Samsung Electronics Co., the world’s top memory chip maker, succeeded in developing in-memory computing technology that allows running a computer without separate processors.
In the standard computer architecture, data is stored in memory chips and data computing is executed in separate processor chips.
Samsung has been working to merge memory and system semiconductors for next-generation artificial intelligence (AI) chips through in-memory computing, a new computing paradigm that seeks to perform both data storage and data computing in a memory network.
The South Korean tech giant on Thursday announced its demonstration of the world’s first in-memory computing based on magnetoresistive random access memory (MRAM). The paper on this innovation, titled ‘A crossbar array of magnetoresistive memory devices for in-memory computing’, was published by Nature, the world's leading multidisciplinary science journal.
Samsung Advanced Institute of Technology (SAIT) led the development in collaboration with the company’s foundry unit as well as the semiconductor research and development center. The first author of the paper was Jung Seung-chul from SAIT. Co-corresponding authors were Ham Don-hee, a research fellow at SAIT and a professor at Harvard University, and Kim Sang-joon, SAIT’s vice president.
SAIT, which develops Samsung’s future technologies, has been paying attention to problems of the existing computer architecture with separate functions of central processing units and memory chips since data to be processed soared due to the improving AI technology. The performance of CPUs and memory chips are advancing, but the time for data exchange is not decreasing.
The institute tried to find solutions from in-memory computing with designs to group chips of each function into memory chips and modules that process operations to prevent “data bottlenecks.” It has been working on the technology with various chips.
The global industry has been pursuing in-memory computing, especially based on resistive random access memory (RRAM) and phase-change random access memory (PRAM). It has so far been difficult to use MRAM because of its low resistance that increases power consumption although the non-volatile memory has merits such as operation speed, endurance and large-scale production. Samsung’s researchers replaced the standard, ‘current-sum’ in-memory computing architecture with a new, ‘resistance sum’ in-memory computing architecture to address the problem of small resistances of individual MRAM devices.
The performance of the MRAM computer was better than expected. Samsung’s research team tested the performance of this MRAM in-memory computing chip by running it to perform AI computing. The chip achieved an accuracy of 98% in the classification of hand-written digits and a 93% accuracy in detecting faces from scenes.
“By ushering MRAM - the memory which has already reached commercial-scale production embedded in the system semiconductor fabrication - into the realm of in-memory computing, this work expands the frontier of the next-generation low-power AI chip technologies,” Samsung said in a statement.
FOR BIOLOGICAL NEURONAL NETWORKS
Samsung’s researchers suggested the MRAM chip can serve as a platform to download biological neuronal networks.
“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,” said SAIT’s Jung, the first author of the paper.
“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.”
In September 2021, researchers of Samsung and Harvard University published a paper on the neuromorphic electronics vision on Nature Electronics, a journal covering electronics research.