[Gettyimages Bank]
For in-memory computing, researchers have developed a magnetoresistive random access memory (MRAM) array chip that can also serve as a platform to download biological neuronal networks, Samsung said, adding the study showcased efforts to merge memory and system semiconductors for next-generation AI chips.
In the standard computer architecture, data is stored in memory chips with data computing executed in separate processor chips. In-memory computing can realize next-generation low-power AI semiconductor chips because data storage and computing are possible in a memory network, leading to a substantial reduction in power consumption.
Non-volatile memories have been actively used for demonstrating in-memory computing. But it has been difficult to use MRAM for in-memory computing. Despite its merits such as operation speed, endurance and large-scale production, MRAM showed low resistance because it cannot enjoy the power reduction advantage when used in the standard in-memory computing architecture.
Samsung's MRAM array chip demonstrated in-memory computing with a new "resistance sum" architecture that solved the problem of low resistance. In a performance test on AI computing, the chip achieved an accuracy of 98 percent in the classification of hand-written digits and 93 percent in detecting faces from scenes.
Samsung said that the study expanded the frontier of next-generation low-power AI chip technologies. "In-memory computing draws similarity to the brain in the sense that in the brain, computing also occurs within the network of biological memories, or synapses, the points where neurons touch one another," Jung Seung-chul, a staff researcher at Samsung Advanced Institute of Technology (SAIT), said in a statement on January 13.
"While the computing performed by our MRAM network, for now, has a different purpose from the computing performed by the brain, such solid-state memory network may in the future be used as a platform to mimic the brain by modeling the brain’s synapse connectivity," said Jung. The research paper was published online by Nature, a scientific journal based in London.
Copyright ⓒ Aju Press All rights reserved.
