Neuromorphic Engineering

What is Neuromorphic Engineering?

Neuromorphic engineering can be termed as mimicking the neurobiological architecture on a Very

Large-Scale Integrated Circuit (VLSI). The neuromorphic devices have an architecture that is based on

the neuro-biological structure of a human brain. Which means, the devices will be able to process

any complex and noisy data with less power to process. It is a well-known fact that there hasn’t been

a device that has the capability of human brain (Douglas, Mahowald, & Mead, 1995) . The reason for

the same is that, understanding the strategies of human brain is complicated in the present time.

But when it is understood and when devices can think and have intelligence more than humans, is

when we have reached singularity in technology.

The main goal of developing neuromorphic devices is that, the devices will be trained to learn from

the input in such a way that human brain does. But we still do not have a full understanding as to

how the brain receives, stores and processes the data through the neurons. A device can work like a

human brain if its processing pattern is mimicked to a device. This is where Neuromorphic

engineering emerges. In the paper (Soman, jayadeva, & Suri, 2016) has compared the growth of

neuromorphic engineering publications over the last decade and also predict the future outcomes.

There has been a drastic difference since 2005. Some recent developments discussed in the paper

are vision control and robot control applications, biomedical field, perception engineering etc.,

Recent Advancements and study:

Neuroscience is a field that has had an enormous progress in understanding the features and

process of a human brain. Using the principles and information that is generated by neuroscientists,

we have already seen advancements in the machine learning and AI in terms of recognising the

image, audio and video. ("Building brain-inspired computing," 2019) believe that apart from new algorithms the artificial intelligence needs more new devices to make it happen and as per the

Moore’s law our computers are about to reach a point of being stagnant performance.

Spiking neural network.

Spiking neural network (SNN) is a program that helps in the simulation of neural networks on real-

time basis. In other words, they are artificially designed neural network that imitate the working of

the actual neural network. SNN plays a major role in the neuromorphic engineered devices. In the

paper (Chicca, Stefanini, Bartolozzi, & Indiveri, 2014) , the plasticity, silicon neurons and synapse

have been tested and modelled to derive a spiking neuron network.

Event Based Cameras:

Event based cameras are vision sensors that shows the output based on the change in the brightness

of the pixel. The advantage of this sensor is that it works on only the important information rather

that wasting memory on static unwanted data. Also, when the motions are at a high speed, the

sensors are still able to capture the event with a less consumption of power (Gallego et al., 2017) . In

several fields motion sensors are used. Some of the examples would be Augmented reality (AR),

Virtual reality (VR), video games etc., With the older version of sensors, it has been a drawback with

the speed of the regular cameras. Thus, comes the role of event-based cameras. 

In the paper (Jabłoński & Tadeusiewicz, 2016) the author has used the line-scan method to detect the events instead of the frame based method. This has been done to improve the latency and resolution of

event detection. Based on the experiment, the temporal resolution was noticed to be increased.

Silicon Cochlea:

The normal sensors used to replicate the vision, olfactory and auditory senses have a problem that

they generate huge amount data which is unneeded and thus has a drawback of consuming a great

deal of energy. The task of neuromorphic research is to replicate the senses in a human and to do it

with the efficiency of brain that is with usage of minimum amount of energy. There is a lot of

research done in designing and inventing these neuromorphic sensors. These sensors use the (VLSI)

technology to imitate the neuro-biological sensors. They also generate an output which is

asynchronous and is spiking in nature. It is a close representation of how our neural network senses

signals. Needless to say, before having any application for the humans, this technology will be used

and supported by machines. We will discuss below how neuromorphic engineering is being used to

mimic auditory and vision senses.