The Basic Theory of eMRAM, a Chip Optimized for AI and Next-Generation Automotive in the Data-Driven Era

We wouldn’t be able to define the modern age without the word "data." The amount of data generated worldwide is growing exponentially every year, and the increasing importance of applications such as AI, edge computing, and autonomous driving is accelerating this growth even further. In response, the industry is facing high demand for the development and reliable supply of semiconductor chips that achieve high performance while consuming less power and reducing cost. In this type of environment, embedded magnetic random access memory (eMRAM) — non-volatile memory that uses magnetic domains and integrated into system semiconductors and processors such as micro controllers and systems-on-chips (SoCs) — is emerging as a next-generation product that can provide optimal performance, reliability, and cost effectiveness. In our previous tech article, “Developing the ‘Industry’s Most Energy-Efficient’ Next-Generation MRAM,” we evaluated Samsung's technological innovations that enabled MRAMs to achieve high performance, high density, and the highest energy efficiency by improving switching efficiency and miniaturizing Magnetic Tunnel Junction (MTJ) size. We also reviewed Samsung’s plans to expand its eMRAM portfolio in the future. In this article, we will take a deeper look at eMRAM based on the fundamental principles of MRAM. DRAM and MRAM: What’s the Difference? DRAM has evolved with a continued decrease in cell size, an increase in integration density, and bandwidth improvement, while pursuing low-power consumption as mobile devices become more common. However, DRAM has its shortcomings as a charge-based memory. It needs to continuously refresh to make up for charges that are lost as time passes, even when it is not in operation, which leads to standby power consumption. Furthermore, as integration density increases, so does interference between memory cells, making it extremely challenging to continue scaling DRAM. This is where MRAM comes in. This type of memory is based on the change in resistance of the unit cell at the MTJ. Since it stores data using its “spin” instead of charge, it retains information almost indefinitely and does not require standby power. It’s an architecture that reduces the total amount of power consumed by a memory device, resulting in high energy efficiency. New Progress Based on Advanced MTJ Design The core of Samsung’s eMRAM breakthrough is its enhanced MTJ stack process technology. The MTJ comprises a three-layer structure with an insulating film used as a tunnel barrier between two ferromagnetic¹ layers. One ferromagnetic layer is “free” and can have its magnetization direction easily adjusted, while the other is “pinned” with a fixed magnetization direction.

As shown in the diagram, the MTJ’s resistance varies according to the relative magnetization direction of the free and pinned layers. If the magnetization direction of the quantum magnetic layers are parallel,” MTJ has a low resistance value, expressed as “0” in binary. And if the magnetization directions are “antiparallel,” MTJ has a high resistance value, expressed as “1” in binary. Based on this methodology, Reads and Writes are performed using the 1s and 0s that are the building blocks of our digital data. The initial MTJ design had similar disadvantages to DRAM; higher integration density reduced the distance between cells, resulting in interference and errors due to the influence of the magnetic field. To address these drawbacks, the Spin-Transfer Torque (STT) — also known as current-induced switching — has been utilized. The existing MRAM architecture generates a magnetic field generated by an electric current to change the magnetization direction of the free layer, but the STT method adjusts the free layer’s magnetization direction by passing current directly through the MTJ. STT effectively deals with the interference phenomenon between cells, thus overcoming the previous technical limitations of integration. Samsung Electronics is adopting this method to advance its eMRAM technology and is supplying eMRAM to customers as the most energy-efficient memory product in the foundry field. By taking these steps, we will continue to fulfill our mission of “making our customers’ lives easier and comfortable.”

---

_{¹ Ferromagnetic substances have high magnetic permeability, a definite saturation point, and appreciable residual magnetism.}