Engineers Develop 'Material That Computes,' Recognizes Images

A piezoelectric-based material capable of generating its own power and computing information.

Sep 3 2016, 2:30pm

Image: photo illustration by the author

Naturally, we think of computers as more or less self-contained entities capable of processing information in useful ways. Computers are built from materials like silicon and copper and they are powered by electricity. It's difficult to conceive of the alternative, really, but engineers have a vision of computers that exist as materials themselves. Whereas we might imagine wearable technology as having a computer built into some textile, "materials that compute" imagine that textile as the computer. The computer and the material are one and the same—incapable of disintegration.

Researchers from the University of Pittsburgh led by chemical engineer Anna Balazs and electrical engineer Steven Levitan have developed a primitive implementation of the idea: a responsive, hybrid material capable of both computational pattern recognition and of generating its own power. Their work is described in the current issue of the open-access journal Science Advances.

What Levitan and Balazs imagine is a combination of stimuli-responsive materials and non-conventional computing that can be employed for such tasks as sensing and communications. "One means of achieving these objectives is to integrate the capabilities of energy-transducing, soft materials, such as oscillating chemical gels, and modes of computation, such as oscillator-based computing, which can exploit these materials characteristics," they explain in the paper.

Image: Yan Fang

Levitan and Balazs' computational material is built from functional units composed of polymer gels overlaid with piezoelectric (PZ) elements. As the gels deform under stress—as the material changes shape in response to some stimuli—they undergo an oscillatory reaction known as the Belousov-Zhabotinsky (BZ) reaction. That is, they pulse rhythmically. When the gels start oscillating, they pass this mechanical motion on to the piezoelectric elements, which convert it into electricity. The elements are then linked together by wires, with the effect being a device that "senses, actuates, and communicates," according to the researchers.

To perform pattern recognition, the material is first encoded with a "memory" of certain patterns of numbers in the form of electrical polarities. Each of the piezoelectric-gel junctions is loaded up with a small amount of charge and each particular charge arrangement represents useful information. The patterns to be recognized are input as crude binary images of digits, where each pixel corresponds to an individual junction. So, the number "0" encoded as 60 pixels of information would trigger 60 piezoelectric-gel junctions matching the shape of the numerical digit.

The computational task of the material is to identify correctly which junction-pixel patterns correspond to the numbers that have been preloaded into its memory as polarities. Simply, when the in-memory polarities are synchronized with a certain input, the result is a system-wide stability that can be interpreted as a matched pattern.

"We imposed a collection of input patterns onto different PZ-BZ networks, where each network encompassed a distinct stored pattern," the paper explains. "The network encompassing the stored pattern closest to the input pattern exhibited the fastest convergence time to stable synchronization behavior and could be identified as the winner. In this way, the networks of coupled BZ-PZ oscillators achieved pattern recognition."

The computation occurs slowly, however, for the fundamental reason that the oscillations that the whole contraption depend upon occur slowly (they have a long period). So, wearable computers aren't likely to be breaking encryption keys anytime soon. But in scenarios in which long-term phenomena are being sensed and computed, particularly those involved in monitoring the human body, it should work out well. Similarly, the material may find interesting future applications as sensory skin for robots.

Now, according to Levitan and Balaz, we at least "have fundamental and experimentally realizable design rules for creating materials that compute."