SEOUL, Jan. 14 (UPI) -- Samsung Electronics announced it has developed an in-memory computing technology that merges memory and system semiconductors.
The world's largest memory chipmaker said Thursday the new technology is enabled by non-volatile memories, dubbed "magnetoresistive random access memory," for the first time globally.
In the traditional computer architecture, data is stored in memory chips while separate processor chips compute data. In comparison, in-memory computing is able to perform both data storage and computing in a memory network at the same time.
As a result, Samsung said that in-memory computing substantially reduces power consumption as the data do not need to move, a feature that may be suitable for next-generation artificial intelligence chips.
Many have tried to achieve in-memory computing based on MRAM because of its merits, including fast operation speed, endurance and large-scale production, but the technology has been elusive, Samsung said.
The company said its MRAM in-memory computing chip achieved an accuracy of 98% in the classification of handwritten digits and 93% in detecting faces from scenes.
The developments will be published in an upcoming edition of the journal Nature. Samsung researcher Jung Seung-chul is the first author of the paper.
"In-memory computing is similar to the brain in the sense that within the brain, computing also occurs within the network of biological memories, or synapses, the points where neurons touch one another," Jung said in a statement.
Experts say in-memory computing is one of the futuristic concepts for the semiconductor industry.
"Samsung appears to have made a technological breakthrough based on potential-rich MRAM," Professor Han Tae-hee at Sungkyunkwan University told UPI News Korea.
"To fully adopt in-memory computing, however, we need a completely different software ecosystem, including different operating systems. Hence, it remains to be seen whether the new Samsung technology will be commercially viable," he said.