Aim to build molecular neuromorphic computing technology

It may pave the way for computing that mimics the brain and helps establish artificial intelligence

It may pave the way for computing that mimics the brain and helps establish artificial intelligence

Artificial intelligence and machine learning can cause a real revolution in the way the world works today, but their development is hampered by the fact that the current state of the art in electronics does not correspond to the needs.

In a bid to develop devices capable of mimicking the functioning of neurons in the brain, researchers at the Indian Institute of Science, Bengaluru (IISc) have designed neuromorphic devices using organic materials that have not been used until now. ‘now. Their work since 2014 goes in this direction.

Organic materials were considered the poorest of the different types of materials in the manufacture of computer components because they were brittle and unstable. “We chose this genre as our horse for the race because we believed that if there was a way to solve these performance issues, the features we could squeeze out of these materials could blast everything out there,” Sreetosh says. Goswami from the Center for Nanosciences and Engineering (CeNSE), IISc.

He and his collaborators have published important articles in this field since 2017 in Natural materials, Nature’s nanotechnology, Advanced materials and Nature, establishing that organic materials can compute reliably and in some respects are even better than inorganic materials. “The molecular system (transition metal complexes of aromatic azo ligands) is the brainchild of my father, Professor Sreebrata Goswami,” says Dr Sreetosh Goswami in an email to The Hindu.

The plastic brain

The human brain that inspired the researchers in their work, in the words of Sreebrata Goswami, who is now with CeNSE, IISc, “significantly surpasses all artificial electronic analogues in learning, cognition and decision-making ability. “. Its remarkable performance consumes only 20 watts of power over a space of 1260 cc. Some of the properties it exhibits which are desirable include interconnection and reconfigurability.

“Brain neurons operate on the edge of chaos with a highly nonlinear feedback mechanism. We are looking for materials that can capture such properties, an elusive goal…” explains Professor Sreebrata Goswami.

Many features

Molecular materials are characterized by interactions between molecules and ions, which then exhibit a multidimensional landscape of parameter space that can be modified to develop appropriate functionalities. The question they posed in a recent article published in Advanced materials was whether they could manipulate these many-body interactions to achieve plasticity and reconfigurability in devices. To do this, they measured current-voltage curves as a function of temperature over a wide range. They could capture features spanning bipolar, unipolar, non-volatile, and volatile memristors.

In the words of Dr. Sreetosh Goswami, this is an “insane amount of variability”, to describe which, the group had to design a mathematical space that could allow for almost any possible characteristic variation desirable in neuromorphic devices. “The same device could work in both analog and digital regimes just by adjusting the activation energy,” says Dr Sreetosh Goswami.

make it work

The challenge was that during low-temperature measurements, in molecular memristors, the switching responses quenched or flattened out as the temperature was lowered. “We could make it work because our molecular devices are robust and the switching mechanisms have a thermodynamic component that still occurs even when the device is cooled,” says Santi Prasad Rath, who is a postdoc at CeNSE, IISc, and the first author of the article published in Advanced materials.

Dr Sreetosh Goswami says: “We are quite confident that we will be able to develop a functional neuromorphic platform based on our metal complexes which could be the world’s first molecular neuromorphic technology.

Sherry J. Basler