2022 IEEE 3rd Global Conference for Advancement in Technology (GCAT)

Bangalore, India. Oct 7-9, 2022

Evolution of Neuromorphic Computing with

Machine Learning and Artificial Intelligence
Ishu Sharma Vanshika
Chitkara University Institute of Engineering and Technology Chitkara University Institute of Engineering and Technology
Chitkara University, Chitkara University,
Punjab, India Punjab, India
[email protected] [email protected]

Abstract - In the present century, where artificial conditions with its biological structure and constricted
intelligence and machine learning are regenerating the world, architecture of neurons, similarly, neuromorphic chips are
this paper takes inspiration from brain intelligence and designed having behavior to adapt to all changes in the
explores new advanced computing known as Neuromorphic system and give the output by changing accordingly. Thus,
Computing. The energy efficiency and accuracy of the brain every data is maintained in the proper way and memory and
are remarkable and its retaining ability and grasping tendency processing can be done simultaneously. Therefore,
are overwhelming. Motivated by this, our paper discusses this
2022 IEEE 3rd Global Conference for Advancement in Technology (GCAT) | 978-1-6654-6855-8/22/$31.00 ©2022 IEEE | DOI: 10.1109/GCAT55367.2022.9971889

neuromorphic computers are replacing traditional computers

benchmarking technology i.e. neuromorphic computing and
and provide benefits like stochasticity, plasticity, fine-grained
how it is optimizing machine learning techniques by developing
parallelism, fast-paced learning, rapid response, and higher
Spiking Neural Networks (SNNs). Neuromorphic computing is
constructing computers that have a commonality like the
adaptability [2].
human brain and have all biological functionalities. This
research paper discusses the recent neuromorphic systems and
also presents the comparative analysis of the neuromorphic
systems in which various research areas are discussed like
image retrieval, forecasting, prediction and classification.
Machine learning algorithms are applied everywhere
nowadays and neuromorphic computing enables machine
learning techniques to work in an efficient manner. This paper
shows the power of neuromorphic computing as an essential
epitome to existing Von-Neumann architecture and also aims
to make the industry’s future bright with more research work
in this research area.

Keywords - Neuromorphic Computing; Machine Learning;

Fig. 1. Architecture of neuromorphic computing
Von-Neumann Architecture; Spiking Neural Networks (SNNs)

I. INTRODUCTION In figure1, the basic architecture of the neuromorphic

system is explained resembling the human brain and nervous
In short term, Neuromorphic Computing can be described system. Neuromorphic computing has great significance in
as Brain-inspired Computing i.e., it is the computing method Artificial Intelligence and Machine Learning as it will
which is inspired by biological structures of the human brain overcome all the challenges of the long-established
and nervous system and this approach to design the architecture like power consumption, storage capacity and
semiconductor chips with Brain-inspired computing low efficiency and will allow computers to execute brain-
methods. This phenomenon is a great replacement for the like-performance and activities. Machine Learning can be
von-Neumann architecture of computer systems and invent described as the techniques enabling computers to have the
new systems having a condensed architecture which can just ability to learn and achieve self-learning power. With the
behave like the human brain [1]. Traditional neural networks rapidly rising bulk of data, machine learning has become a
have low power and low efficiency but by introducing very important aspect nowadays in the industry and it is also
neuromorphic computing, neuromorphic systems can be mandatory for solving problems in various departments like
developed which have the fast computation, low power price and load forecasting, computational biology, finance,
consumption and are more efficient. Neuromorphic robotics, healthcare, industry 4.0 etc. Like ways,
Computing has its significance in the past also as it was first Neuromorphic chips would increase the use of machine
discovered by Carver Mead in the 1980s, some discovery has learning by enabling the artificially intelligent system to
been done on it, but nowadays neuromorphic computing is work in a well-organized manner. The relatability of machine
the utmost point of research for maintaining the future of learning with neuromorphic computing will be discussed
Artificial Intelligence (AI) and Machine Learning (ML). later in the paper.
In contemporary architecture, either performance or In Section II, neuromorphic computing and machine
power is kept at the priority over the other i.e., only one thing learning are discussed in detail and its correlativity that how
is maintained at one time but by using this advanced neuromorphic computing is enabling machine learning
technology, neuromorphic chips deliver both performances algorithms to work efficiently is also discussed in this
as well as low power, this is the main merit of neuromorphic section. The selected research work for advancements in
systems. Our Brain has the capable nature to adapt to all machine learning with neuromorphic computing is detailed in
circumstances, it also changes to external environment

