Will one-terahertz MRAM memory revolutionize memory access?

A thousand billion changes per second: that’s the mind-boggling speed of MRAM. A true Arlesian of technologies, will this memory, which has been in development since the 1980s, turn our computers upside down?

Will an article published by the University of Tokyo change the face of our electronic devices? Unless you have a time machine and look at what happens in ten years, hard to say. But one thing is certain: this publication has what it takes to dream of those who have been waiting for the advent of ultra-fast memory, which has been in development for four decades.

Read also: Intel and Samsung have been developing the memory of the future… for several decades! (Dec. 2018)

MRAM or Magnetic Random Access Memory is, on paper, a Grail of memory in the broad sense. Judge for yourself: its access speed is of the order of a millisecond, it consumes no energy at all when it is not used and retains information once the device is turned off. Basically, it combines the strengths of RAM (random access memory) and ROM. It’s beautiful, it’s perfect, it could solve a lot of problems, especially in supercomputers, and… it’s still not mass-industrialized.

The advantages of MRAM:

  • It is non-volatile (like our hard drives or SSDs)
  • It is up to 1000 times faster than DRAM or Flash,
  • It is not necessary to erase previously stored data before being able to write other
  • It consumes little energy
  • It is theoretically indestructible

As the first “M” in its name suggests, MRAM is based on magnetic principles. Who says in return “magnetic field”, fields that pose problems for researchers and engineers. In particular the impossibility of arranging the magnetic cells randomly in space. This forces an alignment of the cells (complicated and expensive) which generates, in return, a magnetic field which slows down the reading speed. The researchers who work on it have many tricks, but they often consist of playing with temperatures close to absolute zero… Basically: for the moment, the MRAM works well, but is not yet ready for the mass production of our electronic devices. Well, that was before.

Reduce resistance and work at room temperature

The basic cell of an already available MRAM module.  © Avalanche Technology
The basic cell of an already available MRAM module. © Avalanche Technology

What researchers from the University of Tokyo claim to have achieved in their article published on January 18 in the post Nature under the very sexy title of ” Octupole-driven magnetoresistance in an antiferromagnetic tunnel junction », seems to remove many scientific barriers in the development of a more efficient and easily usable MRAM. Digested and made more readable by our colleagues at ScienceDaily (thanks be to them: popular science is hard!), this article shows two key successes on the part of these researchers.

First, they designed a whole new antiferromagnetic component. Unlike ferromagnetic magnets which generate a field because of the magnetic order which prevails at room temperature, this magnet does not generate this field. You will have understood it here, its absence avoids the slowing down of the writing and reading of information. This allows researchers to promise cell state change speeds of the order of terahertz. That is to say of the order of 10-12 times per second. Yes, it could go very quickly!

Second, this antiferromagnetic component operates at room temperature. There is no need to come to extreme cold to find quantum properties of materials. Which brings MRAM a little closer to our real world.

Other barriers including that of industrialization

Samsung Foundry is also working on mass production of MRAM.
Samsung Foundry is also working on mass production of MRAM.

You think the mass is said, that your time jump in 2033 will show you machines equipped with MRAM? Calm your enthusiasm, even the researchers working on it are not there yet. It is a laboratory researcher’s success on a specific point of science. The sum of the developments necessary for the mass production of a very dense module of MRAM operating on this principle is still enormous.

In the interview with ScienceDaily, Professor Satoru Nakatsuji from the Department of Physics at the University of Tokyo explains that the way the antiferromagnetic magnet is produced is far from trivial: We grow crystals in vacuum, in incredibly thin layers using two processes called molecular beam epitaxy and magnetron sputtering. […] It is an extremely difficult procedure and if we improve it, it will make our lives easier and also allow us to produce more efficient devices. As you will have understood, for the moment this is science coming out of a state-of-the-art laboratory.

Read also: Samsung, memory giant, relies on its ultra-fast MRAM to improve our connected objects (Feb. 2021)

Assuming that the process improves, even if researchers eventually obtain complex memory modules from MRAM, a major challenge will stand in the way of this technology: industrialization. It takes a big leap to get a technology out of a lab into small production for the rest of the research, space for defence. But it often takes a giant leap to get to the magnitude of mass production. The only lever to allow a technology to really take off. The obvious parallel is that of back-illuminated image sensors. While many sensors have been produced in laboratories, especially for space sensors for imaging satellites, it was not until Sony designed the Exmor R industrial manufacturing process that these ultra-sensitive sensors began to appear in low light. First in compacts, then in smartphones.

IBM has also been working on MRAM for a long time and has already considered the different applications for the different MRAM variants.  © IBM
IBM has also been working on MRAM for a long time and has already considered the different applications for the different MRAM variants. © IBM

If researchers manage to take advantage of the discovery of this Japanese unit, a major brick – or even foundations? – may have just been asked today. And the whole industry is listening to the results of this research: the Korean giant Samsung is also betting on it. It must be said that the promises of MRAM memory are enormous. And up to today’s memory challenges. Especially in supercomputers where the speed of processors (CPU, GPU, etc.) has progressed much faster than the speed of memories.

Source :


Leave a Comment

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