Magnetic Page Memory: 3D and 2D Architectures

• Magnetic page memory is a storage architecture that organizes magnetic bits into 2D arrays or 3D stacked layers, enabling coherent transfer and replication of entire pages.
• Three-dimensional nanowire implementations use thermally assisted domain transfer with controlled vertical and lateral switching to achieve effective page-level data replication.
• Two-dimensional bubblecade and strain-engineered nanoisland approaches illustrate alternative methods for coherent page transport and ultra-dense, low-power magnetic storage.

Magnetic page memory is a magnetic-storage concept in which information is organized as a two-dimensional array or as stacked magnetic layers that behave like “pages” of bits, and in which stored patterns are transported or replicated rather than accessed only as isolated one-bit cells. In the literature considered here, the term is used most directly for three-dimensional stacked nanowire memories with vertical and lateral domain transfer and for two-dimensional transport of bubble-domain arrays; related work extends the idea toward page-memory-like patterned media based on strain-defined ferromagnetic nanoislands, while some other magnetic memories are better understood as adjacent but non-equivalent architectures (Ozatay et al., 2018, Moon et al., 2014, Chen et al., 2023, 2207.14427).

1. Definition and conceptual boundaries

In the conventional usage that emerges from domain-based spintronics, magnetic page memory denotes a storage architecture in which a two-dimensional array of magnetic bits is moved coherently past fixed read/write elements, or a stacked set of magnetic layers is treated as multiple pages with controlled inter-page transfer. The paper on bubblecade memory states this explicitly by contrasting racetrack memory, which is a one-dimensional shift register using domain walls in nanowires, with magnetic page memory, where a two-dimensional bit array is moved like a “page” sliding under a sensor (Moon et al., 2014). The three-dimensional nanowire work generalizes the same idea by making each magnetic layer a page of bits and enabling both vertical copying between pages and lateral motion within a page (Ozatay et al., 2018).

This boundary matters because not every magnetic memory with arrays, multibit states, or write protection is a page-memory architecture. The electrically function-switchable domain-wall memory based on a U-shaped Pt/Co/Ru/AlOx device is described as bit-level secure magnetic memory, and the source explicitly notes that it is “not” page memory in the conventional sense because it does not implement page-addressable storage or multi-bit page-level transport (2207.14427). A similar distinction applies to single-cell proposals such as topological-insulator Hall-readout memory cells, Josephson magnetic memory cells, and piezoelectric antiferromagnetic memories: they are magnetic memory elements, but they are not page memories in the architectural sense described above (Fujita et al., 2011, 1711.01681, Yan et al., 2019).

A broader implication is that magnetic page memory is best treated as an architectural category rather than a single device physics. Across the sources, the unifying feature is page-like organization, transport, or replication of many magnetic states, whereas the underlying mechanisms vary from thermally assisted domain transfer, to chiral bubble drift, to strain-defined patterned magnetic units (Ozatay et al., 2018, Moon et al., 2014, Chen et al., 2023).

2. Three-dimensional stacked-page architectures

The clearest direct realization appears in “Three-Dimensional Magnetic Page Memory,” which uses vertical stacks of CoNi/Pd magnetic nanowires with perpendicular anisotropy, periodic constrictions, and a SiN$_x$ spacer that is electrically insulating and of low thermal conductivity while remaining magnetically transparent enough for interlayer stray-field coupling (Ozatay et al., 2018). The constrictions serve simultaneously as bit-storage sites, domain-pinning sites, and local current-density concentrators for Joule heating. The resulting architecture is a cross-wire stack in which one layer runs horizontally and the other vertically.

The stored bit is the presence or absence of a reversal domain at a constriction. Vertical information flow is achieved by thermally assisted, stray-field-induced domain transfer. In the demonstrated sequence, a top layer is first written; then a current pulse heats the adjacent target region in the neighboring layer; heating lowers the local coercive field; and the stray field from the already written layer induces reversal in the heated neighbor. After the pulse, the second layer retains a replica of the first. Magnetometry identifies a thermal write window of roughly $110^\circ \mathrm{C}$ to $125^\circ \mathrm{C}$, and at about $125^\circ\mathrm{C}$ the coercivity drops to nearly zero, allowing the adjacent layer’s dipolar field to complete the switching (Ozatay et al., 2018).

The same platform also supports lateral information flow inside a page. Spin-polarized current pulses drive domains along a nanowire with multiple constrictions. In the double-constriction device, pulses around $2.2\,\mathrm{V}$, $1\,\mathrm{ns}$, corresponding to $\sim 1\times 10^8\,\mathrm{A/cm^2}$, produce direct replication to the next constriction, whereas weaker pulses around $1\,\mathrm{V}$, $1\,\mathrm{ns}$, around $4.5\times 10^7\,\mathrm{A/cm^2}$, produce gradual propagation over multiple pulses. The paper attributes this regime dependence to spin-transfer torque together with current-induced heating that modifies the effective anisotropy and therefore the domain-wall dynamics (Ozatay et al., 2018).

