• Twitter
  • Facebook
  • Google+
  • LinkedIn

Sandip Lashkare

Name: Sandip Lashkare

Email Address: lashkaresandip@gmail.com

Roll number: 154076031

Department: Electrical

Programme: Ph.D.

Year of study: 5

Name of Guide: Prof. Udayan Ganguly

Title: PMO based oscillator neuron for neuromorphic computing

Abstract: Approaching the end of Moore’s law i.e. doubling the number of transistor on a chip in approximately every two years, the researchers of computing and electronics domain turned to new architectural approaches as opposed to Von-Neumann architecture in today's computers to improve the computational efficiency in terms of power, time, and area. Among various architectures explored, the brain-inspired architectures have gained significant attention due to its massively parallel computational power at low energy cost in solving optimization problems like pattern recognition, vertex-colouring, travelling salesman problem.

The cortex region of the brain is a network of weakly coupled oscillatory neurons have been shown to solve the optimization problems with ease. The communication between these oscillators depends on the frequencies at which they operate and the solution lies in the phase difference between oscillators. Such rhythmic activity is observed in nature as well, for example, the synchronous flashing of fireflies.

With such inspiration from brain, oscillatory neural networks or neurocomputers are being developed globally. However, the low power and area scalable nano-oscillator and dense oscillatory network integration is still a challenge.

The experimental and system-level capability at hand and with the neuro-oscillator idea, we have invented a PrMnO3(PMO) based memory device based compact oscillator neuron to develop an oscillatory neural network to solve optimization problems. The neuron oscillator consists of three simple elements which include a PMO device in series with a resistor and a capacitor produces self-sustained oscillations. As compared to the state-of-the-artwork, these oscillator input voltages are low which makes it suitable for use with existing technology without the voltage conversion issues. Finally, the oscillatory neural network has been demonstrated to solve complex problems like pattern recognition and vertex-colouring with high energy and computational efficiency.

This work is published in IEEE EDL-2018 and IEEE DRC-2019.

Some idea on possible savings or benefits: Such oscillatory neural network promises to solve large scale problems which occur in day to day life of every citizen like resource allocation for maximum returns be it for ration shops or grain go-downs to travelling salesman problem and home healthcare system for efficient time and resource management.

Future of the work: As it is believed that the next generation of computers will incorporate the principles of the human brain, such an oscillatory neurocomputers certainly has the capacity to advance the research in that direction.