Since the advent of transistor-based computers, computer speed and complexity have increased at an exponential rate. The central processing unit of a modern computer contains over a billion transistors and can perform hundreds of billions of instructions each second. However, even that level of performance is still soundly beaten in many simple but important tasks by the human brain, which processes merely a thousand instructions each second.

Part of the brain’s advantage originates from the parallel nature of its computational circuits. Anirban Bandyopadhyay at the National Institute for Materials Science in Japan along with colleagues in Japan and the USA have now demonstrated a physical platform for computation that is capable of high degrees of such parallelism.1

The researcher’s constructed their platform, a ‘cellular automaton network’, as a pair of molecular layers on a gold substrate. Scanning the tip of a scanning tunneling microscope (STM) across the layers caused individual molecules to switch among four different conducting states, each producing a different strength of connection with the node's neighbors. Information can be encoded in this network in the state of each molecular node, and in the way nodes are connected. While such networks had been demonstrated previously, the molecular version built by Bandyopadhyay and his colleagues allows for a greater degree of communication among neighboring molecular nodes, potentially allowing for faster computation.

Fig. 1: Schematic illustration of how a network of molecules (background) can simulate diffusion (foreground).

After the network settles into an initial configuration, it can be made to evolve — and compute — by applying a voltage trigger to the STM tip. Choosing a particular initial state, applying a particular voltage, or introducing charge into the right location, the researchers could get the network to perform traditional logic gate functions of the type found in modern-day computers. They also simulated the network's ability to solve problems like diffusion and cancer cell growth without the use of logic gate functions (Fig. 1).

The cellular automaton network is constructed without placing and wiring each node in the network individually, and so represents a simpler pathway to computation using cellular automata compared with previous approaches. Bandyopadhyay says the research team plans to scale up the number of nodes involved, as well as extend the structure to three dimensions. “Ultimately, these ‘nano brains’ may be able to solve problems that are intractable using current technologies, such as predicting natural calamities that are decades away.”