New computer simulations help scientists advance energy-efficient microelectronics

Thanks to advances in microchips, today’s smartphones are so powerful that they would have been considered supercomputers in the early 1990s. But the growing ubiquity of artificial intelligence and the Internet of Things—the vast network of connected devices that has made everything from smart grids to smart homes possible—will require a new generation of microchips that not only surpass previous records for miniaturization and performance, but also are more energy efficient than current technologies.

As part of this effort, Berkeley Lab scientists are working to revolutionize the transistor, one of the fundamental components of computer microchips, for higher performance and energy efficiency. Recent work has shown the potential of new transistor materials that use an unusual property called negative capacitance to enable more efficient memory and logic devices. When a material has negative capacitance, it can store more electrical charge at lower voltages, which is the opposite of what happens in conventional capacitive materials.

A multidisciplinary team of researchers has now developed an atomistic understanding of the origins of negative capacitance, allowing them to improve and tailor this phenomenon for specific device applications. This breakthrough was made possible by FerroX, an open-source 3D simulation framework that the team custom-built for the study of negative capacitance. Their work has been published in the journal Advanced electronic materials.

The work represents an important step in a multi-year project, “Co-Design of Ultra-Low-Voltage Beyond CMOS Microelectronics,” which aims to design new microchips that could perform better and require less power than conventional silicon chips.

While it is not uncommon for materials development to be closely tied to applications, Berkeley Lab’s co-design approach to microelectronics research, where atomistic understanding of materials properties is guided and informed by specific device requirements, tightens the connection between research goals in all aspects of device development and builds on the interdisciplinary team science for which the lab is known in hopes of accelerating the path from R&D to commercialization.

“There’s a lot of trial and error in making new materials. It’s like creating a new recipe. Researchers usually have to work day and night in the lab to tweak that recipe. But with our modeling tool, FerroX, you can use your own computer to target specific parameters that can affect the performance of the negative capacitance effect,” said Zhi (Jackie) Yao, a research scientist in Berkeley Lab’s Applied Mathematics and Research Computational Sciences Division and lead author of the study.

Yao and first author Prabhat Kumar, a postdoctoral researcher in the Division of Applied Mathematics and Computational Research, co-led the development of FerroX for the Microelectronics Co-Design Project.

Discovering the atomic origins of negative capacitance

In 2008, co-author Sayeef Salahuddin, a professor of electrical engineering and computer science at UC Berkeley and a senior research scientist in Berkeley Lab’s Materials Science Division, first proposed the concept of negative capacitance to demonstrate a new approach to designing energy-efficient computers.

Negative capacitance typically occurs in materials with ferroelectric properties. Ferroelectric materials hold promise as energy-efficient computer memories because their built-in electrical polarization can be used to store data, for example, that can be written and erased using a low-power electric field.

In the years since Salahuddin’s pioneering proposal, researchers have learned that the negative capacitance effect in thin films of ferroelectric hafnium oxide and zirconium oxide (HfO₂-ZrO₂) occurs when films are composed of a mixture of phases.

This means that small regions or “grains” in the film have slightly different arrangements of atoms or “phases.” The size of these phase grains is tiny (only a few nanometers in diameter), but the different phases have distinct electronic properties that can interact with each other and give rise to macroscopic phenomena such as negative capacitance.

The Salahuddin group has already exploited this phenomenon to produce record-breaking microcapacitors, but to fully exploit the potential of negative capacitance, the researchers needed a deeper understanding of its atomic origins.

To do this, a multidisciplinary team co-led by Yao and Kumar developed FerroX. The open-source framework allowed them to develop 3D phase-field simulations of a ferroelectric thin film, in which they could vary the phase composition at will and study the impacts on the film’s electronic properties.

“Our goal was to understand the origin of the negative capacitance in these films, which is not well understood,” Kumar said. “Our simulations are the first to help researchers tailor a material’s properties for new enhancements in negative capacitance observed in the lab.”

As a result, Berkeley Lab researchers found that the negative capacitance effect can be enhanced by optimizing the domain structure, reducing the size of the ferroelectric grains and arranging them to have a particular direction of ferroelectric polarization.

“This approach to improving negative capacitance was unknown before our study because previous models lacked scalability to easily explore the design space and lacked physical customization,” Yao said.

Yao attributes this new modeling capability to firsthand work with materials scientists like Salahuddin, who helped the FerroX development team understand how to shape their models around the physics of ferroelectrics, and to Berkeley Lab’s unique multidisciplinary assets, where researchers from across the scientific spectrum are in close proximity to the Perlmutter supercomputer at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC).

Perlmutter supports complex simulations, data analysis, and artificial intelligence experiments that require multiple graphics processing units (GPUs) at once. Yao, Kumar, and their team relied heavily on Perlmutter to develop FerroX, which is now available to other researchers as an open-source framework that’s portable from laptops to supercomputers.

“It’s exciting that FerroX can help such a broad community of researchers across academia, industry and national laboratories,” Yao said.

While the FerroX models in the current study simulate the origin of negative capacitance as it evolves at the transistor gate, the Berkeley Lab team plans to use the open-source framework to simulate the entire transistor in future studies.

“Over the years, we have made significant progress in the physics of negative capacitance and in integrating this physics into real microelectronic devices,” said Salahuddin. “With FerroX, we can now model these devices starting with atoms, which will allow us to design microelectronic devices with optimal negative capacitance performance. This would not have been possible without the strength of this co-design group of researchers in computer science and materials science.”