Scientists have developed a new mathematical model that could make it easier to design computer chips using a promising class of materials called ferroelectrics.
The advance could eventually contribute to faster and more energy-efficient computing, including new systems designed for artificial intelligence.
The model, developed by researchers led by the University of Michigan, predicts how ferroelectric materials behave when exposed to the complicated electrical signals found in real electronic devices.
The study was published in Advanced Materials.
Ferroelectric materials are attracting attention because they can store information in a different way from many conventional electronic devices.
When an electric field is applied, the internal electrical polarization of the material can switch between two directions.
Those two stable states can represent the 1s and 0s used to store and process digital information. Unlike conventional approaches that may depend on electrical charge that can leak away, ferroelectric states can remain in place, potentially reducing energy use.
This could be especially valuable for technologies that combine computing and memory. Such systems could reduce the need to constantly move data between processors and memory, a major source of energy consumption in modern computing and AI.
However, accurately predicting the behavior of ferroelectric materials has been difficult. Many existing models assume that the applied voltage remains constant. In real computer chips, voltage signals can arrive as pulses, change frequency and vary over time.
The new model is designed to handle these more realistic conditions.
The researchers tested it using hafnium zirconium oxide, or HZO, a ferroelectric material considered particularly promising for electronics. HZO is compatible with existing silicon chip manufacturing and can retain its useful properties even when made extremely thin.
When voltage is applied to a ferroelectric material, its polarization does not switch everywhere at once. Instead, switching begins in tiny areas, a process known as nucleation. These switched regions then expand, causing surrounding areas to change direction.
The speed of this process depends on the electric field. To predict it, the researchers combined an existing mathematical approach called the reverse time cone model with statistical thermodynamics.
Importantly, the model keeps track of the material’s entire voltage history. This allows it to account for a kind of memory effect in which previous electrical signals influence how the material responds to later ones.
The researchers then simplified the model to just three main parameters so that it could be incorporated into SPICE, widely used software for designing and testing electronic circuits.
Experiments with HZO devices showed that the model closely matched their actual behavior using a single set of material parameters.
The advance could allow engineers to virtually test huge numbers of ferroelectric memory cells before manufacturing a chip. They could predict factors such as energy consumption, switching speed and responses to different voltage patterns.
This may be particularly useful for neuromorphic computing, which aims to create electronic systems inspired by the way the brain processes information.
By connecting fundamental material physics with practical chip-design tools, the new model could help turn promising ferroelectric materials into useful technologies for future memory, computing and AI systems.
