U.S. and Italian researchers report in the journal Nature Electronics on Monday that they have developed a new magnetic memory device based on antiferromagnetic material, which is small in size and uses very little energy, and is likely to help solve the “memory bottleneck” of the current development of artificial intelligence (AI).
Study Drawings – 1: Structural Indications (from: Nature Electronics)
The rapid development of AI technology is expected to improve healthcare, transportation and other areas, but its enormous potential to play on the basis of sufficient computing power, as a number of AI data sets, computers need to be more powerful memory support. Ideally, AI-enabled storage devices must not only have the same speed as static random memory (SRAM), but also have storage capacity similar to dynamic random memory (DRAM) or flash memory, and, more importantly, they consume less energy. However, there are currently no storage technologies to meet all these requirements, which has led to so-called “memory bottlenecks” that severely limit the performance and application of current AI.
Figure 2: Conversion measurement of current control.
To do this, researchers at Northwestern University and researchers at the University of Messina in Italy have targeted antiferromagnetic materials. Antiferromagnetic materials rely on magnetic and orderly spin to complete the data storage, the data stored can not be erased by external magnetic fields. Because of its fast safety and low energy consumption, it is regarded as the potential material of storage device, and how to control the internal magnetic sequence of the material has become a research difficulty.
Figure 3: Micromagnetic simulation.
In the new study, the team used a column-shaped antiferromagnetic material, a geometry that scientists had never explored before. Studies have shown that antiferromagnetic platinum manganese (PtMn) columns that grow on heavy metal layers with diameters as low as 800 nanometers can be converted irreversibly between different magnetic states through very low currents. Multi-stage storage can be achieved by changing the amplitude of the write current.
Figure 4: Asymmetric number of write pulse experiments.
The researchers point out that memory devices based on antiferromagnetic platinum manganese columns are only one-10th of the existing anti-ferromagnetic material-based storage devices, and more importantly, the new devices are manufactured in a way that is compatible with existing semiconductor manufacturing specifications, meaning storage equipment manufacturers can easily adopt new technologies without having to buy new equipment.
Figure 5: The result of the conversion of different AFM film thicknesses to metal materials.
The researchers point out that the new magnetic memory devices are small and low energy consumption, which is expected to make anti-ferromagnetic memory practical applications and help solve AI’s “memory bottlenecks”. They are now trying to further reduce the size of their devices and improve the way data is written to use energy so that new technologies can be put into practice as quickly as possible.
Figure 6: Measurements on nanoscale devices.
Editor-in-chief circle point
Memory has always been a bottleneck for computer power, because memory requires fast reading and writing speed and stability. In recent decades, we have been using semiconductors to make memory, and magnets have been used for hard drives that are not very fast to read. If “magnetic memory” is created, it will greatly expand the computer’s “brain capacity” and “intelligence”. All this to the progress of materials science as a prerequisite, the industry advanced can not be separated from the basic scientific research input.
Science daily Washington, February 10th, by wire (reporter notch Ying)