In the present day, as techpowerup reported, Samsung Electronics formally introduced the world’s first pc geared up with MRAM reminiscence. MRAM stands for Magnetoresistive Random Entry Reminiscence. Associated papers are on the “Nature” web site. The title of the thesis is “A crossbar array of magnetoresistive reminiscence gadgets for in-memory computing.” It shows Samsung’s management in reminiscence expertise and on the identical time showcases the corporate’s effort to merge reminiscence and system semiconductors for next-generation synthetic intelligence (AI) chips.

Actually, the expertise was collectively developed by Samsung SAIT Analysis Institute and Samsung Electronics Foundry Enterprise and Semiconductor R&D Middle. The primary creator of the paper is Dr. Sang Joon Kim, Vice President of Know-how at SAIT.

What’s MRAM?

Many of the present computer systems use impartial RAM reminiscence and impartial onerous disk storage. Nonetheless, with the intention to enhance computing effectivity, the trade has been creating non-volatile reminiscence. The latter ought to consider the features of onerous disk and reminiscence, and assist to enormously scale back energy consumption.

To attain this, researchers have developed RRAM (resistive random entry reminiscence), PRAM (phase-change random entry reminiscence), and so forth. On this sense, the MRAM (magnetoresistive non-volatile random entry reminiscence) utilized by Samsung has the benefits of excessive pace, sturdiness, and simple mass manufacturing. However attributable to energy consumption points, Samsung hasn’t given inexperienced gentle to it.

Samsung Electronics researchers present options by means of revolutionary architectures. Particularly, the researchers efficiently developed a brand new MRAM array. It replaces the present structure with a brand new “resistance” and computational structure.

What’s extra attention-grabbing, the pc with MRAM reminiscence has handed the AI ​​computing efficiency check. Surprisingly, the accuracy of recognizing handwritten numbers has reached 98%, and the accuracy of face recognition has reached 93%.

The researchers mentioned that in-memory computing is the primary place the place they use MRAM chips. They’re very appropriate for neural community operations as a result of this computing structure is just like the mind neuron community.

“In-memory computing attracts similarity to the mind within the sense that within the mind, computing additionally happens throughout the community of organic recollections, or synapses, the factors the place neurons contact each other,” mentioned Dr. Seungchul Jung, the primary creator of the paper. “Actually, whereas the computing carried out by our MRAM community, for now, has a special goal from the computing carried out by the mind, such solid-state reminiscence community could sooner or later be used as a platform to imitate the mind by modeling the mind’s synapse connectivity.”