, understanding the physical foundations of brain-inspired
which are a class of neural network models that mimic the systems and their implications for the field of neuromorphic
way neurons in the brain communicate through electrical computing.
spikes or impulses. We delve into their operation principles,
types, and how they can be implemented in hardware to II. RELATED STUDY
achieve brain-inspired information processing. With a thorough understanding of the material properties,
new design ideas that go beyond conventional
Synaptic plasticity, the ability of connections between semiconductor depending on the charges or the spinning of
neurons to change their strength based on the activity and the electrons are being developed. For example,
history of the neurons, is a key aspect of neuromorphic Fundamental physics provides a method for simulating
computing. We review different types of synaptic plasticity technology based on the human brain. As "neurons in the

655
Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on April 23,2024 at 17:39:43 UTC from IEEE Xplore. Restrictions apply.
 6th International Conference on Engineering Technology and its Applications 2023- (6thIICETA2023)
brain" devices acquire "metal phases" and regulate
conductivity through metallic filament growth, synapses
encode memories by shifting and altering the density of
defects (e.g., ox-ygen vacancies). In general, intriguing
design ideas for memories, networks, and central nervous
system are emerging from memristor occurrences in
combination with luminous oxides. Synapse activity, which
requires flexibility in the regress or of materials, can be
achieved via thermoplastic transistors with many functions,
as well as optoelectronic and metal - oxide systems [12].

On the reverse hand, twist devices profit from substantial

semi in atomic components' magnetic properties. Because of Fig 2: Working of artificial neural network
the low-energy adaptability and quasi of magnetism, The reset power would inevitably increase, so some
“technologically human brains can be modified to create materials science ought to be carried out to take that into
reduced networks, and magnet caused by abnormal have consideration (Fig.3).
been suggested as a method of network training. Nanofibers
can also be made using magnetic microscopic phenomena.
Superconductors Jo-sephson devices can generate moment
responses that simulate synaptic and branching trees in Normalized mean square error
addition to brain spikes. Numerous phenomena, like in-
memory computation and reservoir computation, are being
researched as potential ways to replicate the amazing
functions carried by the mind [13]. However, the projecting segment, which is parallel towards
Emergent behaviour is at the core of biological neural the amorphous region segment when read, primarily
networks' and synapses' intricacy and remarkable regulates the gadget resistance.
effectiveness. Many identity principles are demonstrated by
the concepts of quantum nanomaterials and associated
nonlinearities, which allow for an array of dynamic as well
as static modes helpful for simulating sophisticated brain
systems and networks. A diversity of different memory
states, good of critical behaviour, and the formation of
unique collective occurrences are possible because to the
quantum superposition’s provided by the diverse energy
topography of the quantum state. It is crucial to consider the
system where the devices live in when tackling the problem
of building a neurologically machine [14]. The net is the
systems, to borrow a phrase from Herbert Irrespective and
apply it to neuromorphic computing inspired by biology.

III. PROPOSED METHODOLOGY

Fig.3 Proposed Model

A promising method for correcting PCM areas of mutual
like 1/f distortion and drifting is projecting sequence
storage. In these systems, the length of includes applications IV. RESULT AND DISCUSSION
is parallel to a piece of quasi projection. Using the The reconfigurable volume of the PCM device is scaled in
extremely nonlinear I-V characteristics of phase-change an ongoing effort to lower the RESET current. Either by
semiconductors, one may make absolutely sure that perhaps reducing the device's overall size in a constrained geometry
the projected segment seems to have no impact on the or by enlarging a mushroom-shaped device's bottom
device's performance during the Frame procedure (Figure electrode dimensions. The scaling problem might be helped
2). by the investigation of novel material classes like antimony
with a single element. With multi PCM synaptic topologies,
the restricted endurance and several other non-idealities

656
Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on April 23,2024 at 17:39:43 UTC from IEEE Xplore. Restrictions apply.
 6th International Conference on Engineering Technology and its Applications 2023- (6thIICETA2023)
related with the accumulative behaviour, such as limited closing the gap to all-photonic chip-scale information
dynamic range, nonlinearity, and stochasticity, can be processing, in addition to standard electrical PCM devices.
largely avoided. Recently, a multi-PCM synaptic design that The analogue multiplication of an incoming optical signal
uses a successful counter-based arbitration system was by a scalar value encoded in the state of the phase change
proposed (Figure 4). material was accomplished by integrating phase-change
materials onto an integrated photonics device. It was also
demonstrated that convolution operations might be carried
out in a single time step by taking advantage of wavelength
division multiplexing. This opens up the possibility of
designing phase-change materials that transition between the
crystalline and amorphous phases more quickly and with
greater optical contrast (Figure 5-7).

