Sep 9, 2009
Resistive switching memory configured as synapse circuit
Neural computing is based on artificial networks, which have similar characteristics to biological neural systems, including fault tolerance and a capacity for adaptive learning. The adaptive function is accomplished by changing the synaptic weight values of neurons according to a learning algorithm.
By applying these properties to resistive random access memory (RRAM), a promising candidate for future nonvolatile memory devices, researchers in Korea hope to provide a way of compensating for common scalability issues such as high defect rates, high device variability and device ageing and facilitate economical mass production.
In a recent study, which was published in Nanotechnology, scientists from Gwangju Institute Science and Technology have fabricated a cross-point cell array using a resistive switching layer of GdOx and Cu doped MoOx.
The group operated its cell array as an electrically modifiable synapse array circuit and used the configuration to perform a weighted sum operation.
According to the team, the results pave the way for future ultrahigh density synapse circuits and large-scale neural network systems.
About the author
Hyejung Choi, is a PhD candidate in Hyunsang Hwang's group in the Department of Materials Science and Engineering at Gwangju Institute Science and Technology, Gwangju, South Korea. She is exploring resistive switching memory and its application in synapse circuits.