Neuromorphic Computing: A Chipset To Imitate Human Brain

Neuromorphic Computing
A CHIPSET TO IMITATE HUMAN BRAIN

Members
Name Roll No
Siddesh Bhor 207
Rudresh Chaudhary 208
Adnan Chaudhary 209
Prathamesh Deshmukh 210
Swaraj Gadhave 211
Ashish Gupta 212
Guide: Prof. Geetha G

Contents
• Traditional hardware architechture
• Limitations of traditional chips
• How does nature compute?
• Software implementation of neurons
• Physical Implementation of Neuromorphic Chips
• Neuromorphic chips and start-ups
• Existing systems
• Applications
• Conclusion
Traditional hardware architechture
Von-Neumann Model
• The system is divided into several major,
physically separated, rigid functional units
such as memory (MU), control processing
(CPU), arithmetic/logic (ALU), and data paths.
• This separation produces a temporal and

energetic bottleneck because information
has to be shuttled repeatedly between the
different parts of the system.
• This “von Neumann” bottleneck limits the

future development of revolutionary
computational systems.
Limitations of traditional chips
• Moore's Law depends on
transistors shrinking, and
eventually, the laws of
physics intervene.
• Physical limit is 5nm.
• Basically transistors
cannot be built
smaller than atoms!
How does nature compute?
• Dendrites: provide multiple,
distributed inputs into the
neurons.
• Axons: provide a long-

distance output connection
between prenaptic synapse
and postsynaptic synapse.
• Synapses: provide
interconnections between
neurons.
Software Implementation of Neurons
Feed-Forward: Data enters at the inputs and passes through the network, layer by
layer, until it arrives at the outputs.
Back-Propagation: It works by computing the gradient of the loss function with

respect to each weight by the chain rule.
Physical Implementation of
Neuromorphic Chips
• Every connection is a synapse which is the data transfer point.
• Green bars are the inputs of that layer and Red bars are the inputs of next
layer.
• By stacking up the basic layer,a
network can be created in hardware
chips.
• It is called Memristive Crossbar

Array Structure.
• Switching layer is made up of memristors.
• Memristors are the building blocks of artificial neurons in hardware.
• Memristors can store the resistance state.
• Hence the traditional approach of storing 1’s or 0’s is overcame by this new
technology.
Existing Systems
IBM – TrueNorth
• IBM with the collaboration of the DARPA
program built the chip TrueNorth.
• The chip has actually proven

useful in relation to low energy
consumption compared to GPUs.
• IBM’S objective is to use the chip on

cognitive applications such as robotics,
classification, action classification,
audio processing, stereo vision, etc.
Intel - Loihi
• Intel released its neuromorphic chip
Loihi in 2018.
• Loihi is the most effective and energy-

efficient chip for cognitive applications
(among the neuromorphic chips).
• Intel believes that it should be related to

the robotic sector, which is the one where
the neuromorphic chips can more markedly
express their competitive advantages, ie: a
real-time interference with low energy consumption.
Neuromorphic research chips
 SpinNNaker:
SpiNNaker-1 is the world’s largest neuromorphic
computing platform and will assist EBRAINS.
 BrainScaleS:
It is an accelerated system that runs 10000 times faster
than biological speed, thus enabling the discovering of
the various parts of the brain quicker.
Braindrop & Neurogrid:

Braindrop chip claims to consume lower energy than
the Intel Loihi and the “energy-efficient” AI accelerator
of the Tesla chip.
Neuromorphic hardware start-ups
 Brainchip – Akida :
• The main advantage of Akida is that they can be on the market very soon,
thus giving them the advantage of spreading out the technology among the
early adopters of the SNN.
 aiCTX :
• They have modified and improved this original chip into different
purposes. The chip DYNAP-SE2 is suitable for real-time applications in the
area of robotics and medical health applications.
Applications
Mobile Applications
• Intelligent software is essential for the
current usage of mobile applications. They
cover from image processing to text
processing, audio processing, etc.
• These chips can handle some edge AI

applications such as face recognition, real-
time translation, photo segmentation,
voice recognition.
Adaptive Robotics
• GPU’s have difficulties to process perception
and learning.
• GPU’s consume a lot of energy to handle large-

scale operations.
• Neuromorphic chips are one of the best

candidates to handle edge-robotic applications
with their brain-like energy efficiency and
scalable neuron systems.
Conclusion
• In our emerging and dynamic AI-based society, research and development on
AI is to a large extent focused on the improvement and utilisation of deep
neural networks and AI accelerators.
• However, there is a limit in the architecture of traditional von Neumann

systems, and the exponential increasing of data-size and processing requires
more innovative and powerful solutions.
• Neuromorphic chips are compatible with event-based sensors applications and

emerging technologies such as photonic, graphene or non-volatile memories.
They have a huge potential in the development of AI and can certainly be a
dominant technology in the next decade.
Thank You!

Neuromorphic Computing: A Chipset To Imitate Human Brain

Uploaded by

Neuromorphic Computing: A Chipset To Imitate Human Brain

Uploaded by

Neuromorphic Computing

A CHIPSET TO IMITATE HUMAN BRAIN

Guide: Prof. Geetha G

• This separation produces a temporal and

• This “von Neumann” bottleneck limits the

• Physical limit is 5nm.

• Axons: provide a long-

Back-Propagation: It works by computing the gradient of the loss function with

• It is called Memristive Crossbar

• Memristors are the building blocks of artificial neurons in hardware.

• Memristors can store the resistance state.

• The chip has actually proven

• IBM’S objective is to use the chip on

• Loihi is the most effective and energy-

• Intel believes that it should be related to

Braindrop & Neurogrid:

• These chips can handle some edge AI

• GPU’s consume a lot of energy to handle large-

• Neuromorphic chips are one of the best

• However, there is a limit in the architecture of traditional von Neumann

• Neuromorphic chips are compatible with event-based sensors applications and

You might also like