Power/Performance Bits: Mar. 6

Neural network chip
Neural networks are both slow and consume a lot of power. This made researchers at MIT examine the important aspects of the nodes within a neural network and to see how each part of the computation could be improved. The outcome was a dedicated chip that increases the speed of neural-network computations by three to seven times over its predecessors, while reducing power consumption 94 to 95%.

The gains come from a number of elements in the chips architecture. “The general processor model is that there is a memory in some part of the chip, and there is a processor in another part of the chip, and you move the data back and forth between them when you do these computations,” explains Avishek Biswas, an MIT graduate student in electrical engineering and computer science. He notes that this data transfer is the dominant contributor to energy consumption. So instead, they implement the calculation of the multiply/accumulate operations within the memory.

The second modification they made was to assume that all weights are either 1 or -1. They indicate that recent research suggests that neural nets trained with only two weights should lose little accuracy — somewhere between 1 and 2 percent. This, coupled with converting a nodes input value into analog signals, means that summing the products is just a matter of combining the voltages. Only the combined voltages are converted back into a digital representation and stored for further processing.

The researchers hope that this development will enable handheld devices to incorporate neural networks rather than having such computations done in the cloud.

Li-ion battery sensor
Lithium-ion batteries could be safely charged up to five times faster than the current recommended charging limits, according to researchers at the University of Warwick.

The team developed a range of methods of that allows direct, highly precise internal temperature and “per-electrode” status monitoring of lithium-ion batteries of various formats and destination. These methods can be used during a battery’s normal operation without impeding its performance and it has been tested on commercially available automotive-class batteries. The data acquired by such methods is much more precise than external sensing, according to the researchers.

The in-situ battery sensing employs miniature reference electrodes and Fibre Bragg Gratings (FBG) threaded through bespoke strain protection layer. An outer skin of fluorinated ethylene propylene (FEP) was applied over the fiber, adding chemical protection from the corrosive electrolyte. The result is a device that can have direct contact with all the key parts of the battery and withstand electrical, chemical and mechanical stress inflicted during the batteries operation while still enabling precise temperature and potential readings.

A lithium battery temperature sensor. (Source: WMG, University of Warwick)

To avoid damage from battery overheating or overcharging, manufacturers stipulate a maximum charging rate or intensity for batteries based on what they think are the crucial temperature and potential levels to avoid. However until now internal temperature testing (and gaining data on each electrode’s potential) in a battery has proved either impossible or impractical without significantly affecting the batteries performance.

According to the data gathered by the researchers, commercially available lithium batteries today could be charged at least five times faster than the current recommended maximum rates of charge.

Said Tazdin Amietszajew, a researcher at Warwick, “This technology is ready to apply now to commercial batteries but we would need to ensure that battery management systems on vehicles, and that the infrastructure being put in for electric vehicles, are able to accommodate variable charging rates that would include these new more precisely tuned profiles/limits.”

Lead-free perovskites
Researchers at Brown University and University of Nebraska – Lincoln (UNL) developed a new titanium-based material for making lead-free, inorganic perovskite solar cells.

“One of the big thrusts in perovskite research is to get away from lead-based materials and find new materials that are non-toxic and more stable,” said Nitin Padture, a professor at Brown and director of the Institute for Molecular and Nanoscale Innovation. “Using computer simulations, our theoretician collaborators at UNL predicted that a class of perovskites with cesium, titanium and a halogen component (bromine or/and iodine) was a good candidate. The next step was to actually make a solar cell using that material and test its properties, and that’s what we’ve done here.”

“Titanium is an abundant, robust and biocompatible element that, until now, has been largely overlooked in perovskite research,” said Padture.

Titanium is an attractive choice to replace the toxic lead in the prevailing perovskite thin film solar cells. (Source: Padture Lab / Brown University)

The team made semi-transparent perovskite films that had bandgap of 1.8 electron volts, which is considered to be ideal for tandem solar applications. The material had a conversion efficiency of 3.3%, which is well below that of lead-based cells (those have reached efficiencies up to 22%), but a good start for an all-new material, the researchers say.

The titanium-perovskite also has an open-circuit voltage of over one volt. Other lead-free perovskites generally produce voltage smaller than 0.6 volts.

The researchers say that material’s relatively large bandgap compared to silicon makes it a prime candidate to serve as the top layer in a tandem solar cell. The titanium-perovskite upper layer would absorb the higher-energy photons from the sun that the lower silicon layer can’t absorb because of its smaller bandgap. Meanwhile, lower energy photons would pass through the semi-transparent upper layer to be absorbed by the silicon, thereby increasing the cell’s total absorption capacity.

“Tandem cells are the low-hanging fruit when it comes to perovskites,” Padture said. “We’re not looking to replace existing silicon technology just yet, but instead we’re looking to boost it. So if you can make a lead-free tandem cell that’s stable, then that’s a winner. This new material looks like a good candidate.”