Samsung: World-first MRAM with In-memory technology

The scientific paper, titled “A crossbar array of magnetorestive memory devices for in-memory computing”, presents Samsung’s decisive contribution to memory technology and the company’s effort to merge memory and system semiconductors into a next-generation artificial intelligence (AI) chip.

The research was conducted by the Samsung Advanced Institute of Technology (SAIT) in collaboration with Samsung Electronics Foundry Business and Semiconductor R&D Center. Dr. Seungchul Jung, Researcher at SAIT, is the lead author along with Dr. Donhee Ham, SAIT Member and Harvard University Professor, and Dr. Sang Joon Kim, Vice President of Technology at SAIT.

Read also: Incredible: New update for Samsung Galaxy S8

In the typical computer architecture, the data is stored in a memory chip, while the data calculation is performed on a separate processor chip.

In contrast, In-memory computing is a new computational model that aims to perform data storage and data calculation on a memory network.

This computational standard can process large amount of data stored within the memory network itself without having to move it, and as data processing on the memory network is performed in parallel, power consumption is significantly reduced.

The In-memory computing system has emerged as one of the most promising technologies for creating next-generation semiconductor chips, with built-in AI technology and low power consumption.

The researchers of Samsung Electronics, managed to develop a MRAM chip that integrates in-memory computing technology, replacing the standard, “cumulative” computational memory architecture with a new computational “resistance” memory architecture, which addresses the problem of small resistances of individual MRAM devices.

Then, Samsung’s research team tested the performance of the new MRAM device, performing AI (artificial intelligence) functions. The new chip yielded 98% accuracy to the classification of handwritten digits and 93% accuracy to the detection of faces from scenes.

By integrating MRAM memory, already in commercial-scale production during the manufacture of semiconductor systems, into the realm of in-memory computing, Samsung’s research is expanding the boundaries of next-generation semiconductor chip technologies, with AI technology and low power consumption.

The researchers also proposed the use of the new MRAM chip not only in in-memory computing systems, but also as a platform for capturing biological neural networks, related to neuromorphic electronic vision. This is something that Samsung researchers recently proposed in an episcopal research paper published in the September 2021 issue of the journal Nature Electronics.

This research, based on state-of-the-art memory technology combined with semiconductor technology, reflects Samsung’s plans for continuous evolution and innovation in next-generation computing systems and artificial intelligence chips.

[via]

Spread the love

Leave a Reply

Your email address will not be published. Required fields are marked *