Scalable Cell Technology Utilizing Domain Wall Motion for High-speed MRAM

Abstract

We propose a new MRAM cell that stores data in the form of the domain wall (DW) position. The DW is moved by the spin-polarized current that flows in the free layer. The cell was fabricated and the writing characteristics were investigated. A writing current of the cell was scalable, and the current density was reduced by using a new material. The cell is suitable for a high-speed MRAM that will compete with an eSRAM.Introduction An MRAM is a strong candidate for a high-speed non-volatile RAM due to its high speed random access, unlimited endurance, and compatibility with a scaled-down CMOS processes [1][2][3] . A 2T1MTJ cell was proposed for a high-speed MRAM macro in SoC operating at over 200 MHz [4] . Toggle MRAM makes it possible to enlarge the operating margins [3] , however, such field writing MRAMs have difficulties in scalability of the writing currents. Developing and adopting new magnetic technologies will be required for size reduction [5][6] . We propose a new MRAM cell for scalability of the writing current. The proposed cell has 2T1MTJ structure and it is suitable for a high-speed MRAM [4] .DW-seesaw Cell The proposed cell includes an MTJ with a U-shaped free layer ( Fig. 1a-b). The magnetization directions of the two end parts are fixed, and the two corners are DW trapping sites. A DW moves in the central part, and the DW positions are corresponded to the stored data. A tunnel barrier and a pinned magnetic layer are formed on the central part, and the DW position is detected by measuring the MTJ resistance. In writing, the fixed magnetization region acts as the source of the spin-polarized electrons. The electrons flow from one fixed magnetization region to the other through the DW moving region. The DW should move in line with the electron flow direction [7] . Its behavior is like a seesaw, and we call the cell a DW-seesaw. While a Race-track-memory deals with a number of DWs [8] , a single DW is treated in the DW-seesaw cell. Each fixed magnetization region is connected to a transistor (Fig. 1c). The circuitry of the cell (Fig. 1d) is the same as the 2T1MTJ cell in [4] and it suitable for high speed operation. This memory cell has some advantages. As the cell size is reduced, the writing current and time decrease. The current paths of the writing and reading operations are separated, and the writing current goes through only the free layer. Consequently, the tunnel barrier will be free from the voltage stress and the stored data will be free from the reading disturbance. This is a completely different point to a usual spin injection MRAM. Magnetic ConfigurationThe magnetic configurations of a 300-nm-wide, 10-nm-thick NiFe element were calculated using OOMMF [9] , a micro magnetic simulator. Figs. 2a and b depict the results for a 0-state and a 1-state, respectively. The magnetic configuration was directly observed by magnetic force microscopy (MFM) (Fig. 2c). The bright spots in the right corners indicated DWs. Moreover, we measured dc resistance of the left (R left ) and right (R right ) legs...

Emerging memories

“…The alternative domain wall, proposed in 2007 [24], in which programming does not imply the polarization reversal in the 'soft' layer, but only the lateral motion of the wall separating two domains with opposite polarization. The reduction of the stress on the tunnel oxide is obtained at the expenses of a larger cell.…”

Current-Driven Domain Wall Motion, Nucleation, and Propagation in a Co/Pt Multi-Layer Strip with a Stepped Structure

“…One of the applications is magnetic random-access memories (MRAMs), where reducing driving current with sufficient thermal and magnetic stability is required [2]. The threshold current density for the widely used NiFe nanostrips is around A/m for a small DW pinning field of 5 Oe [3].…”

Reducing shift penalty in Domain Wall Memory through register locality

“…When DWM was first introduced, the write operation was performed by MTJ using Spin-Transfer-Torque mechanism (STT) [19] which is similar to write operation in STT-RAM [13]. This method of write operation is slow and energy hungry.…”