IBM (since a lot of folks out there have asked, we’re going to put this here: IBM stands for International Business Machines when you’re talking about the technology industry) has finally managed to produce a phase-change capable neuron that is entirely artificial in nature.
More specifically, IBM researchers have managed to build what many are calling as the beginnings of an artificial human brain (or at least a brain like computer).
Moreover, the artificial neuron is constructed from materials that have been trusted by the scientific community for a long time and can scale down to sizes that are microscopic in length.
Of course, this isn’t the first time human brains have been compared with computers or supercomputers if we’re talking about actual potential.
But this time, IBM has taken concrete steps beyond just debates and theories by building an artificial neuron that works in a similar way to a human brain. Not quite at the same scale but it still manages to mimic a bunch of neurons in the human brain.
If there is one thing that readers should take away from this news, it’s that IBM is at the forefront of this particular technology. The company seems determined to make comparisons between a human brain and a computer, more than just comparisons. With this artificial neuron, it is safe to say that IBM has taken a step further in the future and has followed the comparisons with some actual work.
IBM’s research centers in Zurich produced 500 of these artificial neurons and then applied the technology to that bunch to imitate a signal transfer.
The artificial neurons managed to carry out the process in a remarkably similar fashion to how an actual human brain(or an organic brain to be on the safe side) would have sent a signal transfer.
But this isn’t the only time researchers have tried to dabble with the idea of an artificial brain. Other studies done on synthetic signaling have exhibited that the real breakthroughs come only when the elements used to form the artificial neurons can be scaled down to microscopic levels.
Not only that, but these microscopic sized artificial neurons also have to work properly on the microscopic level to be considered completed.
And that is exactly why IBM’s achievement is seen as crucial at the moment.
IBM’s imitative neurons are built with materials that are well-known materials regarding stability and reliability.
These materials can scale down to the nanoscale without becoming useless. Not only that, but these materials can also get activated with extremely low energy as has been pointed out by Ars Technica in their recent piece of IBM’s artificial neurons.
Perhaps, we shouldn’t also forget the fact that it is efforts like these from companies like IBM that brain-like computers are now being considered as a definite possibility instead of a science fiction gimmick.
One of the other benefits of these artificial synapses is that they use less power than real organic neurons.
The big reason, if not the most major reason, why people haven’t been able to see a brain like computer outside specialized labs is because the artificial synapses that make a brain like computer work need to eat up a lot of power.
Way more power than the actual thing. And because of researchers like those at IBM’s Zurich research centers and others; low energy consumption from artificial neurons is not a dream anymore.
Researchers, not related to IBM’s Zurich branch, had been able to build nanowire synapses that consumed about 1.23 femtojoules of energy which is extremely low energy demand from an artificial synapse.
Researchers were able to achieve that by wrapping two organic materials together and then using that wrap to release and trap ions.
That process is pretty much how real nerve fibers work inside an organic brain.
But scientists need to go even further if brain-like computers are to become a reality within a reasonable period of time. Researchers would need even thinner nanowires and would need highly advanced 3D printing techniques to form structures that are more like organic brains regarding their function.
Remember, biological neurons contain membranes that act like signal gates. These signal gates require a particular amount of energy to work properly.
The artificial neurons created by IBM doesn’t have any membranes because IBM researchers have replaced membranes with a square that is made of Germanium Antimony Tellurium compound. It’s also called GST which is a common material in optical disks.
In other words, IBM has replaced the role of membranes in organic neurons with GST squares in its artificial ones.
Now, Germanium Antimony Tellurium is a fascinating material. If you heat the material for a given period of time, it changes its physical phase.
The Germanium Antimony Tellurium material goes from an amorphous phase (in which the material acts as an insulator) to a crystalline phase (in which Germanium Antimony Tellurium material becomes a conductor).
More simply put, researchers were able to pass a signal through the “fake” membrane (that is Germanium Antimony Tellurium square material) when it was shot with sufficient amount of electricity.
That happened because the GST material changed into its crystal phase. Afterward, the Germanium Antimony Tellurium material changed back to its amorphous phase.
As impressive as that sounds, it wasn’t the only trait scientists were looking for when they selected the Germanium Antimony Tellurium material.
Researchers need their artificial neuron to have stochasticity. In order to have the major characteristics of an organic neuron in an artificial neuron, researchers needed some form of randomness in the timing of signals that were fired.
The Germanium Antimony Tellurium material, IBM researchers say, enabled them to achieve this effect because the material never went back to the same configuration as before.
This feature of Germanium Antimony Tellurium material allows a group of IBM’s artificial neurons to carry out actions that they otherwise couldn’t if the results of those actions were always predictable.
Now, remember, that the organic brain works with great efficiency because of its features like parallel processing design.
In order to produce computers that were able to mimic that behavior, scientists needed to apply this style of approach to processing sensory data and decision-making processes.
These artificial neurons will help these scientists achieve that objective. The folks at Ars Technica have also noted that the construction of such artificial neurons was actually the easier part.
The tricky part, which scientists haven’t addressed yet, would be to write software applications that could work with that kind of a setup.