Ferroelectric and Antiferromagnetic Memory: A Solution to AI Data Storage Challenges
Effectively storing data remains a key issue in modern information systems and a challenge for electronics engineers who need to design devices with the thought of robust data storage as a primary concern.
In the data torrent of modern information systems, AI algorithms play a major role in improving data storage and processing accuracy, for example, in automotive and IoT systems.
However, the struggle to solve the memory bottleneck problem of AI applications is still current. A possible solution to designing robust data storage and processing devices comes from innovations in ferroics, using ferromagnetic, ferroelectric, and antiferromagnetic materials to build efficient memory and logic devices.
Atypical Memory Qualities of Ferroelectric Materials
Ferroelectric materials are usually crystalline compounds that have particles of electric dipoles with a separate positive and negative charge, which, when exposed to a strong electric field, line up and produce the same polarizing effect in the material itself.
Once the electric field is removed, ferroelectric materials remain polarized because they retain the memory. Devices with ferroelectric properties have the advantage of possessing several important memory characteristics: non-volatility, low power consumption, high endurance, and high-speed writing.
When a new electric field is introduced, ferroelectric compounds lag behind in changing the direction of the polarization, a phenomenon called hysteresis, which went largely unexplained until very recently.
A diagram of the ferroelectric phase of amorphous hafnium oxide. Image used courtesy of Ferroelectric Memory
Ferroelectricity and the Puzzling Hysteron Particles
Back in 1935, when Franz (Ferenc) Preisach first came up with ferroelectric material properties, he named the randomly polarizing fields hysterons, while the belated polarization effect in the crystalline stacks was appropriately named hysteresis.
Hysteresis was wrapped in quite a mystery for more than 80 years. For long scientists couldn’t explain why hysterons lack uniformity and immediacy until the 2018 breakthrough made by the researchers from the universities in Linköping and Eindhoven—who demonstrated the existence of hidden particles hysterons in two ferroelectric materials: semi-crystalline copolymer P(VDF-TrFE) and the polycrystalline molecular ferroelectric trialkylbenzene-1,3,5-tricarboxamide (BTA).
Researchers used the findings to improve the predictive curve of hysteresis on a nanoscale level, looking for better insights about how ferroelectric materials behave when exposed to changing electric fields.
Ferroelectric memory or, more precisely, ferroelectric RAM (FRAM) is a worthwhile contender for memory embedded applications. Instead of using magnetic hard drives that require a large current to store data, ferroelectric memory devices use less power. As a general rule, ferroelectricity has a scaling problem, making the dipoles critically unstable when used in tiny surfaces measured in nanometres.
Ferroelectric Nanoscale Devices
However, it seems that the full scope of properties of ferroelectric materials is not well known, such as in this example of hafnium-based coatings.
Contrary to the typical behavior of ferroelectrics, memory and logic devices with hafnium-based thing films from this research became more robust when their size was reduced.
The propensity for stability when a tremendous amount of pressure that was applied on epitaxially strained Hf0.5Zr0.5O2 thin films is yet to be explained, but it could be used to add another layer of ferroelectric memory to improve the data storage capacity of magnetic drives.
For the time being, though, we are far from seeing ferroelectric drives in widespread use.
Examples of potential applications of spintronics. Image used courtesy of K. Inomata and the Research Center for Magnetic and Spintronic Materials
Another solution which holds more promise for designing powerful chips for data-hungry applications is a memory device based on antiferromagnetic materials (AFM).
Unlike ferromagnetic materials, whose magnetic domains align in the same direction even when they are not exposed to external magnetic fields, (examples include metals such as iron, nickel, cobalt, and metal alloys), antiferromagnetic materials align in opposite directions.
A well-known example of an antiferromagnetic material is manganese oxide (MnO). In a recent discovery, researchers from the McCormick Northwestern University and the University of Messina in Italy developed the smallest of its kind; an AFM device made from antiferromagnetic platinum manganese pillars with a diameter of only 800 nm.
Since it is compatible with current semiconductor manufacturing processes, the practical device can be used without significant investments in new equipment.
AFM Memory Devices
AFM memory devices are a new stage in the development of MRAM (magnetic random-access memory), a technology with blanket data storage qualities, that includes both the capacity to store large data volumes and to do it fast.
AFM memory is, in a way, both static and dynamic RAM, necessary for the development of today’s AI applications that feed on computing power and need to possess non-volatile memory capabilities.
AFM memory devices don’t need a constant electric current as a power supply but could be instead powered by an electric voltage, a future task on the agenda of the same research team.
Furthermore, the tightly packed nanoscale devices cannot interact with external magnetic fields, in turn securing data storage because the data can't be easily erased.
An Emerging Technology
Due to their relatively unstable and unpredictable properties, all ferroic materials, including ferroelectrics and antiferromagnets, require extensive research in the field of nanoelectronics if we want to integrate them in chip-sized devices with giga-scale complexity.
Spintronics holds some promise in this area, too, as an emerging technology that explores the usage of electron spins instead of electron charge for information storage.