Samsung achieves breakthrough in MRAM technology that 'mimics the brain'

In MRAM, information is stored by the magnetometer of the magnetic thin film. The information bits are inverted when the degrees are reversed. Information is read through the change in resistance of the magnetic contact layer. When the magnetic degrees of the magnetic layers are in the antiparallel state, the resistance of the magnetic contact is large, corresponding to bit (1), while when the system is in the parallel state, the resistance decreases sharply, and corresponds to the bit ( 0).

Samsung is one of the pioneer manufacturers in the field of MRAM research and development for many years. However, so far, the application of MRAM technology in practice (and especially on a commercial scale) is still extremely limited. Low resistance makes MRAM significantly more power hungry than other RAM technologies. The Korean manufacturer has been working to overcome this barrier with a new initiative, which is in-memory computing technology using MRAM.

In-memory computing can be understood as a new computer model designed to perform both data storage and data computation in a memory network. This is one of the promising technologies to realize next-generation low-power AI semiconductor chips.

Explaining the benefits of MRAM and how the technology could prove useful in the development of next-generation AI chips, said Dr. Seungchul Jung, one of Samsung's senior engineers:

'The computational process in memory is intrinsically similar to the brain in certain respects. In the brain, for example, computational processes also occur in biological memory networks, or synapses - points where nerve cells (neurons) come into contact with each other. In fact, although the computational process performed by our MRAM network currently has a different purpose than that performed by the brain, such a solid-state memory network could be used in the future. as a platform to mimic brain activity by modeling neural synaptic connectivity'.

To make this breakthrough, Samsung researchers have developed an MRAM array chip that replaces the current 'current-sum' in-memory computing architecture) with a 'resistance' in-memory computing architecture. sum' (total resistance), thereby helping to solve the problem of the small resistance of individual MRAM devices. In some AI computing tests, the chip achieved accuracy of up to 98% when recognizing handwritten digits, and 93% in facial recognition situations in various environments.

In addition, Samsung is also particularly interested in improving the resistivity of MRAM. Solving this bottleneck opens up the range of potential benefits that MRAM can offer, including increased operating speed, durability, and ease of mass production of hardware. Thereby, the system can process a large amount of data stored in the memory network itself without the need to move the data.