The Brainchip Story Collated by Fact Finder Dec 2020 Chapter...

The Brainchip Story Collated by Fact Finder Dec 2020 Chapter One.
The Brainchip story is not one which is well known by the general public for a number of
reasons not the least of which is the fact that Brainchip for obvious commercial reasons has
been operating as much as possible under the radar to protect its commercial and
intellectual advantage in the field of Spiking Neural Network technology.
Recently however with the signing of its first commercial Intellectual Property Licence
agreement it has moved out of the shadows of being a start up in self-imposed stealth mode
to becoming a major international semiconductor provider heralding a new era of artificial
intelligence at the edge and far edge both here on Earth and in Space.
The journey to this point commenced over twenty years ago when Peter van der Made
commenced to formalise his ideas for how to conceive artificial general intelligence. These
ideas were developed by him until in 2008, when he filed the foundational Patent number
8250011 for what was to become Brainchip’s spiking neural network technology known as
AKIDA.
Along this journey other visionaries joined him and in 2015 as a result of Peter van der
Made’s previous experiences with the venture capital model Brainchip Technologies Ltd
was born out of a reverse takeover of the dormant minerals explorer Aziana on the
Australian Stock Exchange. Supported by Mick Bolto Chairman, Adam Osserian Non
Executive Director, Anil Mankar Chief Operating Officer and Senior Engineer and Neil
Rinaldi Non Executive Director, Peter van der Made commenced the journey towards the
commercialisation of his original idea for artificial general intelligence. This caused much
excitement among investors on the ASX and many captivated by the idea of artificial
general intelligence bought shares priced purely on a collective vision of what the future
could hold for Brainchip’s technology.
In October, 2015 an investor presentation by the company announced the introduction of
SNAP. SNAP was the first iteration of SNN technology and stood for Spiking Neuron
Adaptive Processor. The business model involved initially the sale of the intellectual
property however the early high expectations were not to be delivered. There are lots of
reasons but suffice to say having the most advanced revolutionary technology in the world
can be a hard sell to a world which is happy with the status quo and who have no idea what
you are talking about and no interest in learning. In short SNAP in all its variants was ahead
of its time and not market ready and Moore’s Law had yet to reach breaking point.
On 1 September, 2016 the company completed the acquisition of a French based company
Spikenet Technology which specialised in Ai computer vision technology with a view to
accelerating up take of SNAP technology as a video processing solution. Spikenet had a
relationship with Cerco Research Centre where Dr Simon Thorpe (now a member of the
Brainchip Scientific Advisory Board) undertook research with others and in that position had
developed the JAST learning rules.
The term JAST was made up from the first initials of the four inventors and Brainchip saw
the benefit these rules would bring to the development of its technologies and on 20 March,
2017 entered an agreement which gave them exclusive rights to the JAST learning rules.
Bringing together the expertise of Spikenet, Brainchip and the JAST learning rules on 12
September, the company produced a new video processing product which it named
Brainchip Studio. It was a very impressive product and was unlike anything else available
on the market and on 21 August 2017 Brainchip announced that it had been awarded New
1

Product of the Year for 2017 for Video Analytics by Security Today’s panel of independent
experts. Shortly after this on 12 September, 2017 the company announced an upgrade to
Brainchip Studio called Brainchip Accelerator which was capable of 16 channel
simultaneous video processing, with ultra low power and 6x speed boost to Brainchip
Studios CPU-based Artificial Intelligence Software for Object Recognition and performance
levels 7 times more efficient than existing GPU accelerated deep learning systems.
Brainchip Accelerator went on to also win an industry award however the market place in
which it was released was already dominated by well known global brands and proved to be
impenetrable until the then new CEO Lou Dinardo announced the signing of what was in
effect a Heads of Agreement with Gaming Partners International to develop a system built
on Studio technology for game table security in the Casino industry. The agreement
involved the upfront payment by Gaming Partners International of $500,000 as well as an
additional payment of $100,000 to cover engineering works. These payments were made
and it was full steam ahead and the project was reportedly on track and successful
demonstrations of the prototype had gone down well at a Gaming Industry conference in the
USA.
The saying ‘never seems to be able to catch a break’ comes to mind as not long into the
process of designing and building this table gaming solution Gaming Partners International
became the subject of a successful takeover bid by a Japanese gaming company and in the
wash up of that takeover the Japanese owners were not interested in pursuing this direction
as their principle reason for acquiring GPI was to add GPI’s chip technology to their playing
card business. After much negotiation it became obvious to Brainchip that it was a zero sum
game trying to revive the project although both parties have agreed to revisit it in the future.
