Samsung announces the world's first computer equipped with MRAM non-volatile magnetoresistive memory: it can be used for AI computing

IT Home January 13 news, according to foreign media techpowerup reported that Samsung Electronics today officially announced the world's first computer equipped with MRAM memory, the relevant paper published on the "Nature" website, and will be published in the magazine soon. The title of the paper is "Cross-array of Magnetoresistive Storage Devices for In-Memory Computing Computers", and the related computers can be used for AI operations.

This technology was jointly developed by Samsung SAIT Research Institute and Samsung Electronics Foundry Business and Semiconductor R&D Center, and the first author of the paper is Dr. Seungchul Jung.

Most of the current computers use independent RAM memory and independent hard disk storage, but in order to improve the efficiency of computing, the industry has been developing non-volatile memory, which can take into account the functions of hard disk and memory, while helping to greatly reduce power consumption.

In order to achieve this goal, researchers have developed RRAM (resistive random access memory), PRAM (phase change random access memory), etc., and already have prototypes. The MRAM (Magnetoresistive Nonvolatile Random Access Memory) used by Samsung has the advantages of high speed, durability, and easy mass production, but due to power consumption problems, it has not been officially unveiled before.

Researchers at Samsung Electronics provide solutions through innovative architectures. Specifically, the researchers successfully developed a new MRAM array that replaces the current architecture with a new "resistor" and computing architecture.

IT Home learned that the computer equipped with MRAM memory has been tested by AI computing performance, and the accuracy of recognizing handwritten digits has reached 98%, and the accuracy of face recognition has reached 93%.

The researchers said that MRAM chips are applied to in-memory computing computers, which are very suitable for neural network operations, etc., because this computing architecture is more similar to the brain neuron network.