978-1-6654-6855-8/22/$31.00 ©2022 IEEE 1

Authorized licensed use limited to: The National Institute of Engineering. Downloaded on October 15,2024 at 11:03:23 UTC from IEEE Xplore. Restrictions apply.
Section III. Comparative analysis of the discussed research are invented so far are INTEL’S LOIHI [5] , BSS2
work and results are given in section IV. The research paper PROTOTYPE [6], IBM’S TRUENORTH. Loihi chip was
is concluded in section V with a discussion on future the neuromorphic chip excogitated by Intel replacing the
directions. conventional chip system and it proves to provide more
results and performance as it will give support to the spiking
II. WHY NEUROMORPHIC COMPUTING FOR MACHINE neural networks (SNN). Loihi also solves the many ranges of
LEARNING? the problems and delivers quantitative computation, lower
In this section, Machine learning techniques and energy consumption, feature recognition, and precise
neuromorphic computing technology are discussed to relationships as it has been made by integrating 131 072
elaborate on how these techniques are beneficial. The neurons into the small mesh type container allowing spike
neuromorphic systems are enabling the efficient messages to operate and carry events independently putting
implementation of ML techniques. Researchers have no pressure on the framework. Loihi’s merits over the
productively employed neuromorphic technology in machine contemporary neuromorphic chips validate the progress in
learning models to overcome the challenges faced in the neuromorphic computing architecture overtaking the von-
contemporary architecture. This section also summarizes the Neumann architecture [5].
brain-inspired chips discovered to promote research in BrainScaleS2 Prototype (BSS2) is also a Neuromorphic
neuromorphic computing. chip manufactured using 65nm CMOS -process ASICs and
A. Machine Learning Methodologies designed using Plasticity Processing Unit (PPU) inspired by
the pong-videogame having fixed-pattern noise, particularly
As the term specifies, Machine Learning techniques are mimicking the brain’s form and dynamics., also having three
the processes which use mathematical models of data to orders of magnitude more efficiently. Reinforcement
make computers learn and work by themselves without Learning was deployed in biological Spiking Neutral
proper instruction and are at the core of Artificial Intelligence Networks using the Reward Modulated Spike-Timing
and Data Science. The adoption and enhancement of Dependent Plasticity (R-STDP) learning rule and it
machine learning algorithms are found throughout many reproduces Pavlovian conditioning on the neuromorphic chip
walks of life and allow more prominent decision-making in to obtain real-time speed [6]. Using R-STDP statistical
every aspect. Machine Learning is different due to its own learning rule, one form of discrete weight updates is
identity and its terminologies and is applicable in many employed:
domains like molecular biology, computer vision and text
processing. In today’s trend, we are discovering advanced ο࢝ࢇ࢈ ൌ ࢻǤ ሺࡴ െ ࢎሻǤ ࢘ࢇ࢈ (1)
technology i.e., Neuromorphic Computing which enables where ߙ is the learning rate, ‫ܪ‬is the reward, ݄ is the
Machine learning algorithms to work more efficiently and baseline and ‫ݎ‬௔௕ is the Spike-Timing Dependent Plasticity
replace the stereotypical architecture of computers which will (STDP) eligibility trace as the result of the post and
be discussed in later parts of Section 2. presynaptic spikes of synapse connecting neurons ܽ and ܾ.
The machine learning paradigm is used in probabilistic Optoelectronic Synapses are one of the cheering
Information retrieval; many neural networks (Hopfield applications for artificial intelligence which were proposed
Network), symbolic learning Implementation (ID3 AND from all-inorganic perovskite nanoplates. For achieving the
ID5R) and genetic algorithms are applied [3]. Also, Text promising approach of neuromorphic computing,
classification methodologies are also applicable in the micro- optoelectronic synapses are discovered that could process
text (chat post) which are very smaller in length and have visual signals, retain all the data, have lower power
informal grammar and language allowing point-to-point consumption, and detect the historical optoelectronic
conversation via any protocol which conveys important information whenever needed. Backtracking function is one
information within many military forces at a small pace of main function in the optoelectronic synapses which allow to
time. One more operation of machine learning algorithms is see back the information of the history when it is required,
that it is used to predict Coronary artery calcification using a this is the key benefit of the optoelectronic perovskite
sibship-based design [4]. Hence, machine learning is setting synapses [7].
its remarkable progress in the present century and will be at
its peak with new advancements. In this research paper, we
are demonstrating the new concept i.e., Neuromorphic
Computing which will become the future of AI and ML.
B. Neuromorphic Computing – Next Generation Computing
Neuromorphic Computing represents the computing
methods resemblance to the human brain and nervous system
and replaces the old armature of the computer system with a
new configuration which is tomorrow’s demand in the
industry. Pre-predictions have already been made by Gartner
that old computer systems will be soon at the edge by 2025,
evoking the new paradigm i.e., neuromorphic computing. As
the demand for new computing systems is heading upward,
Fig. 2. Optoelectronic Perovskite Synapses For Neuromorphic Computing
so it is necessary to look up more advancements and
neuromorphic computing can be the best bet forward. Till In figure 2, a general representation of the optoelectronic
now very few researchers can demonstrate that neuromorphic synapses is shown which are made from the perovskite
chips are more efficient, some of the neuromorphic chips that nanoplates like polystyrene sulphonate (PEDOT: PSS),