Chapter Two
The decision was made by Brainchip to go for the end goal in a make or break effort to
achieve the companies ‘holy grail’ and Peter van der Made stepped aside from his position
on the Board of Directors to work exclusively on the final stage of the development of the
AKIDA technology. The company engaged in brutal cost savings and shedding of all staff
not engaged in the technology engineering side of the company.
All of the knowledge and research that had gone into the various attempts at
commercialisation had also driven the advance of the AKIDA technology platform so Peter
van der Made and Anil Mankar had the plan already laid out as to what was required, but
now it became a race not just to be the first to market that would be a given but a race to
finalise the research and engineering in record time before the company was no longer
financially viable.
The various decisions as to how to make the final push to market of this revolutionary
product were brilliant. They knew that this would be make or break. They knew they had the
technology but they knew from the past failures that they had to pick an area or segment of
the market where they would not have any true competition. They also knew that even then
they needed to have found out before they built the product what clients thought of it and
what they liked and did not like about it all on a shoestring budget.
The idea of the Akida Development Environment (ADE) was conceived and it was released
to early access customers. Studio had been released to customers after being produced
and as a result feedback from customers came too late in the cycle but nonetheless this
feedback was available to inform Brainchip as to what the ADE should look like. However by
releasing the ADE in advance it was intended that any further feedback would be taken on
board and designed in which in fact occurred however to accommodate these customer
2

requests the release date of the IP and the production of AKIDA1000 was delayed by twelve
months as Peter van der Made and Anil Mankar went back to the drawing board.
The state of the Brainchip balance sheet and the delay that this redesign caused did not go
down well with many shareholders and a great deal of angst and anger were directed at the
company in particular the CEO which suffice to say he fielded with a dignified air but which
to those who had been involved watching the company since he took up his role clearly
weighed heavily on his shoulders. In an extremely positive sign on 14 December, 2019 a
joint presentation of the ADE performing hand gesture identification took place at a trade
show in combination with Tata Consulting Service engineers and to the delight of many
shareholders one of the presentation slides stated that the Tata engineers were looking
forward to building a robot based upon an AKIDA platform.
Anil Mankar has a strong background in bringing IP to successful implementation on chip
and it is reported that he has over 100 successful chip designs to his credit. As a result
despite the speed which was required he ensured that the company took the time to
implement the AKD1000 design in FPGA before taking it to Socionext for final engineering
and design prior to sending it to TSMC for production of the wafer. In an effort to contain the
costs of producing the AKD1000 chip the company decided that it would share a wafer and
on 8 April, 2020 TSMC commenced production.
These actions of going back to the drawing board following feedback from early access
customers and implementing the design on FPGA produced delays which clearly did not
impress the market and it was reflected in an ever falling share price but the company stuck
with what had to be done to achieve the end result.
Chapter Three
While the engineering and tech people were burning the midnight oil the task of marketing the ADE the IP and the eventual AKD1000 boards and raising capital fell to the CEO and Mr Roger Levinson. Despite Covid-19 these two individuals managed to leverage their contacts around the globe to the extent that despite significant shareholder criticism at the time the company announced a 29 million dollar funding arrangement with LDA Capital ensuring the companies financial viability. Shareholders waited to find out if the AKD1000 which had been through testing and packaging by Socionext would also pass validation tests conducted by the company before installing them on Boards and shipping them to customers. Everything seemed to be falling into place.
In a tribute to the chip design work of Anil Mankar the CEO announced that final validation and testing proved the AKD1000 worked perfectly matching and exceeding all predicted benchmarks something which he had not witnessed in all his years in the semiconductor industry.
While all these milestones were being achieved the CEO was keeping shareholders up to date with progress on the marketing and engagement side of the company and over time announced that Brainchip had well north of 100 Non Disclosure Agreements in place with various interested companies and that they had selected from the interested companies about two dozen which stood out as household names, Fortune 500 companies and most likely to bring commercial benefit. It is indeed difficult to know when the CEO found the time, given that after Peter van der Made stepped down from the Board to concentrate all his efforts on finalising the design of AKD1000 all administrative and sales staff were let go, yet he was able to finalise and announce early access and development agreements with Ford and Valeo both well known Tier 1 companies in the automotive industry where such
3

agreements involved these companies paying Brainchip for the privilege of receiving early access to the AKD1000 test boards.
