PORTLAND, Ore. — Learning is usually associated with software — as in "deep learning." Now U.K. researchers have found a way to teach nanotube circuitry to learn its proper functions without changing its random pattern. Next they plan to "evolve" polymer-nanotube composite circuits with the help of a liquid crystal that allows them to change their underlying patterns. Separately, U.S. researchers have found a way to build the random patterned base automatically.
"We have already shown the optimization of simple logic gates (including a half adder circuit) within the polymer/nanotube composite. In the future we are expanding the search for other functions," Mark Massey, Research Associate at Durham University (U.K.) told EE Times. "One of particular interest is the evolution of a neuron within the material. Another interesting application is that of classifying sets of data — another problem that we are working on."
The method works by laying down a random pattern of nanotubes in the polymer composite, then patterning numerous metallic electrode pads on the input side and output side (see photo). The desired input voltages are then applied to the input electrodes, and the voltages are measured at the output electrodes. The extra additional pads then have adjustable voltages applied to them until the desired function is realized, whereupon they are fixed.
Nanotube circuits (middle) learn their functions by applying the corresponding voltages to the electrodes (around edge).
(Source: Durham University)
"An input is applied to the input pads, then the output is sampled and compared with the desired output for that given input. Configuration voltages are applied to the additional pads to train or 'evolve' the material," Massey told us. "These configuration voltages are adjusted by a computer algorithm during each iteration to adjust the material until it gives the desired output."
So far the team is not trying to change the random patterning of the nanotube/polymer composite, but only learn the pathways that are already there that will realize the function they want it to perform.
"We are not evolving or training a circuit in the conventional sense, the material is a combination of pathways with differing conductivities / insulating regions. This complex network allows us to exploit it for our work," Massey told us. "We can image the overall network with electron microscopy, but this does not reveal the actual 'circuit', but only its physical structure."
Their next step, however, is to submerge the nanotube-polymer composite into liquid crystals that will allow the underlying nanotube circuits to realign, permitting them to evolve into their proper function.
"Next we want to use liquid based materials, where we do expect to see a change in the structure and self organization," Massey told us.
Cleaning the Slate
The biggest problem with forming such nanotube films in the first place is the fact that some nanotubes are metallic and some of semiconducting, forcing most researchers to sort them before forming a film. Various methods have been tried solve this problem, but none has been both low cost and effective. Now researchers at the University of Illinois at Urbana-Champaign (ULIC) claim to have invented a method that is both cheap and 100 percent effective.
It works by placing the film to be "cleansed" of metallic nanotubes in contact with a metal substrate. Then a heat sensitive polymer film is placed on the other side of the film. When an voltage is passed through the metal substrates is heats up the metallic nanotubes — which conduct current — just enough to cause a trench to rupture in the polymer overcoat revealing the metallic nanotube's location (see picture). A standard desktop procedure can then remove all the metallic nanotubes at once and chemically dissolve the polymer overcoat so the 100 percent purified nanotube film can be transferred to its final substrate.
— R. Colin Johnson, Advanced Technology Editor, EE Times