The idea of artificial intelligence has been the fascination of science fiction since its inception. While science fiction’s depiction of artificial intelligence is often insidious, it brings with it the promise of technologies that are more effective and more efficient. What was once fiction may be reality thanks to a group of researchers at the University of Massachusetts Amherst. Although countless technologies have attempted to emulate the computing power of the human brain, few have come as close as the diffusive memristor. In essence, they succeeded in the production of a man-made neuron that has the unique ability to mimic the connections present in the human brain.
This new technology has the potential to create a new class of neuromorphic computers that bring with them the promise of energy efficiency and an increased capacity for learning. Previous efforts to duplicate the phenomenon of biological synapses have had limited success. The secret of this new class of artificial neurons lies in their ability to imitate the “synaptic Ca2+ dynamics that occur in biological systems” (Wang et al. 2016). The ability of memristors to incorporate these types of Ca2+ dynamics gives rise to both long- and short-term plasticity. Synaptic plasticity, the ability of synapses to strengthen or weaken over time in response to their level of activity, has been linked to postsynaptic calcium release. Rather than using calcium, memristors make use of silver nanoclusters.
The basis on which the device functions can best be understood by investigating its structure. The diffusive memristor consists of two platinum or gold electrodes that sandwich a film with embedded silver nanoclusters. When a current passes from one electrode to the other, the silver nanoparticles begin bridging the gap between the electrodes. This results in a conductive channel that dramatically increases conductivity, and thus the speed of signal transmission. This conductive channel is only maintained in the presence of an electric current. Once the current stops running through the system, the silver particles relax back to their ground state and the conductive channel is broken.
Now we can observe how this tendency of the memristors can give rise to plasticity. If the time between pulses is shorter than the time it takes for the silver particles to relax, more particles are pushed into the gap. Over time, this can result in the formation of a fully conductive bridge. The researchers that observed this phenomenon called it “paired-pulse facilitation, or PPF” (Wang et al. 2016). PPF is similar to the way in which neurons increase the fidelity and strength of the signals they transmit in short-term plasticity. On the other hand, if a pulse excites the silver particles for too long, they being to migrate to one electrode. This decreases the number of silver nanoparticles in the gap between the two electrodes, and results in slower signal transmission. This phenomenon was deemed “paired-pulse depression, or PPD” (Wang et al. 2016). PPD is analogous to the refractory period that occurs after exciting a neuron.
Finally, when diffusive memristors were assembled into simple networks they gave rise to spike-timing-dependent plasticity. In other words, memristors that fired together reinforced each other by increasing the speed and strength at which a signal traveled through that frequently-used network. Networks of memristors that were not used as often had weaker signal transmission that those that were used frequently. This phenomenon arises without the need for complex pulse engineering and can lead to long-term plasticity. The remarkable promise that this technology shows in these early stages lends itself to the usefulness of its possible applications.
Summarizing a research article in a way that can be understood by a larger audience is a task I severely underestimated. Primary literature often has a narrow audience of highly trained experts and students, due to its reliance on jargon. For this reason, I had to carefully comb through the article and decide what I wanted to include in my summary. Often, I found myself having to look up words and concepts so I could better explain these crucial elements in my summary. It was a challenge to find a balance between including critical parts of the study and excluding nonessential, convoluted details.
I based what I included in my article in part on what the news article included from the study. While the news article did an adequate job summarizing the important findings of the study, it lacked essential background knowledge and the mechanism behind memristor plasticity. When I first read the article, I had to look up several key terms, like plasticity and spike-timing-dependent plasticity. This could be because the article is intended for an audience that has a scientific background however, it still detracted from the flow of the article. This problem could easily be solved by including short definitions of key terms. This is why I chose to include definitions of words that a general audience may not know in my summary. Furthermore, the article lacked an explanation for the mechanism behind memristor plasticity. While this may be a personal preference, I found the mechanism behind the machinery to be an important discovery that was easy to simplify.
Putting myself in the shoes of a journalist has developed my respect for the unique challenges they face. Their task is to provide an accurate summary of the material in a study while capturing the attention of a large audience. Understanding a research article thoroughly enough to decided what should be summarized and what should be excluded requires a great deal of experience and intuition.
Wang Z., Joshi S., Savel’ev S. E., Jiang H., Midya R., Lin P., Hu M., Ge N., Strachan J. P., Li Z., Wu Q., Barnell M., Li G. L., Xin H. L., Williams R. S., Xia O., Yang J. J. (2016 March 23). Nature Materials, 16, 101-108. Retrieved Mary 10, 2017.
The research article had no link to it since it was sent to me in the form of screenshots.