SAMSUNG testing first MRAM computers: New era of AI computers coming

techpowerup reports that Samsung Electronics has officially announced the first MRAM-equipped computer in the world. MRAM stands to Magnetoresistive Random Access Memory. You can find related papers on the Nature website. The thesis title is “A crossbar array magnetoresistive memory device for in-memory computation.” This paper highlights Samsung’s memory technology leadership and showcases the company’s efforts to combine memory and system semiconductors in order to create next-generation artificial intelligence (AI).

samsung emram

The technology was actually developed jointly by Samsung Electronics Foundry Business and Semiconductor R&D Center. Dr. Sang Joon Kim is Vice President of Technology at SAIT.

MRAM: WHAT IS IT?

The majority of current computers have independent RAM memory and hard disk storage. Non-volatile memory has been developed in an effort to increase computing efficiency. These memory should be able to take into consideration the different functions of hard drive and memory and will greatly reduce power consumption.

Researchers have created PRAM (phase-change randomly access memory), RRAM (resistive Random Access Memory), and others to achieve this. Samsung’s MRAM (magnetoresistive, non-volatile random address memory) has many advantages, including high speed, durability and easy mass production. However, Samsung has not given the green light due to power consumption.

Samsung Electronics researchers offer innovative solutions. The researchers developed a new MRAM array. It replaces the existing architecture with a new “resistance”, and computational architecture.

The computer with MRAM memory passed the AI computing performance tests. Surprisingly the computer with MRAM memory has passed the AI computing performance test. The accuracy in recognizing handwritten numbers is 98% and face recognition is 93%.

According to the researchers, MRAM chips are most commonly used in memory computing. These chips are ideal for operations in neural networks, as their computing architecture is very similar to that of the brain neuron network.

“In-memory computation has a similarity to the brain in that computing occurs within the network or biological memories (or synapses) of neurons touching one another,” Dr. Seungchul Jung was the first author of this paper. Although the MRAM network’s computing serves a different purpose than the computing done by the brain, the solid-state memory network could be used in the future to model the brain’s connectivity.

Get latest news from African Startup ecosystem

Latest stories

You might also like...