Experimentally, the work reports several direct signatures of page operation. Magnetic force microscopy shows top-layer writing, followed by bottom-layer reversal under pulse-plus-stray-field conditions. When the lower layer is also switched, the MFM signal increases by about $110^\circ \mathrm{C}$0. Electrical detection on the bottom wire shows a reversible resistance change of about $110^\circ \mathrm{C}$1, consistent with domain-wall anisotropic magnetoresistance. Finite-element thermal analysis further indicates that the target wire can reach the writing temperature while a neighboring wire remains around $110^\circ \mathrm{C}$2, below the write threshold, which is central to limiting thermal cross-talk in stacked operation (Ozatay et al., 2018).

In this architecture, the term “page” is literal rather than metaphorical: each magnetic layer is a page, vertical copying transfers information from one page to another, and spin-torque shifting provides lateral motion within a page. This is the strongest direct example of magnetic page memory as a three-dimensional storage system (Ozatay et al., 2018).

3. Two-dimensional page transport by magnetic bubblecade

A second direct route replaces stacked nanowires with a moving sheet of bubble domains. “Magnetic Bubblecade Memory” demonstrates coherent, unidirectional motion of an array of magnetic bubble domains in a film, without current injection and without structural modulation, by exploiting chiral domain walls in asymmetric multilayers such as Pt/Co/Pt (Moon et al., 2014). In the paper’s formulation, this provides two-dimensional data-storage capability and therefore functions as a page-like transport scheme.

The physical mechanism is Dzyaloshinskii–Moriya-interaction-induced chirality. The bubble wall has a right-handed chiral configuration in which the domain-wall magnetization $110^\circ \mathrm{C}$3 points radially outward. Under a purely out-of-plane field $110^\circ \mathrm{C}$4, a bubble expands or shrinks circularly. When an in-plane field $110^\circ \mathrm{C}$5 is added, the Zeeman interaction between $110^\circ \mathrm{C}$6 and the chiral wall magnetization creates an asymmetric domain-wall-energy distribution around the bubble. Because the domain-wall speed depends on wall energy, one side of the bubble moves faster than the other. Alternating $110^\circ \mathrm{C}$7 and then $110^\circ \mathrm{C}$8 causes asymmetric expansion followed by asymmetric shrinkage, but in both cases the bubble center shifts laterally in the same direction. Repetition yields net drift. The field sequence can be generated with one coil tilted by $110^\circ \mathrm{C}$9 relative to the film normal (Moon et al., 2014).

The key dynamical quantity is the bubble velocity,

$125^\circ \mathrm{C}$0

with the approximate linear relation

$125^\circ \mathrm{C}$1

In the creep regime, the coefficient is given by

$125^\circ \mathrm{C}$2

The paper reports $125^\circ \mathrm{C}$3 at the maximum $125^\circ \mathrm{C}$4. At the maximum $125^\circ \mathrm{C}$5, $125^\circ \mathrm{C}$6 reaches about $125^\circ \mathrm{C}$7 of $125^\circ \mathrm{C}$8, and the maximum observed $125^\circ \mathrm{C}$9 is about $125^\circ\mathrm{C}$0 (Moon et al., 2014).

The page-memory significance lies in coherent pattern preservation. The paper demonstrates motion of a $125^\circ\mathrm{C}$1 bubble array over about $125^\circ\mathrm{C}$2 while maintaining the same pattern. This is structurally simpler than nanowire ratchets because it works in unpatterned thin films and avoids high current densities and Joule heating associated with current-driven racetrack devices. The authors propose a device with writing and reading sections, alternating magnetic pulses as the operation clock, thermomagnetic bubble writing in the experiment, and prospective spin-transfer-torque writing and tunneling magnetoresistive sensors in future implementations (Moon et al., 2014).

An important operational refinement is control of radius oscillation during transport. Under sinusoidal alternating fields, the bubble-radius variation $125^\circ\mathrm{C}$3 decreases inversely with frequency $125^\circ\mathrm{C}$4, while the average bubble speed $125^\circ\mathrm{C}$5 remains nearly constant. This means transport can be made more page-like, in the sense of preserving the bit pattern during motion, by increasing frequency within the bandwidth limits of the field source (Moon et al., 2014).

4. Materials-defined patterned media and page-memory-like storage

A different lineage does not move a page of domains through a uniform film, but instead defines isolated magnetic units intrinsically within a laterally heterogeneous material. “Miniature Magnetic Nano islands in a Morphotropic Cobaltite Matrix” presents this explicitly as a route to magnetic page memory–like patterned storage by creating switchable ferromagnetic nanoislands inside a nonmagnetic oxide matrix (Chen et al., 2023). The motivation is the need for higher-density magnetic memories that avoid superparamagnetic collapse and minimize inter-bit crosstalk without relying solely on top-down lithographic patterning.

