Magnetic Memory Devices

Figure 1: My research on non-volatile magnetic memory and comparison with existing memory technologies.

Spin-transfer torque (STT) magnetoresistive random access memory (MRAM) is advancing as a commercial product for a significant system-level improvement by replacing the battery-backed SRAM and DRAM in standalone, embedded, and cache memory applications. It uses magnetic tunnel junctions (MTJ) as bits, comprising two magnetic layers sandwiching a thin MgO barrier (see Fig. 1a), which are compatible with the back end of line CMOS processes and operate at CMOS-compatible voltages. State-of-the-art MTJs exhibit only a 3 times on-off ratio and require a large write current ~107 A/cm2, which leads to reliability issues, high energy (~1 pJ/bit), and density limitations.

My research includes:

  1. Current-controlled, SOT MRAM.

  2. Voltage-controlled, REC MRAM.

I have identified pathways to lower the energy to ~10 fJ/bit and enhance the on-off ratio to 103/cell.


Spin-Orbit Torque MRAM (SOT-MRAM)

The demonstration of the giant spin Hall effect (Liu et al., Science, 2012) led to the proposal of SOT-MRAM (see Fig. 1b), where the write current is applied in a separate metallic or semiconducting layer under the storage (free) magnet of the MTJ. This mechanism separates the read and write paths, and partly addresses the reliability issues in STT-MRAM, as the large write current do not flow through the MTJ.

Let's look at the underlying mechanism. The write current density (Jc) is determined by spin Hall effect as

where θSH is the spin Hall angle. The switching spin current density (Js) is fixed by the properties magnetic bit, which ~106 A/cm2 for a 50kT stable magnetic bit with 50 nm diameter. Existing transition metals (e.g. Pt and Ta) exhibits a spin Hall angle in the order of ~0.1, which yields a minimum write current density as high as ~107 A/cm2.

Thus, just from the physics of spin Hall effect, it is obvious that we need a spin Hall angle ~1000 if we want to lower our switching current by 4-orders of magnitude , and there is an ongoing search for efficient materials.

A pathway to achieve large spin Hall angle for efficient SOT-MRAM

I developed a semiclassical model that identified new scaling trends of the widely known figure-of-merits in terms of density of states (DOS), which agree with the experiments on diverse materials. I have theoretically pointed out that the spin Hall angle or spin torque efficiency in a spin-orbit material is related to its intrinsic degree of spin momentum locking (p0), number of modes or density of states (mn), and the spin-absorption at the magnetic interface (gso), as

which agrees well with the existing experiments on diverse classes of materials. Here, the degree of spin momentum locking can be any number between 0 and 1 and represents the % of spin-splitting in the spin-orbit coupled channel.

My model suggests that, for a given spin-orbit coupling and a given interface spin conductance, the spin Hall angle can be enhanced by reducing the density of states in the spin-orbit layer, i.e., a material with 103x lower DOS than Pt can achieve a write current of 104 A/cm2.

Related manuscript:

  • S. Sayed, S. Hong, X. Huang, A. S. Everhardt, L. Caretta, R. Ramesh, S. Salahuddin, and S. Datta, "Unified Framework for Charge-Spin Interconversion in Spin-Orbit Materials", Phys. Rev. Applied, 15, 054004, May 2021. (Link).

  • S. Sayed, S. Hong, and S. Datta, "Transmission-Line Model for Materials with Spin-Momentum Locking", Phys. Rev. Applied 10, 054044, November 2018. (Link).

New materials for efficient SOT MRAM

I have identified a new material with low DOS, strontium iridate (SrIrO3) and calculated a high SOT efficiency for this material and observed experimentally on a non-epitaxial SrIrO3/Py system with a collaboration with the Department of Materials Science & Engineering at UC Berkeley and Department of Electrical and Computer Engineering at University of Minnesota. In the semimetallic phase of SrIrO3, we have observed that the SOT efficiency increases for thicker devices while the resistivity of the sample decreases. This observation is counter-intuitive to the observations in metals where higher resistivity exhibits higher SOT efficiency. On the other hand, the Hall carrier concentration measured on bare SrIrO3, which is a measurement of the DOS, indicated that the concentration is also decreasing for thicker samples, which indicated that the resistivity and DOS in these materials can behave differently due to a modulation in the mean free path. Interestingly, the scaling of the SOT efficiency with respect to the carrier concentration agreed well with my model and indicates that there is a scope to tune the interface strain state to further enhance the SOT efficiency.

I have further collaborated with the Department of Materials Science & Engineering at UC Berkeley and Department of Physics at Cornell University to demonstrate a high SOT efficiency in an epiataxially grown SrIrO3/LSMO heterostructure which exhibits a novel trend with respect to the SrIrO3 thickness. The epitaxially grown interface gives a unique interfacial effect that enhances both SOT efficiency of the SrIrO3 and effective magnetization of the LSMO within the bilayer, as compared to individual layers. The heterostructure promises efficient SOT devices with minimal current shunting.

