Korea, January 13, 2022 – Samsung Electronics, a world leader in advanced semiconductor technology, announces its demonstration of the world’s first in-memory computing based on MRAM (Magnetoresistive Random Access Memory). The article on this innovation was published online by Nature January 12 (GMT), and is published in the next print edition of Nature. TitleA cross-linked set of magnetic memory devices for in-memory computing‘, this article shows Samsung’s leadership in memory technology and its efforts to merge memory and system semiconductors for next-generation artificial intelligence (AI) chips.
The research was led by Samsung Advanced Institute of Technology (SAIT) in close collaboration with Samsung Electronics Foundry Business and Semiconductor R&D Center. The first author of the article, Dr. Seungchul Jung, Staff Researcher at SAIT, and the co-authors Dr. Donhee Ham, SAIT Fellow and Harvard University Professor and Dr. Sang Joon Kim, Vice President of Technology at SAIT, led the investigation.
In the standard computer architecture, data is stored in memory chips and data computing is implemented on separate processor boards.
In contrast, in-memory computing is a new computational paradigm that seeks to perform both data storage and data computing on a memory network. Because this scheme can process a large amount of data stored within the memory network itself without having to move the data, and also because the data processing in the memory network is carried out in a very parallel way, power consumption is considerably reduced. In-memory computing has thus emerged as one of the promising technologies for realizing next-generation low-power AI semiconductor chips.
Therefore, research on in-memory computing has been intensely treated worldwide. Non-volatile memories, particularly RRAM (Resistive Random Access Memory) and PRAM (Phase-change Random Access Memory), have been actively used to demonstrate in-memory computing. In contrast, until now it has been difficult to use MRAM ─ another type of non-volatile memory ─ for in-memory computing despite the merits of MRAM such as speed of operation, endurance and large-scale production. This difficulty stems from the low resistance of MRAM, due to which MRAM cannot enjoy the power reduction advantage when used in the standard in-memory computing architecture.
Samsung Electronics researchers have provided a solution to this problem through architectural innovation. Specifically, they succeeded in developing an MRAM array chip that displays in-memory computing, replacing the standard, “now-total” in-memory computing architecture with a new, “resilient sum” in-memory computing architecture that addresses the problem. of small resistances of individual MRAM devices.
Samsung’s research team later tested the performance of this MRAM in-memory computing chip by rolling it out to perform AI computing. The chip achieved 98% accuracy in handwriting digit classification and 93% accuracy in scene face detection.
Introducing MRAM ─ the memory that has already reached commercial-scale production embedded in the semiconductor manufacturing system ─ in the realm of in-memory computing, this work expands the frontier of next-generation low-power AI chip technologies.
The researchers also suggested that this new MRAM chip could not only be used for in-memory computing, but could also serve as a platform for downloading biological neural networks. This is in line with the neuromorphic electronic vision that Samsung researchers recently presented in a prospective article published in the September 2021 issue of the journal. Natural Electronics.
“Memory computation draws a resemblance to the brain in the sense that in the brain, computation also occurs within the network of biological memories, or synapses, the points where neurons touch each other,” said Dr. Seungchul Jung, the first author of the paper. “In fact, while the computing done by our MRAM network currently has a different purpose than the computing done by the brain, such a solid-state memory network can in the future be used as a platform to mimic the brain by modeling the synapse of the brain. Connectivity.”
As emphasized in this work, building its leading memory technology and merging it with systematic semiconductor technology, Samsung plans to continue to expand its leadership in next-generation computing and AI semiconductors.