The platform is a LaCoO$125^\circ\mathrm{C}$6 cobaltite homostructure whose magnetic state is highly sensitive to epitaxial strain. The paper defines a morphotropic cobaltite matrix as a laterally heterogeneous LaCoO$125^\circ\mathrm{C}$7 film in which adjacent regions are forced into different crystallographic and strain states by different underlying templates. One region is grown directly on LaAlO$125^\circ\mathrm{C}$8, which produces compressive strain and a nonferromagnetic state denoted LCO$125^\circ\mathrm{C}$9. Another region is grown on freestanding ultrathin SrTiO$2.2\,\mathrm{V}$0 membranes, producing tensile strain and a ferromagnetic state denoted LCO$2.2\,\mathrm{V}$1. Because the membranes are freestanding and only weakly bonded, they can partially relax and thereby impose a distinct local epitaxial environment within the same film. Atomically sharp grain boundaries between LCO$2.2\,\mathrm{V}$2 and LCO$2.2\,\mathrm{V}$3 isolate magnetic crosstalk, which is essential for defining discrete memory cells (Chen et al., 2023).

The microscopic mechanism is a strain-driven spin-state transition in Co$2.2\,\mathrm{V}$4, mediated by strong spin–orbital coupling and oxygen-cobalt hybridization. Tensile-strained LCO$2.2\,\mathrm{V}$5 favors in-plane orbital occupation and stabilizes a higher-spin ferromagnetic state, whereas compressive LCO$2.2\,\mathrm{V}$6 favors a low-spin, weakly magnetic or nonferromagnetic state. The paper supports this with XRD and STEM for strain state and interface sharpness, SHG for symmetry-group differentiation, NV-center magnetometry showing a strong temperature-dependent magnetic signal only in LCO$2.2\,\mathrm{V}$7, MFM imaging of field-switchable domains only in tensile regions, and XAS/XLD showing negative XLD in LCO$2.2\,\mathrm{V}$8 and positive XLD in LCO$2.2\,\mathrm{V}$9. The XLD is defined as

$1\,\mathrm{ns}$0

The discussion also invokes the balance between crystal-field splitting $1\,\mathrm{ns}$1 and exchange interaction $1\,\mathrm{ns}$2, and gives a qualitative strain-dependent band-overlap expression,

$1\,\mathrm{ns}$3

to explain how reduced tensile strain increases $1\,\mathrm{ns}$4–$1\,\mathrm{ns}$5 overlap and stabilizes the higher-spin ferromagnetic configuration (Chen et al., 2023).

Quantitatively, the smallest fabricated ferromagnetic nanoislands have a diameter of about $1\,\mathrm{ns}$6, more precisely $1\,\mathrm{ns}$7 in supplementary STEM analysis, corresponding to an areal density of about $1\,\mathrm{ns}$8. The paper states that reducing the lateral bit size below $1\,\mathrm{ns}$9 would push patterned media toward the $\sim 1\times 10^8\,\mathrm{A/cm^2}$0 regime. The LCO films are about $\sim 1\times 10^8\,\mathrm{A/cm^2}$1–$\sim 1\times 10^8\,\mathrm{A/cm^2}$2 thick, and the FS-STO membranes are about $\sim 1\times 10^8\,\mathrm{A/cm^2}$3–$\sim 1\times 10^8\,\mathrm{A/cm^2}$4 unit cells thick, roughly $\sim 1\times 10^8\,\mathrm{A/cm^2}$5. The tensile strain on FS-STO-supported LCO is reported around $\sim 1\times 10^8\,\mathrm{A/cm^2}$6–$\sim 1\times 10^8\,\mathrm{A/cm^2}$7, compared with about $\sim 1\times 10^8\,\mathrm{A/cm^2}$8 tensile strain on bulk STO substrates and about $\sim 1\times 10^8\,\mathrm{A/cm^2}$9 compressive strain on LAO-supported LCO (Chen et al., 2023).

From a page-memory perspective, the importance of this work is that it defines nanoscale magnetic bits through materials-intrinsic lateral phase separation rather than by patterning a continuous ferromagnet. The authors present the nanoislands as programmable, addressable magnetic bits and sketch a crossbar geometry for writing and reading via vertical and horizontal line architectures, with readout envisioned through spin-transfer torque or magnon-based detection in crossbars made from heavy metals or 5d oxides with strong spin-orbit coupling. Because the patterning is created by membrane-transfer epitaxy, the method is proposed as scalable to silicon and flexible substrates; epitaxial LCO$1\,\mathrm{V}$0 growth is directly demonstrated on FS-STO membranes transferred to silicon, and compatibility with glass and sapphire is also noted (Chen et al., 2023).

