Heraeus and RWTH Aachen University develop adaptive artificial synapses

May 11, 2020

  • Application in deep learning and artificial intelligence: so-called memristive components are regarded as ideal candidates for computers modeled on the brain
  • Influence of foreign atoms in the oxide layer so far not considered by experts
  • Researchers achieve new switching time record

HANAU, Germany - Researchers from the technology group Heraeus and the Jülich Aachen Research Alliance (JARA) have discovered how the switching properties of artificial synapses can be specifically influenced. Memristors - electrical components that can be switched between a low and high value like a "resistance with memory" - are considered a promising alternative to conventional components in computer chips. In principle, they function like a synapse of the biological nervous system and also get by with extremely little electricity. "Memristive elements are regarded as ideal candidates for neuro-inspired computers modelled on the brain, which are arousing great interest in connection with deep learning and artificial intelligence," explains Dr. Ilia Valov from the Peter Grünberg Institute (PGI-7) at Research Centre Jülich.

So far unnoticed by experts: The secret is foreign atoms in the oxide layer.

Inforgraphics: Synapse vs. Memoristor Copyright: Forschungszentrum Jülich / Tobias Schlößer
Synapses, the connections between neurons, have the ability to transmit signals with varying degrees of strength when they are excited by a quick succession of electrical impulses. One effect of this repeated activity is to increase the concentration of calcium ions, with the result that more neurotransmitters are emitted. Depending on the activity, other effects cause long-term structural changes, which impact the strength of the transmission for several hours, or potentially even for the rest of the person’s life. Memristive elements allow the strength of the electrical transmission to be changed in a similar way to synaptic connections, by applying a voltage. In electrochemical metallization cells (ECMs), a metallic filament develops between the two metal electrodes, thus increasing conductivity. Applying voltage pulses with reversed polarity causes the filament to shrink again until the cell reaches its initial high resistance state.

The components with which the research team has conducted systematic experiments consist of an ultra-thin, amorphous fused silica layer (silicon dioxide) between a platinum and a copper electrode. In the current issue of the open-access journal Science Advances, the researchers describe how the switching properties of memristic components can be specifically influenced. According to this, the purity of the central oxide layer is the decisive factor: the researchers have deliberately introduced foreign atoms into the 99.999999 percent pure (8N) silicon dioxide (doping). "By introducing foreign atoms, we influence the solubility and transport properties of the thin oxide layers," explains Dr. Christian Neumann of the Heraeus technology group. This effect had previously been unnoticed by experts. It can be used specifically for the design of memristic systems, similar to the doping of semiconductors in information technology. The specifically doped glasses were specially developed and manufactured by fused silica specialists Heraeus Conamic, which also holds the patent for the materials.

With the help of the insights gained, manufacturers can now specifically develop memristive elements with the desired functions. This is because memristive elements behave in a similar way to their biological role model: the brain's ability to learn and remember is largely due to the fact that the connections between nerve cells become stronger, so to speak, when they are used frequently. In artificial synapses, conductivity increases with the number of incoming voltage pulses. The changes can also be reversed by applying voltage pulses of opposite polarity. The higher the doping, i.e. the number of foreign atoms in the oxide layer, the slower the resistance of the elements changes with the number of incoming voltage pulses and the more stable the resistance remains. "We have thus discovered a way to construct differently excitable types of artificial synapses," explains Ilia Valov.

Proof of theory: new record in switching time achieved

There are different variants of memristive components such as electrochemical metallization cells (ECMs) or valence change cells (VCMs). Based on their series of experiments with ECMs, the researchers were able to show that the switching times change with the amount of foreign atoms. If the middle layer consists of 8N silicon dioxide, the memristive component switches in 1.4 nanoseconds. So far, the fastest value ever measured with ECMs was about 10 nanoseconds. By doping the oxide layer of the components with up to 10,000 ppm (parts per million) of foreign atoms, the scientists have specifically extended the switching time into the range of milliseconds. Based on generally applicable theoretical considerations, supported by experimental results documented in the technical literature, the research team is convinced that the doping effect occurs not only in ECMs and VCMs, but in all memristive elements.

All research results on this topic can be found in the original publication:  Design of defect-chemical properties and device performance in memristive systems, Lübben et al., Science Advances, 2020