In a further coup the CEO announced the appointment of Mr Rob Telson as Vice President of World Wide Sales. Rob Telson had been at ARM and had filled the roles of Vice President of Foundry Sales World wide and also Vice President of Sales for the Americas. It is interesting to note that with this appointment Brainchip seemed to be confirming that its stated intention in 2015 to use ARM as its model for building its global footprint still was in play.
The CEO not one to let grass grow also jointly announced with Socionext an agreement to include the AKD1000 chip as part of the Socionext SynQuacer product offering and also a joint development agreement with Magik-Eye the inventor of a revolutionary 3D point sensor capable of depth perception. Then out of left field the CEO announced an Early Access Customer engagement with Vorago for the purpose of Vorago hardening the AKD1000 chip for space deployments in a fully funded Phase 1 program with NASA. Not long after this announcement the CEO announced that NASA had directly been admitted to the Early Access Program.
Chapter Four
While all this has been taking place Peter van der Made having completed the AKD1000
design has with Anil Mankar moved on to the designs for AKD2000 and AKD3000 while at
the same time working with Adam Osseiran and others on the AERO project. The AERO
project started life as a way to leverage the ADE to identify gases and in the process
created a design for a scientific instrument powered by the ADE using only 200 AKIDA
neurons which was capable of state of the art identification of the 20 gas data set (of ten
gases) with latency of only 3 seconds. This breakthrough research has led to trialling the
ADE as the (processor) for identifying Volatile Organic Compounds of various diseases from
samples of exhaled human breath.
As luck would have it Adam Osseiran’s work with Noble Laureate Professor Barry Marshall
on the invention of the Noisy Gut Belt has led to Professor Marshall becoming aware of the
AKIDA technology and accepting an invitation to join the BrainChip Scientific Advisory
Board for the purpose of developing medical diagnostic applications relating to AERO and
its proven capacity to identify gases for the purpose of identify various medical conditions in
real time from sampling exhaled human breath.
The Brainchip Scientific Advisory Board is presently chaired by Adam Osseiran with
Professor Barry Marshall and Dr Simon Thorpe. In a presentation by Peter van der Made to
the Brain Inspired Computing Congress in November, 2020 he confirmed that AKIDA had
been successfully implemented as the processor for the purpose of identifying Covid-19
with a success rate of 94%. It is well recognised that various diseases including some
cancers, diabetes and Covid-19 express themselves in identifiable patterns when exhaled
on an infected person’s breath.
The potential low cost, unconnected, portable nature of an AKIDA medical scientific
instrument for this purpose could be life changing for remote populations and people of
developing countries. The ability to incorporate this technology into telehealth applications
which have grown exponentially in developed countries as a result of Covid-19 infection
promises huge commercial opportunities and together these factors could provide greater
equality of access to modern medical services than at any time in history of the world. In
fact researchers at Australia’s own Flinders’ University have in recent months announced
that they have identified VOC markers in exhaled human breath for head and neck cancer.
4

Chapter Five
On more than one occasion the CEO of Brainchip has stated to shareholders in
presentations during the second half of 2020 that the company was very close to
announcing its first commercial engagement and that this announcement may come late
2020 but certainly early 2021. Well to the amazement of shareholders as Christmas
descended and the markets started to wind down the CEO was true to his word as a trading
halt was called on the 22 December, 2020 and after the close of trade on 23 December,
2020 the CEO and the Board of Brainchip released two announcements.
The first was confirmation that Brainchip had been awarded a sole source contract with
NASA for the AKIDA early access development kit. The second and pivotal announcement
was that the first Intellectual Property licence had been entered with Renesas with the CEO
commenting: “This is an exciting and significant milestone in obtaining the Company’s first
IP licensing agreement. Furthermore, this is a market validation of our technology.”
The salient terms of this agreement as announced are: Aliso Viejo, Calif. – 23 December
2020 – BrainChip Holdings Ltd (ASX: BRN), a leading provider of ultra-low power, high-
performance AI technology, today announced the signing of an intellectual property license
agreement with Renesas Electronics America Inc., a subsidiary of Japan-based Renesas
Electronics Corp., a tier-one semiconductor manufacturer that specializes in microcontroller
and automotive SoC products.