In addition, recently our group has demonstrated a large SOT efficiency in silicides like Fe1-xSix, where the silicide shows high SOT and negligible magnetization when x < 50% and a sizable magnetization and negligible SOT when x > 50%. This implies that both magnetic bit and SOT write mechanism can be incorporated in the same material within a silicon platform. More interestingly, the large SOT is observed in a purely amorphous phase of the material where the concept of typical energy vs. momentum band diagram is not well defined. Thus, this observation is counter-intuitive to the established understanding on metallic systems that a large SOT can come from a band structure within certain crystalline form of materials. However, the DOS is a well-defined parameter in the amorphous phase of material and a model description in terms of DOS holds.

Related manuscript:

  • S. Susarla , X. Huang , S. Sayed , L. Caretta , H. Zhang , S. Salahuddin , P. Ercius, and R. Ramesh, "Atomic scale understanding of the electronic structure of 5d-3d perovskite oxide heterostructures using STEM-EELS", Microscopy and Microanalysis, 27(S1), 356-358, July 2021. (Link).

  • X. Huang, S. Sayed, J. Mittelstaedt, S. Susarla, S. Karimeddiny, L. Caretta, H. Zhang, V. A. Stoica, T. Gosavi, F. Mahfouzi, Q. Sun, P. Ercius, N. Kioussis, S. Salahuddin, D. C. Ralph, and R. Ramesh, "Novel spin-orbit torque generation at room temperature in an all-oxide epitaxial La0.7Sr0.3MnO3/SrIrO3 system", Advanced Materials, 2008269, May 2021. (Link).

  • S. Sayed, S. Hong, X. Huang, A. S. Everhardt, L. Caretta, R. Ramesh, S. Salahuddin, and S. Datta, "Unified Framework for Charge-Spin Interconversion in Spin-Orbit Materials", Phys. Rev. Applied, 15, 054004, May 2021. (Link).

  • A. S. Everhardt, D. C. Mahendra, X. Huang, S. Sayed, T. Gosavi, Y. Tang, C.-C. Lin, S. Manipatruni, I. Young, S. Datta, J.-P. Wang, R. Ramesh, "Tunable charge to spin conversion in strontium iridate thin films'', Phys. Rev. Materials 3, 051201(R), May 2019. (Link).

  • C.-H. Hsu, J. Karel, N. Roschewsky, S. Cheema, D. S. Bouma, S. Sayed, F. Hellman, S. Salahuddin, "Spin-orbit torque generated by amorphous FexSi1−x", arXiv:2006.07786, June 2020 . (Under Review).

Figure 2: Theoretical model for spin Hall angle or spin torque efficiencies in diverse classes of materials.

Spin Funnel for Efficient SOT-MRAM

In addition to the new materials search, I have predicted a composite structure using an existing SOT material (see Fig. 1c) that exhibits ~6x larger SOT compared to the SOT material operating by itself. The idea is to funnel spins from the large area of the SOT material into the small area of the storage magnet, using a material with lower resistivity and higher spin diffusion length like copper. In order to avoid current shunting, we have inserted a thin layer of pure spin conductor, implemented with coupling between electron spin and magnon within ferromagnetic insulator like yttrium-iron-garnet. The ferromagnetic insulator do not transport charge and transmits only spins to the copper layer. This prediction of mine has been partially demonstrated.

Related manuscript:

  • S. Sayed, V. Q. Diep, K. Y. Camsari, and S. Datta, “Spin Funneling for Enhanced Spin Injection into Ferromagnets”, Scientific Reports 6, 28868, July 2016. (Link).

  • V. Ostwal, A. Penumatcha, Y.-M. Hung, A. D. Kent, and J. Appenzeller, "Spin-orbit torque based magnetization switching in Pt/Cu/[Co/Ni]5 multilayer structures", Journal of Applied Physics 122, 213905, Dec 2017 (Link).

Related patent:

  • S. Sayed, V. Q. Diep, K. Y. Camsari, and S. Datta, “Apparatus for Spin Injection Enhancement and Method of Making the Same'', U.S. Patent, US20180182954A1, 28 June 2018. (Link)

Figure 3: SOT-MRAM using a spin funnel structure.

New phenomena enabling an all-metallic SOT-MRAM without MTJs

I have predicted a long-range three-state magnetoresistance in a multi-terminal lateral spin-valve where one of the antiparallel states is higher than the parallel states and the other one is symmetrically lower. This is surprisingly different from conventional cases, where antiparallel states are usually higher than the parallel states and observed only up to a few hundred nm of separation between the two magnets. My predictions have been independently demonstrated on a semiconductor (InAs) and a 2D material (PtSe2) at room temperature by my collaborators in Korea Institute of Science and Technology (see Fig. 4) and Chalmers University-Sweden, respectively. They have observed the long-range nature of this phenomenon and the magnetoresistance did not degrade up to 1.62 mm of separation between the magnetic contacts at room temperature. This phenomenon offers a new read mechanism for the SOT MRAM without requiring MTJs (see Fig. 1d) and have been used to demonstrate a reconfigurable logic for memory-in-compute.

