Navigation and service

Improved PCRAM for More Efficient Computers

17 September 2019

Artificial intelligence, big data and supercomputing are set to revolutionize how we live and work. However, the continued development of these technologies could experience a significant slowdown. The "culprit" here is the architecture of current computer chips: built according to the principles of "von Neumann architecture", they process and store data in areas which are physically separated from each other. This has negative consequences: the necessary communication between the two units limits the computing speed attainable and also causes energy wastage of around 40 percent.

In contrast, our brain works extremely efficiently and therefore a chip architecture modelled on neuronal information processing could provide a way out of the dilemma. An international team of researchers, including scientists from Forschungszentrum Jülich and RWTH Aachen University, is now proposing an improvement in random access memory (RAM) based on phase change materials, which can be used in industrial production. These PCRAMs are regarded as promising candidates for the realization of a chip architecture based on a neuro-inspired model.

PCRAMs store binary data in a phase change layer in the form of two different atomic arrangements: disordered and crystalline. An electrical current is used to switch between the two states and their different resistances enable data to be read out. However, the performance of the layers is not yet good enough for use in neuromorphic computing: when used intensively, the composition and structure of the layers change, and as a result their resistances vary considerably. The work of the scientists is now bringing about significant improvements.

This is down to two changes: the use of an alternative phase change material and its spatial application in nano-layers, which are stacked together with a different material, thereby increasing performance stability. "Our approach is suitable for industrial production because multi-layer deposition does not significantly increase manufacturing costs or require complex processes. This facilitates integration into state-of-the-art device configurations for high-performance neuromorphic chips," explains Prof. Chunlin Jia from the Ernst Ruska-Centre in Jülich and the Jiaotong University in Xi'an, China, who was involved in evaluating the test results.

Original publication:

Phase-change heterostructure enables ultralow noise and drift for memory operation;
Keyuan Ding, Jiangjing Wang, Yuxing Zhou, He Tian, Lu Lu, Riccardo Mazzarello,
Chunlin Jia, Wei Zhang, Feng Rao, Evan Ma
Science  22 Aug 2019: eaay0291, DOI: 10.1126/science.aay0291

Further information:

Article: Paper accepted by Science: "Phase-change heterostructure enables ultralow noise and drift for memory operation” from 23 August 2019 der Jiaotong-Universität Xi’an, China (engl.)

Forschungszentrum Jülich, Ernst Ruska-Centre – Physics of Nanoscale Systems (ER-C-1) / Peter Grünberg Institute – Microstructure Research (PGI-5)

RWTH Aachen, Work Group of Prof. Riccardo Mazzarello