2
Authorized licensed use limited to: The National Institute of Engineering. Downloaded on October 15,2024 at 11:03:23 UTC from IEEE Xplore. Restrictions apply.
thiocyanate (CuSCN) and the backtracking function is also TensorFlow materials. Some models are discovered using
explained in this which helps in finding the previous data. NengoDL liking Spiking MNIST having all features,
therefore NengoDL allows the use of all neural networks and
C. A Stochastic Approach To Machine Learning With Nengo methods.
Neuromorphic Computing
In paper [12], content-based image retrieval is discussed
Machine learning algorithms have turned to be the best which is finding the similarity between the query image and
techniques for providing the endless performance in the dataset of product images using the Loihi chip as discussed
ample of applications and scenarios. With the advent of in paper [5]. In content-based image retrieval, visual search is
neuromorphic systems, machine learning algorithms can be done using neuromorphic neural networks and SNNs are
implemented in an energy-efficient manner. The new trained on Fashion-MNIST (image classification dataset). By
framework is discovered by the scientists which will have doing this, the nearest visual neighbors of the query image
more SNN (Spiking Neural Networks) and charge pump are obtained using the neuromorphic chip Loihi considerably
discharge strategies that increase the workload and design the showing the power efficiency of neuromorphic systems. This
neuromorphic chips which will have more appropriate will also accelerate the process by providing 12.5 times more
reliability and give workload-specific performance [8]. energy than GPU and 2.5 more times than CPU. Thus, it
IBM’s TrueNorth Neuromorphic system was designed using shows that by using neuromorphic systems, machine learning
16 TrueNorth neuromorphic processors and this system applications can become more reliable.
called NS16e is an advanced structure invented to make ML
techniques more reliable and scalable. NS16e has just like One more application of machine learning i.e., short-term
brain computation having neurons and synapses and gives wind power forecasting can be enhanced using neuromorphic
output like the brain comparing to Von-Neumann computing. This paper builds up more ML algorithms using
architecture. In the field of radar, NS16e proves to be the best neuromorphic models for efficient forecasting of wind power
armature for finding the direction, distance and sending the and uses NengoDL software [11] for making deep spiking
pulses of radio waves [9]. neural networks by ANN(Artificial Neural Network) -SNN
conversion. SNNs are made more useful by overcoming all
Inspired from the brain and by using SNN’s, the Neu the challenges by using the error backpropagation algorithm
Cube framework is established through which new data and ANN-to-SNN algorithm and making them more capable
machines are invented known as STDM (Evolving Spatio- of forecasting wind [13].
Temporal Data Machines). eSTDM is the promising
approach used for predictions in all fields like stroke In paper [14], lower energy efficiency maintenance is
prediction, personalised event prediction, and ecological discussed that how we can use system software for lowering
predictions. These evolving connectionist systems can be the energy used in the neuromorphic system for solving the
used for new knowledge discovery and for the better machine learning problems and applied to fields like
acquiring of all the input and output information [10]. In Embedded systems and Internet-of-Things (IoT). Thus, it
these spatial data machines, it is very important to learn all provides the clear recognition that system software helps to
patterns correctly and data streams, for this data machine divide the SNN model into clusters, mapped clusters into
requires SSTD (Spatio or Spectro-temporal data). SSTD is neurosynaptic cores and arrangement of neurons and
synapses into the core. This will lower the energy level by
ࡱǣ ࡭ሺο࢚ሻ  ՜ ࡮ (2) 20% making the system more considerable for machine
In equation 2, A(t) = (ܽଵ (t), ܽଶ (t),……...ܽ௡ (t)) and learning problems. The author also discusses the heuristic-
t=1,2,3,4…..,݉ where ‫ ܣ‬is a set of independent variables, ‫ܤ‬ based mapping approach to arrange neurons and synapses in
is the set of dependent output variables, ‫ݐ‬is the time window such a manner that would take lower energy consumption.
and ‫ ܧ‬is the linked function between whole input data. Thus, Authors in paper [15] mention about PyCarl which is
by using neuromorphic computing systems like Neu Cube PyNN interface that increases the neuromorphic development
system, more machine learning data machines are established cycle by allowing the combination of both machine learning
which can do predictions, retrieve data and information, and models and code sharing. In this PyCarl tool, CARL-sim
give output accordingly. SNN simulations are done using biological details of neurons
Evolution Of Machine Learning With Neuromorphic and synapses and hardware-oriented simulations. It allows all
Computing the users to apply machine learning models and
neuromorphic hardware early during implementation. This
In this section the recent research work for the evolution will take the field of neuromorphic computing at its peak.
of machine learning techniques or applications with
neuromorphic computing is elaborated. In the paper [16], the author mentions SpiNNaker which
is the programmable neuromorphic platform in which a
NengoDL is discussed in the paper [11], which is a machine learning algorithm like the k-NN algorithm can be
software tool designed for the implementation of SNN-based applied. This article tells that K- nearest neighbor search can
algorithms. The main aim of NengoDL is to allow all the be performed with more accuracy absorbing low power
users to use this framework to construct hybrid models and particularly for the classification tasks. This platform allows
form dynamic neural networks combining both deep learning the simulation of large-scale spiking neural networks.
as well as neuromorphic systems. NengoDL offers higher Classification task on the Irish dataset is also performed
speed and allows more models to run at the same time i.e., using SpiNNaker platform and the output obtained is correct
inducement of the model is done once keeping all different with higher accuracy it is also seen that on an increasing
models in parallel rather than running all different models in number of K values, k-NN algorithm gives more correct
a row taking a long time. NengoDL has some prominent output. Software simulation is done using NEST to run
features in comparison to other tools like forming of spiking SpiNNaker platform.
versions from deep learning networks, and the insertion of

