Power/Performance Bits: June 18

Multi-value logic transistor
Researchers at the University of Texas at Dallas, Hanyang University, Gwangju Institute of Science and Technology, Yonsei University, Kookmin University, and Ulsan National Institute of Science and Technology developed and fabricated a transistor capable of storing intermediate values between 0 and 1. Such a multi-value logic transistor would allow more operations and a larger amount of information to be processed in a single device.

Based on zinc oxide, the device is capable of two electronically stable and reliable intermediate states between 0 and 1, boosting the number of logic values per transistor from two to three or four.

“The concept of multi-value logic transistors is not new, and there have been many attempts to make such devices,” said Kyeongjae Cho, professor of materials science and engineering at UT Dallas. “We have done it.”

The device uses two forms of zinc oxide. Crystalline zinc oxide quantum dots are embedded into amorphous zinc oxide to form a composite nanolayer, which is then incorporated with layers of other materials in a superlattice.

The image on the left shows two forms of zinc oxide combined to form a composite nanolayer in a new type of transistor: Zinc oxide crystals (inside the red circles) are embedded in amorphous zinc oxide. The image on the right is a computer model of the structure that shows electron density distribution. (Source: University of Texas at Dallas)

“A device incorporating multi-level logic would be faster than a conventional computer because it would operate with more than just binary logic units. With quantum units, you have continuous values,” Cho said. “The transistor is a very mature technology, and quantum computers are nowhere close to being commercialized. There is a huge gap. So how do we move from one to the other? We need some kind of evolutionary pathway, a bridging technology between binary and infinite degrees of freedom. Our work is still based on existing device technology, so it is not as revolutionary as quantum computing, but it is evolving toward that direction.”

“By engineering this material, we found that we could create a new electronic structure that enabled this multi-level logic behavior,” added Cho. “Zinc oxide is a well-known material that tends to form both crystalline solids and amorphous solids, so it was an obvious choice to start with, but it may not be the best material. Our next step will look at how universal this behavior is among other materials as we try to optimize the technology.

The researchers have applied for a patent on the technology.

Graphene band gap
Researchers at Purdue University, University of Michigan, and Huazhong University of Science and Technology stressed graphene to create a record band gap for the material, potentially opening the way for graphene electronics.

The tough and highly conductive material doesn’t naturally have a band gap. To create one, the team turned to a technique called laser shock imprinting, which allowed them to create a record 2.1 electronvolt band gap. Previous studies had reached around 0.5 eV.

“This is the first time that an effort has achieved such high band gaps without affecting graphene itself, such as through chemical doping. We have purely strained the material,” said Gary Cheng, professor of industrial engineering at Purdue, whose lab has investigated various ways to make graphene more useful for commercial applications.

“Researchers in the past opened the band gap by simply stretching graphene, but stretching alone doesn’t widen the band gap very much. You need to permanently change the shape of graphene to keep the band gap open,” Cheng added.

Beyond keeping the band gap open, they found a way to tune the gap width from zero to 2.1 eV for different applications.

The laser shock imprinting technique uses a laser to create shockwave impulses that penetrated an underlying sheet of graphene. The laser shock strains graphene onto a trench-like mold, permanently shaping it. Adjusting the laser power adjusts the band gap.

Cheng notes that while it is still far from putting graphene into semiconducting devices, the technique grants more flexibility in taking advantage of the material’s optical, magnetic and thermal properties.

Finding faked images
Researchers at NYU Tandon propose a way to protect digital images and videos from misleading manipulation with an authenticating watermark.

AI techniques have made it increasingly possible to create convincing faked images and videos, and these so-called ‘deep fakes’ of politicians and celebrities have been created as demonstrations of the power of the technology. But they could also be used by bad actors intentionally trying to pass a faked event off as real, and the techniques are becoming both better and more accessible.

The researchers’ solution starts at the moment a digital image is created, replacing the typical photo development pipeline with a neural network that introduces carefully crafted artifacts directly into the image at the moment of image acquisition. These artifacts, akin to “digital watermarks,” are extremely sensitive to manipulation.

“Unlike previously used watermarking techniques, these AI-learned artifacts can reveal not only the existence of photo manipulations, but also their character,” said Paweł Korus, a research assistant professor in the Department of Computer Science and Engineering at NYU Tandon.

The process is optimized for in-camera embedding and can survive image distortion applied by online photo sharing services, according to the team.

An illustration demonstrating the general principle of the authentication process for a binary manipulation detection problem. (Source: NYU Tandon)

“If the camera itself produces an image that is more sensitive to tampering, any adjustments will be detected with high probability,” said Nasir Memon, a professor of computer science and engineering at NYU Tandon. “These watermarks can survive post-processing; however, they’re quite fragile when it comes to modification: If you alter the image, the watermark breaks.”

In tests, this prototype imaging pipeline increased the chances of detecting manipulation from approximately 45% to over 90% without sacrificing image quality.

The whole imaging and distribution channel is modeled as a fully differentiable Tensorflow model. The camera is replaced with a convolutional neural network and optimized for faithful development of color RGB images from RAW sensor measurements, and reliable manipulation detection at the end of the distribution channel.

The researchers say additional work is needed to refine the system, which is open source and can be freely accessed from GitHub.