Researchers from ETH Zurich, the University of Zurich, and Empa built a new memristor that can operate in multiple modes and could potentially be used to mimic neurons in more applications.
“There are different operation modes for memristors, and it is advantageous to be able to use all these modes depending on an artificial neural network’s architecture,” said Rohit John, a postdoc at ETH. “But previous conventional memristors had to be configured for one of these modes in advance.” The new memristors can easily switch between two operation modes while in use: a mode in which the signal grows weaker over time and dies (volatile mode), and one in which the signal remains constant (non-volatile mode). “These two operation modes are also found in the human brain.”
The new memristors are made from halide perovskite nanocrystals, a semiconductor material often used in photovoltaic cells. “The ‘nerve conduction’ in these new memristors is mediated by temporarily or permanently stringing together silver ions from an electrode to form a nanofilament penetrating the perovskite structure through which current can flow,” said Maksym Kovalenko, a professor at ETH.
The team noted, “This process can be regulated to make the silver-ion filament either thin, so that it gradually breaks back down into individual silver ions (volatile mode), or thick and permanent (non-volatile mode). This is controlled by the intensity of the current conducted on the memristor: applying a weak current activates the volatile mode, while a strong current activates the non-volatile mode.”
“To our knowledge, this is the first memristor that can be reliably switched between volatile and non-volatile modes on demand,” added Yiğit Demirağ, a doctoral student at University of Zurich and ETH.
The researchers tested 25 of these new memristors and carried out 20,000 measurements with them. They were able to simulate a computational problem on a complex network that involved classifying a number of different neuron spikes as one of four predefined patterns.
The memristors will need further optimization before they could be used in computers, but they could also be used for research in neuroinformatics, said Giacomo Indiveri, a professor at the Institute for Neuroinformatics of the University of Zurich and ETH Zurich. “These components come closer to real neurons than previous ones. As a result, they help researchers to better test hypotheses in neuroinformatics and hopefully gain a better understanding of the computing principles of real neuronal circuits in humans and animals.”
Improved heat conductance in silicon nanowires
Researchers from University of California Berkeley, Rice University, University of Massachusetts-Amherst, and Tsinghua University created single-isotope silicon nanowires that can conduct much more heat than is typical of silicon.
Natural silicon’s poor ability to conduct heat is due to the material’s three different isotopes. About 92% of silicon consists of the isotope silicon-28, which has 14 protons and 14 neutrons; around 5% is silicon-29, weighing in at 14 protons and 15 neutrons; and just 3% is silicon-30, a relative heavyweight with 14 protons and 16 neutrons, explained Joel Ager, who holds titles of senior scientist in Berkeley Lab’s Materials Sciences Division and adjunct professor of materials science and engineering at UC Berkeley. When heat-carrying phonons bump into silicon-29 or silicon-30, they can change direction and slow down.
In addition, heat transfer can be worse in silicon nanowires, such as those used in gate-all-around FETs, because the rough surface caused by chemical processing adds further disruption to the travel of phonons.
Bulk isotopically pure silicon-28 had previously been investigated for its ability to conduct heat better, but the increase, about 10%, wasn’t worth the additional processing cost.
The researchers decided to apply these earlier findings to silicon nanowires. Using a technique called electroless etching, the team made natural silicon and silicon-28 nanowires 90 nanometers in diameter.
To measure the thermal conductivity, the researchers suspended each nanowire between two microheater pads outfitted with platinum electrodes and thermometers, and then applied an electrical current to the electrode to generate heat on one pad that flows to the other pad via the nanowire.
“We expected to see only an incremental benefit — something like 20% — of using isotopically pure material for nanowire heat conduction,” said Junqiao Wu, a faculty scientist in the Materials Sciences Division and professor of materials science and engineering at UC Berkeley.
Instead, they found the Si-28 nanowires conducted heat 150% better than natural silicon nanowires with the same diameter and surface roughness.
Part of the improvement came from the formation of glass-like silicon dioxide on the surface of the Si-28 nanowire. Computational simulation showed that the absence of isotope silicon-29 and silicon-30 prevented phonons from escaping to the surface, where the silicon dioxide layer would drastically slow down the phonons. This in turn kept phonons on track along the direction of heat flow.
“This was really unexpected. To discover that two separate phonon-blocking mechanisms — the surface versus the isotopes, which were previously believed to be independent of each other — now work synergistically to our benefit in heat conduction is very surprising but also very gratifying,” Wu said.
Wu said that the team next plans to investigate how to “control, rather than merely measure, heat conduction in these materials.”
Cooling with copper coatings
Researchers from the University of Illinois at Urbana-Champaign and University of California Berkeley propose a method to cool electronics that is more compact than traditional heat spreaders.
It uses a copper coating that completely covers the device, said Tarek Gebrael, a University of Illinois at Urbana-Champaign (UIUC) Ph.D. student in mechanical engineering, including “the top, the bottom, and the sides… a conformal coating that covers all the exposed surfaces.”
This removes the need for a thermal interface material, the researchers note, and entirely replaces the heat spreader and heat sink. The coatings can be used both in air and in water, for immersion cooling applications.
“In our study, we compared our coatings to standard heat sinking methods,” Gebrael said. “What we showed is that you can get very similar thermal performance, or even better performance, with the coatings compared to the heat sinks.” In addition, a device using the new solution is dramatically smaller than one using heat sinks. “And this translates to much higher power per unit volume. We were able to demonstrate a 740% increase in the power per unit volume.”
“This technology bridges two separate thermal management approaches: near-junction device-level cooling, and board-level heat spreading,” added Nenad Miljkovic, an associate professor of mechanical science & engineering at UIUC.
Gebrael gave an example of the space-saving potential. “Let’s say you have multiple printed circuit boards. You can stack many more printed circuit boards in the same volume when you are using our coating, compared to if you are using conventional liquid- or air-cooled heat sinks.” Applications could include power electronics cooling, data center thermal management, and electric machine cooling.
The researchers are still investigating the coatings’ reliability and durability, including reliability in boiling water, boiling dielectric fluids, and high-voltage environments. They will also implement the coatings on full-scale power modules and GPU cards, whereas they used only simple test boards in the initial work.