3
Authorized licensed use limited to: The National Institute of Engineering. Downloaded on October 15,2024 at 11:03:23 UTC from IEEE Xplore. Restrictions apply.
 Machine learning has its ample applications, and these arrays and PARC (Pattern Recognition Engine) are used for
are solved at fastest pace with neuromorphic computing one recognizing chemicals. Olfactory systems made from these
such is explained in paper [17] which is in cheminformatics are used for wide purposes like in chemical factories, and
for band gap prediction and classification of chemicals. In wine industries. The Authors also discuss the data-to-event
this, neuromorphic simulation software CrossSim and encoding using AERO, signal conditioning and pre-
NeuroSim are used for classifying materials and predicting processing.
the band gap of small-molecule organic semiconductors. The
state-of-the-art contemporary architecture uses Dynamic One more architecture of the multi-core neuromorphic
Random Access Memory (DRAM) to obtain the results, but processor chip is discussed in paper [20] which is particularly
crossbar neuromorphic circuitry allows to do parallel work used for the classification of visual signals in DVS (dynamic
and does not use DRAM. This energy-efficient neuromorphic vision sensor) to manage the large traffic in the system. This
system is suitable for performing such tasks in chemical architecture called Dynamic Neuromorphic Asynchronous
organics providing low power and consuming less energy. In Processors (DYNAP)s has different neurons and synapses
this work, the author also aims for more fabrication of routing methods like memory optimized routing method, and
neuromorphic circuits to be specially designed to meet the programmable Address Event Representation (AER) routing
industry requirements for chemical purposes. to minimize memory needs and maximize the
programmability which supports a wide range of real-life
In paper [18], deep medical image analysis is discussed applications. It particularly has 1k Very Large-Scale
which is used for doing detection of diseases to give Integration (VLSI) neurons distributed among neurons and
outcomes and for this, machine learning technique combines routers for meeting these demands. This paper particularly
with neuromorphic computing system like loihi chip to aims at building such neuromorphic systems which have on-
overcome challenges like having less-labelled images, lack chip heterogeneous memory structures and have the
of advanced technology like magnetic resonance imaging. capability to reduce the memory requirements and construct
Some learning methods are described in it like spatial different network methodologies.
representation learning, and transfer learning for providing
the correct output from the large-scale database. It also In paper [21], Design-Technology-Co-Framework
resolves the issue of time management of Magnetic (DTCO) is developed which is constructed for performing
Resonance Imaging (MRI) as it is a long-lasting test taking machine learning algorithms and this is an essential
ample time. Therefore, this benchmarking technology will framework for inference accuracy robustness. It represents
enable countless machine learning applications to apply by how the ML algorithm, emerging memory Verilog-A models
overcoming all challenges. and the circuit interacts to obtain the best output, this is done
using SPICE and PYTHON packages. Also, an example,
The authors in the paper [19], discuss the e-nose systems ReRAM neuromorphic circuit is also studied in which
which have been encouraged using machine learning robustness of Neural Network (NN) is implemented for digit
algorithms. Neuromorphic SNNs are used for doing recognition and in ReRAM neuromorphic circuit, it was
classification of malts and producing olfactory systems found by the author that by using DTCO framework deeper
consuming less energy offering real-time processing giving neural network has less robustness making the neuromorphic
97% more accuracy. In this, the neuromorphic approach of circuit more robust. In table 1, the comparative analysis of
Akida SNN and Address Event Representation Olfaction the studied Neuromorphic computing systems has been
(AERO) encoder is used for the classification of malts in presented.
olfactory systems. Using the neuromorphic approach, sensor

