THE BLOG

Building Artificial Brains With Electronics

04/21/2015 03:18 pm ET | Updated Jun 19, 2015

There are many things about the human brain that computer engineers envy and try to mimic. The grey, wrinkled flesh that sits inside our cranium is capable of massive parallel information processing while consuming very little power. It does so by means of an intricate web of neurons where each neuron acts as a tiny unit of memory and processor rolled into one: neurons "remember" what exited them and respond only when levels of excitation from other neurons exceed a certain value. Wouldn't it be great if we could build computer circuits that did exactly that?

Perhaps we are very close to do this thanks to "memristors", electronic components that behave similar to neurons. Originally envisioned by electronics pioneer Leon Chua, a memristor is a variable resistor that "remembers" its last value when the power supply is turned off. Electronic engineers regard the memristor as a "fundamental circuit element", just like a capacitor, an inductor, or indeed a "normal" resistor. In 2012 a team of researchers from HLR Laboratories and the University of Michigan made news by announcing the first functioning memristor array built on conventional chips. Since then much research effort has gone into developing further the technology behind manufacturing these amazing electronic components.

Only last week scientists at Northwestern University published in Nature Nanotechnology how they managed to transform a memristor from a 2-terminal to a 3-terminal component. This is an important breakthrough because when engineers design electronic circuits they like to create things that can execute "logical" functions; think of them as arithmetic operations; and to do so the minimum requirement is to have two inputs and one output in whatever circuit one designs. Using ultra thin semiconductor materials the scientists were essentially capable of manipulating a memristor so that it could process inputs from two electrical sources, much like a neuron is "tuned" by becoming excited by more than one other neuron.

Other scientists have experimented with connecting a memristor with a capacitor, and getting a so-called "neuristor". This electronic combo mimics the ability of biological neurons not only in remembering what excited them but also in "firing" their own electrochemical message when they reach a certain excitation value. Here's how a neuristor works: as an electric current passes through the memristor its resistance increases, while the capacitor gets charged. However, the memristor has an interesting property: at a given current threshold its resistance suddenly drops off, which causes the capacitor to discharge. This discharge at a certain threshold is akin to a single neuron firing.

Memristors and neuristors are elementary circuit elements that could be used to build a new generation of computers that mimic the brain, the so-called "neuromorphic computers". These computers will differ from conventional architectures in a significant way. They will mimic the neurobiological architecture of the brain by exchanging spikes instead of bits. It is estimated that in order to simulate the human brain on a conventional computer we will need supercomputers a thousand times more powerful that the ones we have today. This requirement is stretching the limits of the current technology in chip manufacturing, and for many it lies beyond the upper forecast of Moore's Law. Memristors and neuromorphic circuits could be a game changer because of their enormous potential to process information in massively parallel way, just like the real thing that invented them.