The unconditional agreement provides for:
• a single-use, royalty-bearing, worldwide IP design license for the rights to use the AkidaTM
IP in the customer’s SoC products, which continues while the customer continues to use the
Akida IP in its products. The parties agreed to customary termination terms;
• BrainChip to provide implementation support services (at an agreed fee to cover costs)
aimed at facilitating the customer’s adoption and commercialization of the Akida-licensed
product during the first year of the license agreement.
• The agreement provides for various payment terms including the payment of ongoing
royalties based on the volume of units sold, commencing at certain agreed volume
threshold and the net sale price of the customer’s products. The royalties remain in effect
throughout the life of the licensed product.
• Brainchip to provide software maintenance services, which attract a separate fee if the
customer elects to continue to use these services after the first two years of the agreement.
So after this announcement the next trading day being 24 December, 2020 trading resumed
and the market closed early because it was Christmas Eve. The next chapter in the
Brainchip story will now unfold with expectations of further IP agreements, mass production
of the AKD1000 and many exciting new ventures being anticipated. The rest of the
Brainchip story is likely to be written in front of an increasingly attentive market both here in
Australia and around the world and is likely not to require someone to put it in writing as
AKIDA technology becomes ubiquitous both on Earth and in Space.
Chapter Six
Patents:
In finalising this first book of Brainchip I have included below some of the patents and
published research to complete the historical picture noting that there are still a number of
patents in progress:
5

Autonomous learning dynamic artificial neural computing device and brain inspired system
Patent number 8250011 was filed by Peter van der Made in September 2008.
It outlines a digital simulation of the biological process known as Synaptic Time Dependent
Plasticity (STDP), which can enable an information processing system to perform
approximation, autonomous learning and the strengthening of formerly learned input
patterns. The information processing system consists of artificial neurons (composed of
binary logic gates), which are interconnected via dynamic artificial synapses. Each artificial
neuron consists of a soma circuit and a plurality of synapse circuits, whereby the soma
membrane potential, the soma threshold value, the synapse strength and the Post Synaptic
Potential at each synapse are expressed as values in binary registers, which are
dynamically determined from certain aspects of input pulse timing, previous strength value
and output pulse feedback.
Method and A System for Creating Dynamic Neural Function Librarieshttps://patents.google.com/patent/US20190012597A1/en
Abstract:
A method for creating a dynamic neural function library that relates to Artificial Intelligence
systems and devices is provided. Within a dynamic neural network (artificial intelligent
device), a plurality of control values are autonomously generated during a learning process
and thus stored in synaptic registers of the artificial intelligent device that represent a
training model of a task or a function learned by the artificial intelligent device.
Control Values include, but are not limited to, values that indicate the neurotransmitter level
that is present in the synapse, the neurotransmitter type, the connectome, the
neuromodulator sensitivity, and other synaptic, dendric delay and axonal delay parameters.
These values form collectively a training model. Training models are stored in the dynamic
neural function library of the artificial intelligent device. The artificial intelligent device copies
the function library to an electronic data processing device memory that is reusable to train
another artificial intelligent device.
Spiking neural network US20200143229A1 - Spiking neural network
Abstract: Disclosed herein are system, method, and computer program product
embodiments for an improved spiking neural network (SNN) configured to learn and perform
unsupervised extraction of features from an input stream.
An embodiment operates by receiving a set of spike bits corresponding to a set synapses
associated with a spiking neuron circuit.
The embodiment applies a first logical AND function to a first spike bit in the set of spike bits
and a first synaptic weight of a first synapse in the set of synapses.
The embodiment increments a membrane potential value associated with the spiking
neuron circuit based on the applying.
The embodiment determines that the membrane potential value associated with the spiking
neuron circuit reached a learning threshold value.
The embodiment then performs a Spike Time Dependent Plasticity (STDP) learning function
based on the determination that the membrane potential value of the spiking neuron circuit
reached the learning threshold value.
Low power neuromorphic voice activation system and methodUS10157629B2 - Low power neuromorphic voice activation system and method
Abstract: The present invention provides a system and method for controlling a device by
recognizing voice commands through a spiking neural network. The system comprises a
6

spiking neural adaptive processor receiving an input stream that is being forwarded from a
microphone, a decimation filter and then an artificial cochlea. The spiking neural adaptive
processor further comprises a first spiking neural network and a second spiking neural
network. The first spiking neural network checks for voice activities in output spikes received
from artificial cochlea. If any voice activity is detected, it activates the second spiking neural
network and passes the output spike of the artificial cochlea to the second spiking neural
network that is further configured to recognize spike patterns indicative of specific voice
commands. If the first spiking neural network does not detect any voice activity, it halts the
second spiking neural network.
