Thursday, April 25, 2024

World’s First SNN Developed at IIT Bombay

- Advertisement -

The world has been taken over by computing technology that uses transistors to create digital gates. However, utilising typical digital computing technologies to replicate a complex processes in the biological brain has considerable inefficiencies. Adapting mature silicon technology by leveraging physics to design and manufacture efficient artificial neurons.

Professor Udayan Ganguly of IIT Bombay proposed the usage of quantum tunnelling current in a silicon on insulator (SOI) technology as a solution. The quantum tunnelling current levels are similar to a classical transistor’s off-current level. The off-current is 10000 times smaller than ordinary transistor on-currents. This results in significant energy and space savings!

Left to Right: Prof. Maryam Shojaei Baghini, Vivek Saraswat, Prof. Udayan Ganguly, Ajay Singh holding the test board with the Neural Network chip. (Credit: Ankit Bende)

A device-circuit-algorithm co-design team led by Prof. Udayan Ganguly and Prof. Maryam Shojaei Baghini of IIT Bombay was formed to take the idea to the next systems level. The team has collaborated to create the world’s first spiking neuro-synaptic core on 45nm SOI technology. Speech recognition and a neural network method inspired by the auditory cortex was used. The Liquid State Machine, a spoken word recognition system, is based on the premise that the neural network produces a neuronal reservoir similar to a liquid reservoir.

- Advertisement -

Vivek Saraswat, a graduate student, worked on converting this method to hardware by including numerous hardware peculiarities in “ideal” software implementations to demonstrate the hardware implementation’s viability. Ajay Singh, another graduate student, lead the circuit demonstration. He built a 20-neuron input layer that sent the neural spike signal into a 36-neuron reservoir with 256 synapses connecting neurons in a random recurrent fashion. Through design, several of the important hardware “quirks” revealed in algorithms were minimised.

On the mature 45RFSOI technology, the neural network circuit was designed and constructed. The acclaimed Global Foundries University Program provided funding for the chip manufacturing. According to Dr Ted Letavic, CTO of GF’s Compute and Wired Infrastructure Business Unit, “This is pioneering work that showcases the advantages of mature SOI technology to demonstrate a new neural compute paradigm based on quantum tunneling.”

Click here to access the paper.


Unique DIY Projects

Electronics News

Truly Innovative Tech

MOst Popular Videos

Electronics Components