5. Adjacent magnetic-memory architectures and why they are not page memory

Several related devices illuminate what magnetic page memory is not. The function-switchable magnetic domain-wall memory based on an asymmetric Pt/Co/Ru/AlOx heterostructure stores information in the magnetization state on the two sides of a domain wall and uses spin–orbit torque for current-driven switching, but the source explicitly states that it is not page memory in the conventional sense because it does not describe a page-addressable storage format (2207.14427). Its key contribution is bit-level secure magnetic memory: small alternating pulses such as $1\,\mathrm{V}$1 support rewritable operation, a sufficiently large pulse such as $1\,\mathrm{V}$2 annihilates the domain wall and converts the cell into a read-only state, the rewritable-to-read-only switching probability is reported as $1\,\mathrm{V}$3 below $1\,\mathrm{V}$4 and $1\,\mathrm{V}$5 at $1\,\mathrm{V}$6, and the function can later be recovered with an in-plane magnetic field plus current. The devices were integrated into wafer-scale arrays on $1\,\mathrm{V}$7-inch Si/SiO$1\,\mathrm{V}$8 substrates, including a $1\,\mathrm{V}$9 array (2207.14427).

The split-cell extraordinary-Hall memory is closer to page-memory goals in density, but not in transport geometry. In “Sixteen-state magnetic memory based on the extraordinary Hall effect,” a logical cell is composed of several independently switchable ferromagnetic dots or multilayer elements with perpendicular anisotropy, and the combined remanent state is read electrically by the extraordinary Hall effect (Segal et al., 2011). The paper demonstrates four-, eight-, and sixteen-state cells. Its relevance is that it raises the number of stored states per physical cell without shrinking the dot size, which is directly useful for higher-density magnetic storage. The source also notes that placing dots of the same cell at separate locations and different heights suggests an effective three-dimensional memory architecture. Even so, the scheme remains a multibit cell design rather than a page-transport architecture (Segal et al., 2011).

These examples clarify a recurring misconception: dense arrays, multibit cells, or secure write-protected bits do not by themselves constitute magnetic page memory. Page memory is defined by page-level organization and controlled movement or replication of magnetic information, not merely by nonvolatility or areal-density enhancement. This distinction is explicit in the source material and is important for comparing proposals across spintronic subfields (2207.14427, Segal et al., 2011).

6. Distinct mechanical usage, common design goals, and open limits

A distinct usage of the phrase appears in “Magnetic poles enabled kirigami meta-structure for stable mechanical memory storage with high information density,” where the memory is mechanical rather than spintronic (Xin et al., 2023). There, a 3D-printable cylindrical kirigami module stores information as a sequence of stable buckled states, and oppositely oriented magnetic N–S poles stabilize the deformed configurations against crushing, impact, and shaking. The architecture scales to an $1\,\mathrm{ns}$0 module array, with a demonstrated $1\,\mathrm{ns}$1 prototype. For a $1\,\mathrm{ns}$2-type nine-layer module, the number of potential patterns is reported to increase to about $1\,\mathrm{ns}$3, and the second stable state has stiffness on the order of $1\,\mathrm{ns}$4, about $1\,\mathrm{ns}$5 times larger than in a cited prior stable-memory mechanical metamaterial (Xin et al., 2023). This is a page memory in a mechanical sense, but it is not a magnetic-domain memory.

Even across these distinct meanings, several design objectives recur. The stacked-nanowire and bubblecade papers both target ultra-dense, reliable, low-power memory by reducing the need for global high-field switching or large current injection (Ozatay et al., 2018, Moon et al., 2014). The cobaltite nanoisland work pursues minimal inter-bit crosstalk and nanometer-scale periodicity through atomically sharp phase boundaries rather than etching (Chen et al., 2023). The split-cell Hall-memory work pursues more states per cell without forcing one-bit-per-dot scaling (Segal et al., 2011). In each case, the central challenge is the same: increasing storage density while preserving state stability, selective addressability, and readout contrast.

Open limits are also explicit in the sources. In the cobaltite patterned medium, the exact lower limit of island size is not yet known because it depends on growth and nanofabrication control (Chen et al., 2023). In bubblecade memory, writing and reading were demonstrated with external methods, and integrated implementations remain future work (Moon et al., 2014). In the stacked-nanowire page memory, reliable operation depends on balancing magnetic coupling with thermal isolation through the spacer and on maintaining well-defined pinning at constrictions (Ozatay et al., 2018). A plausible synthesis is that the field is converging on multiple non-exclusive routes to page-like magnetic storage: true three-dimensional page replication, two-dimensional coherent page transport, and materials-intrinsic bit patterning that could support page-addressable crossbar schemes (Ozatay et al., 2018, Moon et al., 2014, Chen et al., 2023).