When we watch videos or ask AI questions, enormous amounts of data are constantly moving inside computers. In particular, data centers that support AI must process and transfer vast amounts of data at very high speeds. However, current computers have a fundamental limitation: the place where calculations are performed and the place where data is stored are physically separated.
Because of this, data has to travel back and forth many times within a chip. This repeated movement takes time and consumes energy, creating a bottleneck that limits both speed and efficiency.
One promising way to solve this problem is to place memory very close to the computing circuits. This reduces the distance that data needs to travel. Such an approach, where computing and memory are integrated, is known as logic-embedded memory and is being actively studied worldwide.
A material attracting attention for this purpose is aluminum scandium nitride. This material can retain information even when the power is turned off and is also resistant to high temperatures during device fabrication.
Previous studies have mainly focused on making thin layers of this material and investigating how thin they can be. However, real memory devices consist of multiple stacked layers, including electrodes. As a result, it has remained unclear how thin the entire device can be made.
To address this challenge, a research team led by Professor Hiroshi Funakubo at Institute of Science Tokyo (Science Tokyo) set out to minimize the thickness of the entire memory device. The paper is published in the journal Advanced Electronic Materials.
To read more, click here.