Samsung showcases the world's first MRAM-based in-memory computing for next-generation AI computing

The research was conducted by the Samsung Advanced Technology Research Institute (SAIT) together with Samsung Electronics Foundry Business and Semiconductor R&D Center, and was conducted by a team led by SAIT Employee Researcher Dr. Seungchul Jung, SAIT Researcher and Harvard University Professor Dr. Donhee Ham, and DR. Sang Joon Kim, Vice President of SAIT Technology.

In modern standard computer architectures, data is stored in memory chips and data computation is performed in processors. In contrast, in-memory computing is a new type of computing designed to perform both data storage and data computations in an in-memory network. Since the scheme can handle large amounts of data stored in memory itself without transferring the data, and the data processing in memory is carried out in a highly parallel manner, power consumption can be greatly reduced. Therefore, in-memory computing has become one of the most promising technologies to achieve the next generation of low-power artificial intelligence semiconductor chips.

Samsung researchers have provided solutions through architectural innovations, successfully developing a MRAM array chip that applies a new memory computing architecture called "resistance sum" to solve the problem of small resistance of a single MRAM device. Samsung's research team used the MRAM chip to perform AI operations and tested its performance, and the results showed an accuracy rate of 98% in the classification of handwritten digits and 93% accuracy in detecting faces in scenes.

Samsung believes that the MRAM chip can also be used as a platform to download the bion neuron network in the future, because its computing architecture is more similar to the brain neuron network.