TABLE 1. COMPARATIVE ANALYSIS OF NEUROMORPHIC COMPUTING SYSTEMS

Research Neuromorphic Machine Learning

Research Area Tool Used
Paper Computing System Technique
Pattern recognition; early event prediction on
Neural networks, fuzzy
NeuCube neuromorphic EEG data, audiovisual data, seismic data;
[10] ModuleM1 systems
framework neurorehabilitation, stroke prediction,
radioastronomy
Classification of visual signals in dynamic Convolutional neural
[20] DYNAPs Not Mentioned
vision sensor (DVS) networks
BrainScaleS 2
Reinforcement learning using
[6] neuromorphic prototype Radar beam shaping NEST V2. 14.0 SNN
STDP
(BSS2 prototype)
Deep learning training
[11] Spiking MNIST Embedded systems, Internet-of-Things NengoDL methods, convolutional
networks
Ultra-Low power machine
learning techniques,
[9] IBM’s NS16e Optical Communication Not Mentioned
Convolutional neural
networks
Deep learning approaches for
Recognition and Learning, Closed-loop control
[5] Loihi NxSDK SNNs, online approximation
for robotics
for backpropagation
Crossbar neuromorphic Band gap prediction, predicting pharmaceuticals Regression tasks, neural
[17] CrossSim
circuitry properties for drug-like compounds networks
ReRAM neuromorphic SPICE and Python
[21] Hand-written digit image classification Complex mapping algorithm
circuit packages

4
Authorized licensed use limited to: The National Institute of Engineering. Downloaded on October 15,2024 at 11:03:23 UTC from IEEE Xplore. Restrictions apply.
 [16] SpiNNaker Embedded systems, Classification tasks NEST K-nearest neighbor algorithm

Classifier training, learning

Higher accuracy classification of malts in e-nose
[19] Akida SNN architecture PCNose using STDP, K-nearest
systems
neighbor algorithm

