A PDF file should load here. If you do not see its contents
the file may be temporarily unavailable at the journal website
or you do not have a PDF plug-in installed and enabled in your browser.
Alternatively, you can download the file locally and open with any standalone PDF reader:
http://eujournal.org/index.php/esj/article/viewFile/5716/5522
THE ERA OF NEUROSYNAPTICS: NEUROMORPHIC CHIPS AND ARCHITECTURE
European Scientific Journal June 2015 /SPECIAL/ edition ISSN: 1857 - 7881 (Print) e
THE ERA OF NEUROSYNAPTICS: NEUROMORPHIC CHIPS AND ARCHITECTURE
Siddhartha Agarwal Divyanshi Rastogi Aayush Singhal 0
0 Department of Electronics and Communication JSS Academy of Technical Education NOIDA , India
Since its invention the modern day computer has shown a significant improvement in its performance and storage capacity.However, most of the current processor cores remain sequential in nature which limit the speed of computation. IBM has been consistently working over this and with the launching of neurosynaptic chips, it has opened a new gateway of thought process. This paper aims at reviewing the various stages and researches that have been instrumental in the overall development of neuromorphic architecture which aims at developing flexible brain like structure capable of performing wide range of real time computations while keeping ultra-low power consumption and size factor in mind. Inspired by the human brain, which is capable of performing complex tasks rapidly and accurately without being programmed and utilizing very less energy, TrueNorth chips tends to mimic the human brain so as to perform complex computations at a faster pace. This has inspired a new field of study aimed at development of the cognitive computing systems that could potentially emulate the brain's computing efficiency, size and power.The paper also aims to highlight the inadvertent challenges of neuromorphic architecture as posed by the prevailing technologies which are a major field of research in near future.
-
Introduction
Human Brain- synonymous with the central processing unit of a
computer , consists of a wide network of neurons and attached dendrons
which on contraction and expansion help in transferring of information from
one part of body to another. This very transfer is done by Synapse, which
means ‘conjunction’. Thus, at a synaptic site, signal passing neuron comes in
close contact with the target which is rich in extensive array of molecular
machinery.
In actual, the biological neural systems perform wide range of real
time computations and tasks such as pattern recognition, sensory
reconstruction carried out by these low power, dense neural circuits which
within metabolic constraints are more efficient that traditional computers.
The basic computational unit of this system is the neurons that
communicate with each other through the generation and modulation of
spike trains where it may be an all-or-nothing pulse.
The human brain consists of a staggering number of over 100 billion
neurons and over 1 trillion synapses.
The implementation of such a complex system is feasible through use
of supercomputers but power and space has always been a constraint,
preventing its usefulness in mobile systems for real time applications. Thus,
mimicking human brain so as to take a big leap towards bionic systems has
been a major concern over decades. With the aim of producing intelligent
memory cells, materials such as chalcogenides need to be crafted which can
sustain non-Boolean Algebra,thereby offering multi stable states. The
memories so created are called ‘cognitive memory’ and forms the very basis
of OCD’s.
Ovonic computational devices (OCD) have analogous functions to
neurons in brain and their synapses and they tend to offer same plasticity as
organic molecules associated at neurosynaptic sites. Their very composition
comprises of nano-dimensional amorphous structures like chalcogenide
glasses offering high optical quality, which can be used to mimic brain like
computations.
But in order to improve the switching times and reduce power
consumption per synaptic event, the neuromorphic architecture came into
being, which was capable of running spiking neural networks in compact
and low power hardware. This neuromorphic architecture used analog
circuits for biological components and digital asynchronous circuits for the
communication of spike events. Inspite of its compactness and reduced
power consumption , its sensitivity to process variations, ambient
temperature and noisy environment posed a challenge in configuring circuit
that could operate under a wide array of external parameters. This limited
correspondence between analog implementations and neural algorithm was
an obstacle to algorithm development and deployment . Even the lack of
addition of high density capacitors and sub-threshold currents in analog
implementation made it more complex and unreliable. Thus, implementation
of neuromorphic architecture using discrete time, low power event driven
circuits provided a path to both area and power efficient architecture, capable
of one-to-one correspondence with the simulator.
This breakthrough came as a part of IBM’s launch of neurosynaptic
chips, which opened gateways to a new thought process.
The world of synapses
The innovation heralded with development of neurosynaptic core
which had 256 integrate-and-fire neurons (...truncated)