Power/Performance Bits: Aug. 24

Low power AI
Engineers at the Swiss Center for Electronics and Microtechnology (CSEM) designed an SoC for edge AI applications that can run on solar power or a small battery.

The SoC consists of an ASIC chip with RISC-V processor developed at CSEM along with two tightly coupled machine-learning accelerators: one for face detection, for example, and one for classification. The first is a binary decision tree (BDT) engine that can perform simple tasks but cannot carry out recognition operations.

“When our system is used in facial recognition applications, for example, the first accelerator will answer preliminary questions like: Are there people in the images? And if so, are their faces visible?” said Stéphane Emery, head of system-on-chip research at CSEM. “If our system is used in voice recognition, the first accelerator will determine whether noise is present and if that noise corresponds to human voices. But it can’t make out specific voices or words – that’s where the second accelerator comes in.”

The second accelerator, a convolutional neural network, can perform the more complicated tasks of facial recognition and word detection, but consumes more energy. Most of the time only the first accelerator is running, reducing overall power consumption.

The team said the SoC is adaptable to any application where time-based signal and image processing is needed. “Our system works in basically the same way regardless of the application,” said Emery. “We just have to reconfigure the various layers of our CNN engine.”

Flexible MoS2 transistors
Researchers at Stanford University developed a way to manufacture flexible, atomically thin transistors less than 100nm in length.

The process involves placing 2D semiconductor molybdenum disulfide (MoS2) overlaid with small nano-patterned gold electrodes onto a solid slab of silicon coated with glass.

Using the conventional silicon substrate means chemical vapor deposition (CVD) can be used to grow a film of MoS2 three atoms thick. This patterning technique can’t be used with flexible plastic substrates, as they would melt in the high temperatures required.

By first patterning these parts on rigid silicon and allowing them to cool, the researchers were able to apply the flexible material without damage. With a simple bath in deionized water, the entire device stack peels back, now fully transferred to the flexible polyimide.

Illustration of transfer process for 2D semiconductor with nanopatterned contacts (left) and photograph of flexible transparent substrate with transferred structures (right). (Image credit: Victoria Chen/Alwin Daus/Pop Lab)

The researchers said that while entire circuits could be built and then transferred to the flexible material, certain complications with subsequent layers make these additional steps easier after transfer.

“In the end, the entire structure is just 5 microns thick, including the flexible polyimide,” said Eric Pop, a professor of electrical engineering at Stanford.

The team said the device can handle high electrical currents while operating at a low voltage. The gold metal contacts dissipate and spread the heat generated by the transistors while in use, protecting the flexible polyimide.

“This downscaling has several benefits,” said Alwin Daus, a postdoctoral scholar at Stanford. “You can fit more transistors in a given footprint, of course, but you can also have higher currents at lower voltage – high speed with less power consumption.”

The group also built similar transistors using two other atomically thin semiconductors (MoSe2 and WSe2) to demonstrate the broad applicability of the technique and are looking into integrating radio circuitry with the devices.

Batteries without cobalt, nickel
Researchers from Lawrence Berkeley National Laboratory, University of California Berkeley, and Argonne National Laboratory propose a new material for lithium-ion battery cathodes that can eliminate the need for expensive and limited supplies of cobalt and nickel.

The new cathode material is disordered rocksalts with excess lithium, or DRX. It would allow cathodes to be constructed with a variety of metals aside from cobalt and nickel.

“With the current NMC class, which is restricted to just nickel, cobalt, and an inactive component made of manganese, the classic lithium-ion battery is at the end of its performance curve unless you transfer to new cathode materials, and that’s what the DRX program offers. DRX materials have enormous compositional flexibility – and this is very powerful because not only can you use all kinds of abundant metals in a DRX cathode, but you can also use any type of metal to fix any problem that might come up during the early stages of designing new batteries. That’s why we’re so excited,” said Berkeley Lab battery scientist Gerbrand Ceder.

In conventional cathodes, lithium ions travel through the cathode material along well-defined pathways and arrange themselves between the transition metal atoms, usually cobalt and nickel, in orderly layers. But the researchers found that a disordered atomic structure could hold more lithium while allowing for a wider range of elements to serve as the transition metal.

Ceder added that reducing the use of cobalt has been a priority for the U.S. Department of Energy. “The battery industry is facing an enormous resource crunch. Even at 2 TWh, the lower range of global demand projections, that would consume almost all of today’s nickel production, and with cobalt we’re not even close. Cobalt production today is only about 150 kilotons, and 2 TWh of battery power would require 2,000 kilotons of nickel and cobalt in some combination.”

Instead, the team focused on using manganese and titanium in the DRX cathode, which are both more abundant and lower cost than nickel and cobalt.

“Manganese oxide and titanium oxide cost less than $1 per kilogram whereas cobalt costs about $45 per kilogram and nickel about $18,” said Ceder. “With DRX you have the potential to make very inexpensive energy storage. At that point lithium-ion becomes unbeatable and can be used everywhere – for vehicles, the grid – and we can truly make energy storage abundant and inexpensive.”

The researchers hope to expand their team and tackle some of the issues the new battery material currently has, such as improving the life cycle and optimizing the electrolyte.