[12] Loihi Content based image retrieval NxSDK K-nearest neighbor algorithm

[13] Loihi Short-term wind power forecasting Nengo Backpropagation algorithm

computing is very promising and can solve numerous

III. RESULTS AND DISCUSSION computation research problems.
The machine learning and artificial intelligence are
impacting most of the industries by providing highly IV. CONCLUSION AND FUTURE SCOPE
improved solutions. The industries ranging from In our paper, we have discussed the key advantages of the
manufacturing to banking, all are targeting to take decisions neuromorphic computing as it is arising as the utmost
based on the probabilistic results achieved by machine technology in the industry and replacing the conventional
learning intelligence. The advancement in neuromorphic architecture, also it is emerging as the future of artificial
computing can provide optimal solutions with low intelligence by building neuromorphic systems and allowing
processing power and less consuming time. In this paper, we machine learning algorithms to work in an excellent manner.
discussed the recent research work for the evolution of This research article covers the applications of neuromorphic
machine learning with neuromorphic computing. The studied computing that in which domains neuromorphic computing has
research work shows that many machine learning algorithms proved its capability like content-based image retrieval, short-
can be improved using Neuromorphic computing system.
term power forecasting, classification, stroke prediction, IoT
The NeuCube neuromorphic framework [10] provides highly
improved solution for neural network and fuzzy systems, etc. This paper also gives information about the different
which can be widely utilized in healthcare industry for neuromorphic systems like loihi, BrainScaleS 2 prototype and
various applications like medical image processing, early how machine learning algorithms can be more applicable by
detection of diseases and patients pattern recognition. In the using neuromorphic computing. Some research work has been
similar way, Crossbar neuromorphic circuitry [17] is done in neuromorphic computing and some neuromorphic chips
capable to predict the band gap of small size molecules of developed but it is at its infant stage but now the future need is
semiconductors, which is opening a new domain for to take this technology to its successful stage by more research
neuromorphic computing in the direction of in this domain. More research work in the neuromorphic
cheminformatics. The Loihi chip for neuromorphic computing is the mandatory need in the industry for taking this
computing has been experimented in recent research works upwards as it promises to open various new possibilities in the
for inducing efficiency in different application areas for computing, healthcare, robotics, and many other artificial
instance robotics [5], Image extraction based on content or intelligence applications. The neuromorphic computing systems
features [12] and the other promising research area is Wind are intended to solve tremendous data parallel processing
power forecasting [13]. Many researchers have worked to problems.
achieve impressible results for handwriting recognition tasks REFERENCES
employing neural network techniques, however ReRAM
neuromorphic circuit-based solution for handwritten images [1] K. Roy, A. Jaiswal, and P. Panda, “Towards spike-based machine
intelligence with neuromorphic computing,” Nature, vol. 575, no.
clearly indicates the significance of power of neuromorphic 7784, pp. 607–617, 2019.
computation. [2] C. D. Schuman, S. R. Kulkarni, M. Parsa, J. P. Mitchell, and B.
Kay, “Opportunities for neuromorphic computing algorithms and
DYNAPs [20] and Spiking MNIST [11] are two evolving applications,” Nat. Comput. Sci., vol. 2, no. 1, pp. 10–19, 2022.
neuromorphic computing systems that are capable to make [3] M. I. Jordan and T. M. Mitchell, “Machine learning: Trends,
optimal deep neural network for IoT based research perspectives, and prospects,” Science (80-. )., vol. 349, no. 6245, pp.
problems. Akida SNN neuromorphic architecture [19] 255–260, 2015.
demonstrated the experimental results with high accuracy for [4] B. Mahesh, “Machine learning algorithms-a review,” Int. J. Sci. Res.
(IJSR).[Internet], vol. 9, pp. 381–386, 2020.
classification of malts in e-nose systems. Moreover, the [5] M. Davies et al., “Advancing neuromorphic computing with loihi: A
energy efficiency software engineering approach can also be survey of results and outlook,” Proc. IEEE, vol. 109, no. 5, pp. 911–
achieved by incorporation of neuromorphic computing 934, 2021.
system as discussed in the discussed paper [14]. PyCarl [15], [6] T. Wunderlich et al., “Demonstrating advantages of neuromorphic
co-simulation tool encourages researcher to deploy computation: a pilot study,” Front. Neurosci., vol. 13, p. 260, 2019.
[7] F. Ma et al., “Optoelectronic perovskite synapses for neuromorphic
neuromorphic development cycle including hardware and
computing,” Adv. Funct. Mater., vol. 30, no. 11, p. 1908901, 2020.
software resource utilization. The studied research papers [8] A. Balaji et al., “A framework to explore workload-specific
demonstrated the improved performance for widely used K- performance and lifetime trade-offs in neuromorphic computing,”
NN machine learning approach with integration of IEEE Comput. Archit. Lett., vol. 18, no. 2, pp. 149–152, 2019.
neuromorphic computing [16], [19]. The application area [9] M. Barnell et al., “Demonstrating Advanced Machine Learning and
radar beam shaping has been explored with BrainScaleS 2 Neuromorphic Computing Using IBM’s NS16e,” in Science and
Information Conference, 2020, pp. 1–11.
neuromorphic prototype by the authors, that provides [10] N. Kasabov et al., “Evolving spatio-temporal data machines based
platform for reinforcement learning with STDP. With this on the NeuCube neuromorphic framework: Design methodology
discussion, we clearly understand the future of neuromorphic and selected applications,” Neural Networks, vol. 78, pp. 1–14,
2016.