Neural processor based accelerator system and methodUS20170024644A1 - Neural processor based accelerator system and method
Abstract: A configurable spiking neural network based accelerator system is provided. The
accelerator system may be executed on an expansion card, which may be a printed circuit
board.
The system includes one or more application specific integrated circuits comprising at least
one spiking neural processing unit and a programmable logic device mounted on the printed
circuit board. The spiking neural processing unit includes digital neuron circuits and digital,
dynamic synaptic circuits.
The programmable logic device is compatible with a local system bus.
The spiking neural processing units contain digital circuits comprises a Spiking Neural
Network that handles all of the neural processing.
The Spiking Neural Network requires no software programming, but can be configured to
perform a specific task via the Signal Coupling device and software executing on the host
computer.
Configuration parameters include the connections between synapses and neurons, neuron
types, neurotransmitter types, and neuromodulation sensitivities of specific neurons.
Intelligent biomorphic system for pattern recognition with autonomous visual feature extractionUS20170236027A1 - Intelligent biomorphic system for pattern recognition with autonomous visual feature extraction
Abstract: Embodiments of the present invention provide a hierarchical arrangement of one
or more artificial neural networks for recognizing visual feature pattern extraction and output
labelling.
The system comprises a first spiking neural network and a second spiking neural network.
The first spiking neural network is configured to autonomously learn complex, temporally
overlapping visual features arising in an input pattern stream. Competitive learning is
implemented as spike time dependent plasticity with lateral inhibition in the first spiking
neural network.
The second spiking neural network is connected by means of dynamic synapses with the
first spiking neural network, and is trained for interpreting and labelling output data of the
first spiking neural network. Additionally, the output of the second spiking neural network is
transmitted to a computing device, such as a CPU for post processing.
Secure Voice Communications SystemUS20190188600A1 - Secure Voice Communications System
Abstract: Disclosed herein are system and method embodiments for establishing secure
communication with a remote artificial intelligent device.
An embodiment operates by capturing an auditory signal from an auditory source. The
embodiment coverts the auditory signal into a plurality of pulses having a spatio-temporal
distribution.
The embodiment identifies an acoustic signature in the auditory signal based on the plurality
of pulses using a spatio-temporal neural network. The embodiment modifies synaptic
strengths in the spatio-temporal neural network in response to the identifying thereby
7

causing the spatio-temporal neural network to learn to respond to the acoustic signature in
the acoustic signal.
The embodiment transmits the plurality of pulses to the remote artificial intelligent device
over a communications channel thereby causing the remote artificial intelligent device to
learn to respond to the acoustic signature, and thereby allowing secure communication to be established with the remote artificial intelligent device based on the auditory signature.
Intelligent Autonomous Feature Extraction System Using Two Hardware Spiking Neutral Networks with Spike Timing Dependent PlasticityUS20170236051A1 - Intelligent Autonomous Feature Extraction System Using Two Hardware Spiking Neutral Networks with Spike Timing Dependent Plasticity
Abstract: Embodiments of the present invention provide an artificial neural network system
for feature pattern extraction and output labelling. The system comprises a first spiking
neural network and a second spiking neural network.
The first spiking neural network is configured to autonomously learn complex, temporally
overlapping features arising in an input pattern stream. Competitive learning is implemented
as spike timing dependent plasticity with lateral inhibition in the first spiking neural network.
The second spiking neural network is connected with the first spiking neural network
through dynamic synapses, and is trained to interpret and label the output data of the first
spiking neural network.
Additionally, the labelled output of the second spiking neural network is transmitted to a
computing device, such as a central processing unit for post processing.
System and Method for Spontaneous Machine Learning and Feature Extractionhttps://patents.google.com/patent/US20180225562A1/en
Abstract: Embodiments of the present invention provide an artificial neural network system
for improved machine learning, feature pattern extraction and output labelling. The system
comprises a first spiking neural network and a second spiking neural network. The first
spiking neural network is configured to spontaneously learn complex, temporally
overlapping features arising in an input pattern stream.
Competitive learning is implemented as Spike Timing Dependent Plasticity with lateral
inhibition in the first spiking neural network. The second spiking neural network is connected
with the first spiking neural network through dynamic synapses, and is trained to interpret
and label the output data of the first spiking neural network.