Fig 4: Representation of Boolean transformation

However, more study on the impact of device geometries as
well as the randomness connected to crystal development is
needed to optimise the accumulation behaviour at the device
level. Fig.2 shows the number of neurons calculated by the
proposed in model in different time interval.

Fig 7: NMSE as a function

IV. CONCLUSION
In-memory computing can take advantage of the non-
volatile binary storage, analogue storage, and accumulative
behaviour associated with PCM devices (IMC). The main
benefits of PCM over other non-volatile memory
technologies are its easy embeddability in a CMOS
substrate, volumetric switching, and well-understood device
physics. To enable PCM-based IMC, however, a number of
device and fabrication-level difficulties must be resolved,
and this is a hot area of research. Observing how PCM-
based neuromorphic computing is eventually marketed will
Fig.5 number of neurons calculated by the proposed in model in
also be intriguing. Prior to true IMC, a hybrid architecture
time
that uses stacked logic chips for processing and PCM
memory chips to store synaptic weights in a non-volatile
way.

REFERENCES

[1] Najmaei, Sina, et al. "Advancements in materials, devices, and

integration schemes for a new generation of neuromorphic
computers." Materials Today (2022).
[2] Stenning, Kilian D., et al. "Adaptive programmable networks for in
materia neuromorphic computing." arXiv preprint
arXiv:2211.06373 (2022).
Fig 6: Analysis of Number of inputs with time [3] Stenning, Kilian D., et al. "Adaptive programmable networks for in
Photonic memory devices based on phase-change materials, materia neuromorphic computing." arXiv preprint
arXiv:2211.06373 (2022).
which can be written, erased, and read optically, are quickly

657
Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on April 23,2024 at 17:39:43 UTC from IEEE Xplore. Restrictions apply.
 6th International Conference on Engineering Technology and its Applications 2023- (6thIICETA2023)
[4] Prosandeev, Sergey, et al. "Hidden phases with neuromorphic
responses and highly enhanced piezoelectricity in an antiferroelectric
prototype." Physical Review B 105.10 (2022): L100101.
[5] Yu, Jinran, et al. "Bioinspired interactive neuromorphic
devices." Materials Today (2022).
[6] Liu, Xiwen, Keshava Katti, and Deep Jariwala. "Accelerate &
Actualize: Can 2D Materials Bridge the Gap Between Neuromorphic
Hardware and the Human Brain?." arXiv preprint
arXiv:2301.10277 (2023).
[7] Goossens, Anouk S., et al. "Memristive memory enhancement by
device miniaturization for neuromorphic computing." arXiv preprint
arXiv:2301.03352 (2023).
[8] Bradley, Hannah, et al. "Artificial neurons based on
antiferromagnetic auto-oscillators as a platform for neuromorphic
computing." AIP Advances 13.1 (2023): 015206.
[9] Zins, Noah, et al. "Neuromorphic Computing: A Path to Artificial
Intelligence Through Emulating Human Brains." Frontiers of
Quality Electronic Design (QED) AI, IoT and Hardware Security.
Cham: Springer International Publishing, 2023. 259-296.
[10] Wang, Wei Sheng, and Li Qiang Zhu. "Recent advances in
neuromorphic transistors for artificial perception applications:
FOCUS ISSUE REVIEW." Science and Technology of Advanced
Materials 24.1 (2023): 10-41.
[11] Zhai, Yongbiao, et al. "Reconfigurable 2D-ferroelectric platform for
neuromorphic computing." Applied Physics Reviews 10.1 (2023):
011408.
[12] Kim, Joon Pyo, et al. "Dielectric-Engineered High-Speed, Low-
Power, Highly Reliable Charge Trap Flash-Based Synaptic Device
for Neuromorphic Computing beyond Inference." Nano
Letters (2023).
[13] Nilsson, Mattias. Event-Driven Architectures for Heterogeneous
Neuromorphic Computing Systems. Diss. Luleå tekniska universitet,
2023.

658
Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on April 23,2024 at 17:39:43 UTC from IEEE Xplore. Restrictions apply.