With technological innovations such as the Internet of Things (IoT), all manner of electronic devices are being connected by the Internet, and the amount of data processed by IT devices is increasing steadily. The use of artificial intelligence (AI) is particularly important in these technologies, but with the accelerating increase in the amount of data handled, the energy consumption of electronic devices is becoming an issue. Brain-type computing is recently attracting attention as an approach aimed at enhancing the functionality and reducing the power consumption of AI technology. Brain-type computing is an attempt to mimic the activity of neurons and synapses in the brain with electronic elements having similar functions, but synapses store the importance of information as “weight,” which requires a memory function to reproduce. However, existing SRAM and DRAM used as memory are volatile memories that consume power even in the standby state when data processing is not performed, so there is a concern that this will be an obstacle to reducing power consumption. The approach expected to solve this issue is the introduction of non-volatile memory that does not lose data even when the power is turned off, that is to say, memory that does not require standby power. Among these, MRAM that uses magnetic characteristics to impart non-volatility combines features such as high speed, high durability for repeated operation, and high affinity with existing semiconductor processes. As such, MRAM are expected to be a non-volatile memory with applicability not only to Neumann-type computing but also brain-type computing.

Researchers in AIST developed a magnetic tunnel junction element (hereafter, “MTJ element”) with a new structure using a tunnel barrier layer that combines lithium fluoride (LiF) and magnesium oxide (MgO) and successfully enhanced perpendicular magnetic anisotropy, which is an indicator of the memory retention characteristics of magnetic memory (MRAM). It was found that by introducing an extremely thin LiF layer only 1 or 2 atoms thick between iron (Fe) and MgO, the magnetization direction of the Fe can be stabilized in the direction perpendicular to the film surface, and the perpendicular magnetic anisotropy is increased to approximately twice that of the conventional structure using only MgO.

This MTJ element consists of a structure in which a tunnel barrier layer around 1 nm thick is sandwiched by a magnetic thin film and can store data semi-permanently according to the magnetization direction of the magnetic thin film. Utilization of this characteristic achieves non-volatile memory that does not require standby power, and studies are underway for application not only to existing Neumann-type computing, but also to brain-type computing that aims for advanced data processing by mimicking the structure and data processing method of the brain.