Traditional chips face challenges, but until quantum computing and brain-like computing are well developed, chip computing will continue to rely on the advancement and innovation of existing technologies. In-store computing chips have also received a lot of attention in the AI era, but Purdue University engineers developed methods from the perspective of materials innovation, the realization of the chip in the calculation can also be stored at the same time. In the future, the researchers say, further improvements to the chip could help the development of brain computing.
Computer chips use two different components to process and store information. If engineers can combine or place two components together or placed next to each other, there will be more space on the chip, faster from the chip, and more powerful performance.
Engineers at Purdue University have developed a way to process information on millions of microswitches, often called transistors, that can also store information on a chip.
This approach, detailed in a paper published in Nature Electronics, does so by solving another problem: combining transistors with higher-performance storage technologies than most computers use, called ferroelectric RAM.
Researchers have been trying to bring the two together for decades, but the problem is the interface between ferroelectric materialand and silicon, the semiconductor material that makes up transistors. In addition, ferroelectric RAM operates as a separate unit on the film, which limits its potential to significantly improve computational efficiency.
The team, led by Purdue University’s professor of electrical and computer engineering, Peide Ye, Richard J. and Mary Jo Schwartz, discovered ways to overcome the deadly hostile relationship between silicon and ferroelectric materials.
“We used semiconductors with ferroelectric properties. Both materials become one material so that there is no need to worry about interface problems. Ye said.
The result is a so-called ferroelectric semiconductor field-effect transistor, which is constructed in the same way as the transistors used on current computer chips.
The alpha selenium tantalum material not only has ferroelectric properties, but also solves the problem that “band width” usually acts as an insulator rather than a conventional ferroelectric material for semiconductors, which means that current cannot pass through and does not occur without calculation.
The band width of alpha-selenium is much smaller, which makes the material a semiconductor without losing its ferroelectric properties.
Mengwei Si, a postdoctoral researcher in electrical and computer engineering at Purdue University, built and tested the transistor and found that its performance was comparable to that of existing iron-field-effect transistors, and said it would be better to optimize performance in one step. Sumeet Gupta, an assistant professor of electrical and computer engineering at Purdue University, and Atanu Saha, who received his Ph.D., supported modeling.
Si and Ye’s team also worked with researchers at the Georgia Institute of Technology to build alpha-selenium in a chip space called an iron tunnel junction that engineers could use to enhance the chip’s functionality. The team presented the study at the 2019 IEEE International Electronics Conference on December 9.
In the past, researchers were unable to build high-performance iron-electric tunnel junctions because its broadband gap made the material too thick to pass through current. Because the band gap of alpha-selenium is much smaller, the thickness of the material is only 10 nanometers, allowing more current to flow through.
Larger currents reduce the size of the chip to a few nanometers, making the chip’s transistors denser and more energy-efficient. The thinner material – which can even reduce the thickness of the atom – also means that the electrodes on both sides of the tunnel junction can be much smaller, which is useful for building circuits that simulate the human brain, Ye adds.