Artificial city

Iisc team builds a transistor that helps artificial neural networks | Bangalore News

Bangalore: Artificial Neural Networks (ANN) – a system of interconnected electrical circuits or algorithms that communicate with each other much like neurons in the brain – are growing in popularity for their ability to perform complex tasks, especially in pattern recognition.
However, pointing out that existing devices that mimic neurons and synapses (junctions between two interacting neurons) operate at high voltages and exhibit varying conductivity, making it difficult to construct effective ANNs, the IISc researchers said they have developed a new transistor that can help. “… To address these concerns, we have developed a transistor that has been shown to mimic several characteristics of a biological synapse,” read a statement from IISc.
The team, led by Digbijoy Nath at the Center for Nano Science and Engineering (CeNSE), made a transistor from multilayer indium selenide (In2Se3) – an inorganic compound made up of indium and selenium – and tested its response to sequential voltage pulses, similar to electrical signals in neurons. “Communication between biological synapses has some unique signatures. Prolonged activation of a neuron could make the neuron receiving the signal from it (“postsynaptic neuron”) either less or more responsive, depending on the nature of the signal, “the researchers said.
The postsynaptic neuron also responds differently to consecutive pulses separated by a very short time interval compared to a single pulse, the researchers said, adding that the output characteristics of their device captured these essential characteristics. “The researchers used the device’s responses to create a computer model that simulated an ANN. When they trained the ANN with a database of handwritten numbers, it was able to correctly recognize 93% of the numbers … ”IISc added.