Additionally, the output of the second spiking neural network is transmitted to a computing
device, such as a CPU for post processing.
Method, digital electronic circuit and system for unsupervised detection of repeating patterns in a series of eventshttps://patents.google.com/patent/US20190286944A1/en
and
Published Research
Unsupervised learning of repeating patterns using a novel STDP based algorithmhttps://jov.arvojournals.org/article.aspx?articleid=2651951
Abstract: A method of performing unsupervised detection of repeating patterns in a series of
events, includes
a) Providing a plurality of neurons, each neuron being representative of W event types;
b) Acquiring an input packet comprising N successive events of the series;
c) Attributing to at least some neurons a potential value, representative of the number of
common events between the input packet and the neuron;
d) Modify the event types of neurons having a potential value exceeding a first threshold TL;
e) generating a first output signal for all neurons having a potential value exceeding a
second threshold TF, and a second output signal, different from the first one, for all other
neurons. A digital electronic circuit and system configured for carrying out such a method is
also provided
8

Computational vision systems that are trained with deep learning have recently matched
human performance (Hinton et al). However, while deep learning typically requires tens or
hundreds of thousands of labelled examples, humans can learn a task or stimulus with only
a few repetitions. For example, a 2015 study by Andrillon et al. showed that human listeners
can learn complicated random auditory noises after only a few repetitions, with each
repetition invoking a larger and larger EEG activity than the previous. In addition, a 2015
study by Martin et al. showed that only 10 minutes of visual experience of a novel object
class was required to change early EEG potentials, improve saccadic reaction times, and
increase saccade accuracies for the particular object trained. How might such ultra-rapid
learning actually be accomplished by the cortex? Here, we propose a simple unsupervised
neural model based on spike timing dependent plasticity, which learns spatiotemporal
patterns in visual or auditory stimuli with only a few repetitions. The model is attractive for
applications because it is simple enough to allow the simulation of very large numbers of
cortical neurons in real time. Theoretically, the model provides a plausible example of how
the brain may accomplish rapid learning of repeating visual or auditory patterns using only a
few examples.
Unsupervised Feature Learning With Winner-Takes-All Based STDPhttps://www.frontiersin.org/articles/10.3389/fncom.2018.00024/full
present a novel strategy for unsupervised feature learning in image applications inspired by
equivalence between rank order coding Leaky-Integrate-and-Fire neurons and ReLU
AbstractWe
the Spike-Timing-Dependent-Plasticity (STDP) biological learning rule. We show
artificial neurons when applied to non-temporal data. We apply this to images using rank-
order coding, which allows us to perform a full network simulation with a single feed-forward
pass using GPU hardware. Next we introduce a binary STDP learning rule compatible with
training on batches of images. Two mechanisms to stabilize the training are also presented :
a Winner-Takes-All (WTA) framework which selects the most relevant patches to learn from
along the spatial dimensions, and a simple feature-wise normalization as homeostatic
process. This learning process allows us to train multi-layer architectures of convolutional
sparse features. We apply our method to extract features from the MNIST, ETH80, CIFAR-
10, and STL-10 datasets and show that these features are relevant for classification. We
finally compare these results with several other state of the art unsupervised learning
methods.
A Hardware-Deployable Neuromorphic Solution for Encoding and Classification of Electronic Nose Datahttps://www.mdpi.com/1424-8220/19/22/4831
AbstractIn several
application domains, electronic nose systems employing conventional data processing
approaches incur substantial power and computational costs and limitations, such as
significant latency and poor accuracy for classification. Recent developments in spike-based
bio-inspired approaches have delivered solutions for the highly accurate classification of
multivariate sensor data with minimized computational and power requirements. Although
these methods have addressed issues related to efficient data processing and classification
accuracy, other areas, such as reducing the processing latency to support real-time
application and deploying spike-based solutions on supported hardware, have yet to be
studied in detail. Through this investigation, we proposed a spiking neural network (SNN)-
based classifier, implemented in a chip-emulation-based development environment, that
can be seamlessly deployed on a neuromorphic system-on-a-chip (NSoC). Under three
different scenarios of increasing complexity, the SNN was determined to be able to classify
real-valued sensor data with greater than 90% accuracy and with a maximum latency of 3 s
on the software-based platform. Highlights of this work included the design and
implementation of a novel encoder for artificial olfactory systems, implementation of
unsupervised spike-timing-dependent plasticity (STDP) for learning, and a foundational
study on early classification capability using the SNN-based classifier.
