As a 2013 article on in The Economist said about neuromorphic computing (meaning microprocessors configured more like human brains than like traditional chips): “Computers will help people to understand brains better. And understanding brains will help people to build better computers.” In that general context, Professors Joshua Yang and Qiangfei Xia of our Electrical and Computer Engineering (ECE) Department led a 24-person international team of researchers that has just published the second of two defining papers on neuromorphic computing, which mimics neuro-biological architectures present in the nervous system in order to build better computing systems.
As the researchers explained, neuromorphic computing is one of the most promising transformative computing technologies currently under intensive study. The research team’s paper posted on February 8, 2018, in the prestigious scientific journal Nature Electronics represented the “sister paper” to a 2017 piece in Nature Materials.
As Yang explained, “The Nature Materials paper is about realizing artificial synapses, and this Nature Electronics paper is about realizing artificial neurons using memristive devices. In this paper, we have further integrated the artificial synapses and neurons (two major building blocks of a neural network) into the first fully memristive neural network, on which we have demonstrated pattern classification with unsupervised learning functions.”
The new paper in Nature Electronics is titled: "Fully memristive neural networks for pattern classification with unsupervised learning."
As The Economist editorialized about neuromorphic computing in its August 3, 2013, edition: “Though the brain-as-computer is, indeed, only a metaphor, one group of scientists would like to stand that metaphor on its head. Instead of thinking of brains as being like computers, they wish to make computers more like brains. This way, they believe, humanity will end up not only with a better understanding of how the brain works, but also with better, smarter computers.”
Nature also published an editorial on February 6, 2018, titled “Big data needs a hardware revolution,” which stated that “Software companies make headlines, but research on computer hardware could bring bigger rewards.”
An important advancement in the evolution of neuromorphic computing is being carried out by Yang and Xia in their work being reported in their recent publications in the Nature research journals. As the team of 24 researchers explained in its February 8 Nature Electronics paper, “Neuromorphic computers comprised of artificial neurons and synapses could provide a more efficient approach to implementing neural network algorithms than traditional hardware.”
Recently, noted the researchers, scientists have developed artificial neurons based on memristors, but these possess limited bio-realistic dynamics and no direct interaction with the artificial synapses in an integrated network.
“Here we show that a diffusive memristor based on silver nanoparticles in a dielectric film can be used to create an artificial neuron with stochastic leaky integrate-and-fire dynamics and tunable integration time, which is determined by silver migration alone or its interaction with circuit capacitance,” as the researchers explained their groundbreaking study. “We integrate these neurons with nonvolatile memristive synapses to build fully memristive artificial neural networks for the first time. With these integrated networks, we experimentally demonstrate unsupervised synaptic weight updating and pattern classification.”
The new study builds upon the pioneering research reported in the 2017 Nature Materials sister paper. At that time, Yang described it as part of “our collaborative work on a new type of memristive device that can faithfully emulate the functionality of a biological synapse.” Xia added that “This work opens a new avenue of neuromorphic computing hardware based on memristors.”
The researchers believed then that their new approach held considerable advantages over the traditional CMOS approach. Compared to the CMOS approach, they disclosed, “The two-terminal diffusive memristor will lead to a significant reduction in footprint, complexity, and energy-consumption.”
As the researchers said in 2017, “Memristors have become a leading candidate to enable neuromorphic computing by reproducing the functions in biological synapses and neurons in a neural network system, while providing advantages in energy and size.”
With this exciting new paper in Nature Electronics, the research group has made a significant new step toward that revolution in neuromorphic computing. As written in a “News & Views” article titled “Memristors fire away” in the same issue of Nature Electronics that highlights this work, “Neuromorphic computing based on fully memristive neural networks could offer a scalable and lower-cost alternative to existing neural spiking chips based solely on CMOS technology.”
Nature Electronics says it is “interested in the best research from all areas of electronics, incorporating the work of scientists, engineers, and researchers in industry.” Nature Electronics chose research previously reported by COE from this UMass team on “computing engine using large memristor crossbars” as the cover article for its inaugural issue in January of 2018.
In addition to the UMass ECE Department, the 24 authors in the Nature Electronics paper were part of: the Department of Physics, Loughborough University, Loughborough, UK; Hewlett Packard Labs, Palo Alto, CA; the Air Force Research Lab, Information Directorate, Rome, New York; and the Institute of Microelectronics, Tsinghua University, Beijing, China. (February 2018)