Researchers “train” carbon nanotubes to perform computational tasks

carbon nanotubes
Nanotubes dispersed in liquid crystal were rearranged using electrical signals to act as a nanocircuit. The study opens up a whole new field of research and application in electronic circuits (image: Massey, M. K. et al. Evolution of Electronic Circuits using Carbon Nanotube Composites. Sci. Rep. 6, 32197; doi: 10.1038/srep32197 (2016))

A computer-controlled technique inspired by evolutionary biology has been used to rearrange carbon nanotubes suspended in liquid crystal through the application of electrical fields to cause them to act as a nanocircuit capable of performing simple computational tasks. The scientists who conducted the experiment include Brazilian physicist Diogo Volpati. The experiment is described in the paper “Evolution of electronic circuits using carbon nanotube composites”, published in Scientific Reports, an online journal owned by Springer Nature.

Currently a postdoctoral fellow at Mid Sweden University in Sundsvall, Sweden, Volpati participated in the study as part of his research project “Molecular control in nanostructured films of carbon nanotubes”, supported by FAPESP.

“Instead of creating a circuit step by step using discrete components – capacitors, resistors, and so on – our work takes disordered material and ‘trains’ it to act as a circuit that can perform a computational task, in this case, separating and classifying datasets,” Volpati told Agência FAPESP. “The training is done using an evolutionary algorithm based on concepts borrowed from biology.”

In the experiment, the nanotubes were dispersed in a liquid crystal matrix, and drops of about 1 microliter in volume were deposited on a set of electrodes, which supplied the inputs and outputs for the electrical signals. In the absence of a signal, the nanotubes were “disoriented”, positioning themselves randomly in the medium. In response to a signal, they repositioned themselves, moving in the liquid crystal following the lines of force in the electrical field. The researchers tested various concentrations of nanotubes in the liquid crystal matrix.

The diagram above (reproduced from Scientific Reports) schematically illustrates what the experiment achieved. The gray circle represents a drop of liquid crystal containing nanotubes, with electrodes running under it (yellow dots and traces). In the callout magnifying an area demarcated by a dotted red line, four electrodes are shown to be connected by a web of nanotubes (black lines). The nanotubes are initially disoriented in the absence of an electrical signal. The red arrows indicate the electrodes that supply the training signals. The corner electrodes perform the computational task enabled by realignment of the nanotubes in response to the configuration or training signals.

“Basically, the experiment consisted of using electric signals to modify the morphological characteristics and electrical properties of the material, in this case, single-walled carbon nanotube/liquid crystal composites,” Volpati said. “The purpose of the modification was to ‘train’ the material to perform a computational task via the network of electrodes.”

The computational task, which consisted of separating two datasets, was extremely simple. The point of the experiment was not to perform a complex task but to prove the principle that such material can be “trained”.

To do this, the electrical signals that reorganized the nanotubes were applied according to an evolutionary algorithm. “We started with a blend of data belonging to two distinct classes. We ‘asked’ the material to separate the classes. Whenever a major error in separation occurred, we had the material ‘evolve’ by passing a voltage between different electrodes again. This process of training and task performance was repeated several times until the errors were reduced to an acceptable minimum,” Volpati said.

The researchers do not expect to see their method compete with high-speed silicon computers, but it could enable inexpensive low-power devices to be fabricated in the near future for use in analog signal processing, for example. The long-term possibilities are unimaginable. The study has opened up a whole new field of research for exploration.

“Our approach shows that a small amount of material can replace a complex electrical circuit if it’s trained to do the job required,” Volpati said. “Just as living organisms have evolved in nature to perform complex tasks with remarkable ease, we’ve shown that non-biological material can also evolve.”

The article “Evolution of electronic circuits using carbon nanotube composites”, published in Scientific Reports, can be read at