US cracks code for brain-like supercomputers with ultrafast microscopy

Researchers have made use of novel ultrafast electron microscopy techniques to track millisecond changes in a material during electrical pulsing.

A team at the US Department of Energy’s (DOE) Argonne National Laboratory (ANL) is conducting these experiments to develop a more energy-efficient form of next-generation supercomputing using artificial neural networks.

Thanks to the new technique, they determined two previously unobserved ways in which electricity can manipulate the state of the charge density waves.

According to the team, the “melting response mimics how neurons are activated in the brain, while the vibrational response could generate neuron-like firing signals in a neural network,” said Daniel Durham, a postdoctoral researcher at ANL, in a statement .

Energy-efficient supercomputing

Supercomputers of today need so much energy that it would be enough to power thousands of households. In response, scientists are using artificial neural networks to create next-generation supercomputing that is more energy-efficient.

These networks emulate the primary building blocks of the human brain, neurons. Certain materials, such as charge density waves, could be utilized to achieve this imitation.

Electrons, which are negatively charged particles, travel in correlated patterns that resemble waves. These patterns are called charge density waves.

The resistance to electron mobility in the material is increased by the charge density waves. The resistance could be turned on and off quickly if the waves could be controlled.

Then, researchers say this characteristic could be used for ultraprecise sensing and more energy-efficient processing. But how the switching happens is unclear, especially considering that the waves transition between states in a mere 20 billionth of a second.

Charge density wave control

Researchers developed a novel microscopy technique to study charge density waves.

Using the ultrafast electron microscope at the Center for Nanoscale Materials, a DOE Office of Science user facility, they observed the nanosecond dynamics in a material known as 1T-TaS2, a tantalum sulfide that forms charge density waves at room temperature.

The team attached two electrodes to a flake of 1T-TaS2 to generate electrical pulses. Contrary to previous assumptions, the ultrafast electron microscope revealed two unexpected findings.

First, the charge density waves melted due to the heat from the injected current, rather than the charge current itself, even during nanosecond pulses. Second, the electrical pulses induced drum-like vibrations in the material, altering the arrangement of the waves.

Researchers say that these findings highlight new ways in which electricity can manipulate the state of charge density waves. The melting response resembles neuron activation in the brain, while the vibrational response could simulate neuron-like firing signals in a neural network.

According to the team, the technique provides a new approach to studying electrical switching processes. 1T-TaS2’s nanoscale properties make it a promising material for next-generation microelectronic devices.

According to Charudatta Phatak, a materials scientist and deputy division head at ANL, this novel approach provided results with broad potential for energy-efficient microelectronics.

“Understanding the fundamental mechanisms of how we can control these charge density waves is important because this can be applied to other materials to control their properties,” said Phatak.

The details of their research were published in the journal Physical Review Letters.