Improved Computers: Materials with Memory Can Enhance Processor Functionality
Artificial intelligence, machine learning, and edge computing have introduced new computational paradigms aimed at improving the quality of life. These advancements require enhanced performance from computer systems, particularly in terms of data storage and movement within microprocessors. Current methods of storing and moving data have intense energy requirements that are projected to account for over 20% of the global energy consumption over the next ~10 years if no improvements are made. [1,2] Additionally, memory components have less efficiency than data processing components, and there are indications that this "memory–logic" gap will widen in the future.
Current microprocessors rely on static random access memory (SRAM) and dynamic random access memory (DRAM) for data storage. However, both SRAM and DRAM have limitations in terms of retaining data without a continuous power supply. Typical types of SRAM and DRAM are volatile, requiring constant refreshing to avoid data loss, analogous to the human body's need for constant hydration. There is a critical need to improve memory functionality in microprocessors, and a potential solution could be ferroelectric materials.
The Center for Three-Dimensional Ferroelectrics for Microelectronics (3DFeM) is focusing on utilizing ferroelectric materials to address the memory-logic gap, or “bottleneck” as it is commonly referred to. Remarkably, these ferroelectric materials are less than 100 nm thick, which is approximately 1000 times thinner than a human hair. Ferroelectrics are special because of their ability to retain electronic charge from a voltage supply even after it is removed, an intrinsic property referred to as nonvolatility that is useful for memory storage. Specifically, ferroelectrics have a non-linear and hysteretic relationship between charge and applied voltage, meaning that even at zero voltage, there is still charge that can be electrically sensed. This charge dipole, a measure of charge separation in a material, is stable in two states, “up and down”, like north and south poles in a magnet since the orientation of the charge dipole can be reversed with an applied voltage. As a vector quantity, the charge dipole has a magnitude and an orientation. These two charge states can retain information at zero voltage, analogous to zeros and ones in digital logic. Unlike conventional semiconductors and dielectrics used in SRAM and DRAM, ferroelectrics offer nonvolatile behavior, eliminating the need for continuous data refreshing.
Researchers at 3DFeM are synthesizing new ferroelectric materials specifically designed for compatibility with microprocessor manufacturing conditions. These materials are being deposited at temperatures less than 400°C while still obtaining long range crystalline order using plasma-assisted techniques. The electrical performance of these materials is being tested to assess their ability to withstand over one million cycles of fast electrical pulses, simulating data processing in a computer.
Improving memory functionality in microprocessors is crucial for achieving global prosperity, given the increasing demand for advanced computational technologies. The utilization of ferroelectric materials shows promise in bridging the "memory–logic" gap and reducing the energy consumption associated with data storage and movement. Researchers at 3DFeM are currently in the process of demonstrating the viability of ferroelectric materials in microprocessor environments, potentially revolutionizing the way data is processed and stored in future computer systems.
 A.S.G Andrae, Int. J. of Sci. and Eng. Invest. 8, 27, (2019).
 A.S.G. Andrae et al. Challenges 6, 117 (2015). doi:10.3390/challe6010117