Fraunhofer Institute for Silicon TechnologyAs the digitization of our society continues to increase and AI processes spread rapidly, the energy demand for information-processing components is escalating. To counter this, new approaches to information processing are needed. One approach is neuromorphic computing, inspired by the brain, in which the boundaries between memory and computational unit are dissolved. Since the constant transport of information is very energy-intensive, so-called in-memory compute architectures can save several orders of magnitude in energy. These architectures require memory cells that are capable of performing basic computing operations.
Energy-saving information storage and clever switches
Ferroelectric switching of AlScN for voltages up to 10V with complete repolarization starting at about 9 V.
Output characteristics of AlScN-based FeFETs with fully polarized states. Intermediate states lie between the curves.
Erste AlGaN/GaN HEMT (engl.: high-electron-mobility transistor) Prototypen mit integriertem ferroelektrischem Gate.
Ferroelectric memories at Fraunhofer ISIT - optimistic progress in development
At Fraunhofer ISIT, the ferroelectric material aluminum-scandium-nitride (AlScN) is used in the construction of such memory cells and incorporated into ferroelectric field-effect transistors as well as ferroelectric RAM cells. In contrast to the numerous alternative approaches, the scientists at ISIT were able to show that even a single AlScN-based memory cell is capable of storing and processing multiple bits of information. This multi-bit capability potentially allows to overcome previous challenges in the scalability of ferroelectric memories. However, it is important to demonstrate that the new material system can also compete with competing storage media in other target parameters. In particular, the writing voltage and speed are in the focus of future users. The mood among ISIT researchers was correspondingly optimistic when they were able to demonstrate switching frequencies in the MHz range and at less than 10 V for the first time last year. In the previous year, these sizes were only in the kHz range and at voltages 10 times higher.
In addition to determining the limits of AlScN, the scientists are evaluating new integration approaches in CMOS circuits and continuously improving the understanding of the material system.
Synergies with nitride-based semiconductors
In 2022, ferroelectric AlScN was incorporated into gallium nitride (GaN)-based transistors for the first time. They are central building blocks of tomorrow's power electronics because they are particularly energy-efficient. However, GaN-based circuits still lack reliable and scalable memories, as there are no established CMOS equivalents for them to date. AlScN, the only nitride-based ferroelectric to date, is particularly suitable for this purpose. For this reason, its development was included in the project in parallel to the previously discussed classic CMOS technologies.
Currently, the development towards first circuits of such memory elements is very promising and it is approaching the stage where AlScN can be applied in decentralized data processing for the first time. In addition, concepts are being evaluated at ISIT as to how the properties of power transistors can be modified by means of ferroelectric layers. For example, the switch-on point of these components plays a decisive role for reliable and energy-efficient use. However, as demonstrated in the case of memory cells, this point can be modified almost at will. This has resulted in the first concepts of intelligent switches that could replace entire circuits that have so far mimicked this function.