9

Real-Time Classification of Multivariate Olfaction Data Using Spiking Neural Networkshttps://www.mdpi.com/1424-8220/19/8/1841
Abstract: Recent studies in bio inspired artificial olfaction, especially those detailing the
application of spike-based neuromorphic methods, have led to promising developments
towards overcoming the limitations of traditional approaches, such as complexity in handling
multivariate data, computational and power requirements, poor accuracy, and substantial
delay for processing and classification of odours. Rank-order-based olfactory systems
provide an interesting approach for detection of target gases by encoding multi-variate data
generated by artificial olfactory systems into temporal signatures. However, the utilization of
traditional pattern-matching methods and unpredictable shuffling of spikes in the rank-order
impedes the performance of the system. In this paper, we present an SNN-based solution
for the classification of rank-order spiking patterns to provide continuous recognition results
in real-time. The SNN classifier is deployed on a neuromorphic hardware system that
enables massively parallel and low-power processing on incoming rank-order patterns.
Offline learning is used to store the reference rank-order patterns, and an inbuilt nearest
neighbour classification logic is applied by the neurons to provide recognition results. The
proposed system was evaluated using two different datasets including rank-order spiking
data from previously established olfactory systems. The continuous classification that was
achieved required a maximum of 12.82% of the total pattern frame to provide 96.5%
accuracy in identifying corresponding target gases. Recognition results were obtained at a
nominal processing latency of 16ms for each incoming spike. In addition to the clear
advantages in terms of real-time operation and robustness to inconsistent rank-orders, the
SNN classifier can also detect anomalies in rank-order patterns arising due to drift in
sensing arrays.
A Review of Current Neuromorphic Approaches for Vision, Auditory, and Olfactory Sensorshttps://www.frontiersin.org/articles/10.3389/fnins.2016.00115/full
entional vision, auditory, and olfactory sensors generate large volumes of redundant data
neuromorphic sensors have been developed. These sensors mimic the neuro-biological
AbstractConv
and as a result tend to consume excessive power. To address these shortcomings,
architecture of sensory organs using aVLSI (analog Very Large Scale Integration) and
generate asynchronous spiking output that represents sensing information in ways that are
similar to neural signals. This allows for much lower power consumption due to an ability to
extract useful sensory information from sparse captured data. The foundation for research
in neuromorphic sensors was laid more than two decades ago, but recent developments in
sophisticated neuromorphic sensors that provide numerous advantages over conventional
sensors. In this paper, we review the current state-of-the-art in neuromorphic
implementation of vision, auditory, and olfactory sensors and identify key contributions
across these fields. Bringing together these key contributions we suggest a future research
direction for further development of the neuromorphic sensing field.
understanding of biological sensing and advanced electronics, have stimulated research on
Application of a Brain-Inspired Spiking Neural Network Architecture to Odour Data Classificationhttps://www.mdpi.com/1424-8220/20/10/2756
Abstract: Existing methods in neuromorphic olfaction mainly focus on implementing the data
transformation based on the neurobiological architecture of the olfactory pathway. While the
transformation is pivotal for the sparse spike-based representation of odour data,
classification techniques based on the bio-computations of the higher brain areas, which
process the spiking data for identification of odour, remain largely unexplored. This paper
argues that brain-inspired spiking neural networks constitute a promising approach for the
next generation of machine intelligence for odour data processing. Inspired by principles of
brain information processing, here we propose the first spiking neural network method and
associated deep machine learning system for classification of odour data. The paper
demonstrates that the proposed approach has several advantages when compared to the
10

current state-of-the-art methods. Based on results obtained using a benchmark dataset, the
model achieved a high classification accuracy for a large number of odours and has the
capacity for incremental learning on new data. The paper explores different spike encoding
algorithms and finds that the most suitable for the task is the step-wise encoding function.
Further directions in the brain-inspired study of odour machine classification include
investigation of more biologically plausible algorithms for mapping, learning, and
interpretation of odour data along with the realization of these algorithms on some highly
parallel and low power consuming neuromorphic hardware devices for real-world
applications.
Neuromorphic engineering — A paradigm shift for future IM technologieshttps://ieeexplore.ieee.org/abstract/document/8674627
The End.
developments in measurement science have mainly focused on enhancing the quality of
AbstractRecent
measurement procedures, improving their efficiency, and intro
11