I have used this phenomenon to propose a novel SOT MRAM (see Fig. 1d) where we can replace the whole MTJ stack with just the storage magnetic bit. This unique idea makes the read and write paths same again (like STT) but both the currents flow through the SOT material. Thus the proposed SOT-MRAM can be implemented within the same control circuitry as STT-MRAM. The proposed device also allows on-cell self-referencing that will help against the process variation in an array. The read signal depends on the current induced spin voltage in the material which scales inversely with the materials density of states. Existing metallic material can provide a read-out signal in the few mV range. I have identified a new scaling trend of this novel magnetoresistance induced voltage could scale inversely with the DOS of the material, see Fig. 5, which agrees with experiments on diverse materials.

Related manuscript:

  • S. Sayed, S. Hong, and S. Datta, “Multi-Terminal Spin Valve on Channels with Spin-Momentum Locking”, Scientific Reports 6, 35658, October 2016. (Link).

  • J.-H. Lee, H.-J. Kim, J. Chang, S. H. Han, H.-C. Koo, S. Sayed, S. Hong and S. Datta, “Multi-terminal spin valve in a strong Rashba channel exhibiting three resistance states”, Scientific Reports 8, 3397, February 2018. (Link).

  • S. Sayed, S. Hong, E. E. Marinero, and S. Datta, "Proposal of a Single Nano-Magnet Memory Device", IEEE Electron Device Letters 38, 1665-1668, December 2017. (Link).

  • J.-H. Lee, S. Hong, H.-J. Kim, J. Chang, and H. C. Koo, "Reconfigurable spin logic device using electrochemical potentials", Appl. Phys. Lett. 114, 152403, April 2019.

Related patent:

  • S. Sayed, S. Datta, and E. E. Marinero, “Single nanomagnet memory device for magnetic random access memory applications”, U.S. Patent, US20180233188A1, 16 Aug 2018. (Link)

Figure 4: Prediction and demonstration of a novel magnetoresistance in a multiterminal lateral spin-valve with spin-orbit materials.

Figure 5: Proposed SOT-MRAM without MTJs.

Resonant Exchange Controlled MRAM (REC-MRAM)

I have predicted a voltage-controlled write mechanism that can be engineered into conventional MTJs by replacing the MgO barrier with a MgO/Ru/MgO barrier (see Fig. 1e). Such a barrier can tune the exchange coupling between the magnets using resonant tunneling and enable a resonant-exchange-controlled (REC) switching with low write energy in the order of 10 fJ. Bidirectional switching is possible with the same polarity of the voltage, unlike conventional magnetic memory devices where a bidirectional current or a magnetic field is necessary. REC switching energy is independent of the dynamic parameters like Gilbert damping due to conservative nature of the exerted torque, where the switching speed scales inversely with the damping. Thus, REC switching energy and delay are decoupled where conventional mechanisms show a trade-off. Based on detailed simulations using existing materials parameters, CoFeB/MgO/Ru, we show that the device current at the switching voltage can be lowered by three orders of magnitude than conventional spin-torque devices while the switching speed is in the order of ns. The calculated energy and delay of the device are very close to that observed in embedded DRAM and SRAM, respectively see Fig,1 .

For the materials combination used for simulations, the magnetoresistance (MR) is expected to be ~35%. Such a REC MTJ integrated into a CMOS gain-cell made with standard CMOS models (see Fig. 6) promises a 103x change in the read current between '0' and '1' states, while retaining the low-energy advantages of the voltage-controlled switching.

Currently, we are working to demonstrate the mechanism experimentally.

Related manuscript:

  • S. Sayed, C.-H. Hsu, N. Roschewsky, S.-H. Yang, and S. Salahuddin, "Resonant enhancement of exchange coupling for voltage-controlled magnetic switching", Phys. Rev. Applied 14, 034070, September 2020. (Link).

  • S. Sayed, C.-H. Hsu, and S. Salahuddin, "A voltage-controlled gain cell magnetic memory", IEEE Electron Device Letters, 42(10), 1452 - 1455, Aug 2021. (Link).

Related patent:

  • S. Sayed and S. Salahuddin, "Electric-Field Controlled Interlayer Exchange Coupling for Magnetization Switching", U.S. Patent, US2021/0012940 A1, 14 Jan 2021. (Link)

  • S. Sayed, and S. Salahuddin, "Voltage-controlled gain cell magnetic memory'', UC case number: BK-2021-065.

Figure 6: Proposed voltage-controlled REC-MRAM and its operations.