5
Authorized licensed use limited to: The National Institute of Engineering. Downloaded on October 15,2024 at 11:03:23 UTC from IEEE Xplore. Restrictions apply.
[11] D. Rasmussen, “NengoDL: Combining deep learning and Computing Simulation in Band Gap Prediction and Chemical
neuromorphic modelling methods,” Neuroinformatics, vol. 17, no. Reaction Classification,” ACS omega, vol. 7, no. 1, pp. 168–175,
4, pp. 611–628, 2019. 2021.
[12] T.-Y. Liu, A. Mahjoubfar, D. Prusinski, and L. Stevens, [18] N. Getty, T. Brettin, D. Jin, R. Stevens, and F. Xia, “Deep medical
“Neuromorphic computing for content-based image retrieval,” PLoS image analysis with representation learning and neuromorphic
One, vol. 17, no. 4, p. e0264364, 2022. computing,” Interface Focus, vol. 11, no. 1, p. 20190122, 2021.
[13] J. M. G. Sopeña, V. Pakrashi, and B. Ghosh, “Neuromorphic [19] A. Vanarse, A. Osseiran, A. Rassau, and P. van der Made,
computing for short-term wind power forecasting,” 2022. “Application of Neuromorphic Olfactory Approach for High-
[14] T. Titirsha, S. Song, A. Balaji, and A. Das, “On the role of system Accuracy Classification of Malts,” Sensors, vol. 22, no. 2, p. 440,
software in energy management of neuromorphic computing,” in 2022.
Proceedings of the 18th ACM International Conference on [20] S. Moradi, N. Qiao, F. Stefanini, and G. Indiveri, “A scalable
Computing Frontiers, 2021, pp. 124–132. multicore architecture with heterogeneous memory structures for
[15] A. Balaji et al., “PyCARL: A PyNN interface for hardware-software dynamic neuromorphic asynchronous processors (DYNAPs),” IEEE
co-simulation of spiking neural network,” arXiv Prepr. Trans. Biomed. Circuits Syst., vol. 12, no. 1, pp. 106–122, 2017.
arXiv2003.09696, 2020. [21] H. Cao, T. Lam, H. Nguyen, A. Venkattraman, D. Parent, and H. Y.
[16] I. Sugiarto, A. B. Gumelar, and A. Yogatama, “Embedded Machine Wong, “Study of ReRAM neuromorphic circuit inference accuracy
Learning on a Programmable Neuromorphic Platform,” in Recent robustness using DTCO simulation framework,” in 2021 IEEE
Advances in Electrical and Electronic Engineering and Computer Workshop on Microelectronics and Electron Devices (WMED),
Science, Springer, 2022, pp. 119–128. 2021, pp. 1–4.
[17] B. Li, H. Sun, H. Shu, and X. Wang, “Applying Neuromorphic

6
Authorized licensed use limited to: The National Institute of Engineering. Downloaded on October 15,2024 at 11:03:23 UTC from IEEE Xplore. Restrictions apply.