Mechanical Racetrack Memory

• Mechanical racetrack memory is a system that encodes digital information in buckling domains of elastic metamaterials and transports them via controlled domain-wall motion.
• It utilizes cyclic modulation of beam stiffness and asymmetric preloads to achieve a quantized shift of exactly one cell per cycle through topological boundary-mode instabilities.
• The platform offers tunable logic functionalities and low-energy operation by leveraging negative couplings and precise control of cyclic loading parameters.

Mechanical racetrack memory is a racetrack architecture in which digital information is encoded in spatial domains of an elastic metamaterial and transported by controlled motion of domain walls rather than by spin torques. In the experimentally realized topological boundary ratchet, information is encoded in buckling domains and transported in a quantized manner via cyclic loading. The implementation uses a patterned chain of bistable and monostable beams whose interfaces host topological boundary modes; cyclic loading renders these modes unstable through inter-domain pressure, which drives domain-wall motion. The transport is rooted in topological boundary-mode instabilities rather than adiabatic Chern-number pumping, yet over one full cycle the soliton shifts exactly one unit cell, making the approach a pathway toward racetrack memories in neutral systems (Omidvar et al., 1 Sep 2025).

1. Elastic-metamaterial realization and state encoding

The unit cell contains four beams, with two “main” bistable sites and two “coupling” monostable sites coupled in series. The local stiffnesses are modulated cyclically to realize a four-site pump. The elastic energy is written as

$$V \;=\;\sum_{n,j}\Bigl[\tfrac{\lambda}{4}\,q_{n,j}^4+\tfrac12\bigl(\omega_0^2+a_{0,j}+a_{1,j}(\theta)\bigr)\,q_{n,j}^2\Bigr] \;+\;\sum_{\langle\ell,m\rangle}\tfrac{c}{2}\,(q_\ell-q_m)^2\,,$$

where $q_{n,j}$ is the lateral displacement of site $j\in\{1,2,3,4\}$ in cell $n$, $\lambda$ is the Duffing nonlinearity, $\omega_0$ is the bare oscillation frequency, and $c$ is the nearest-neighbor coupling. The static terms

$$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$

encode the DC compression $\beta_{1,2}$, while the AC modulation is

$$a_{1,1}(\theta)=-a_{1,3}(\theta)=\gamma\,\alpha\sin\theta,\quad a_{1,2}(\theta)=-a_{1,4}(\theta)=-\gamma\,\alpha\sin\theta\,.$$

The information-bearing states are buckling domains. Each strongly compressed “main” beam sees an effective negative quadratic coefficient

$q_{n,j}$0

together with a positive quartic $q_{n,j}$1, producing a double-well potential

$q_{n,j}$2

These are the two stable buckled states, with $q_{n,j}$3 experimentally. A “soliton” is a “1”-domain between two “0”-domains, i.e. two domain walls. This multistable encoding supplies the non-volatile degree of freedom, while the controlled displacement of the walls supplies the racetrack functionality (Omidvar et al., 1 Sep 2025).

2. Topological structure of the pump and the status of quantization

If $q_{n,j}$4 are treated as independent pump parameters and a closed loop is traced around the origin in the $q_{n,j}$5–$q_{n,j}$6 plane, the Bogoliubov bands of small fluctuations acquire a quantized Chern number

$q_{n,j}$7

where $q_{n,j}$8 is the Berry connection of band $q_{n,j}$9. In that two-parameter picture, the pump displaces linear wave-excitations by one cell per cycle as a Thouless pump.

The experimentally relevant sweep is different. In the single-parameter sweep $j\in\{1,2,3,4\}$0 along a line in parameter space, the bulk gaps collapse at one point, so $j\in\{1,2,3,4\}$1 is ill-defined. Nonetheless, strongly localized boundary modes persist at each domain wall for most of the cycle. This distinction is central: the observed quantized transport does not rely on a fully gapped two-parameter bulk pump during the experimental sweep. A common misconception is therefore to identify the device with a conventional adiabatic Chern pump. The more precise statement is that neighboring domains act as different topological pumps for their Bogoliubov excitations, so their interface hosts topological boundary modes, and cyclic loading uses the instability of those boundary modes to ratchet the wall (Omidvar et al., 1 Sep 2025).

3. Equations of motion and the boundary-ratchet mechanism

The Newton equations derived from the elastic potential are

$j\in\{1,2,3,4\}$2

Linearization about the steady state gives the boundary-mode problem

$j\in\{1,2,3,4\}$3

At a domain wall, the neighboring domains exert inter-domain pressure because the adjacent domain has its own steady-state displacement $j\in\{1,2,3,4\}$4. To leading order, the boundary-mode frequency is renormalized to

$j\in\{1,2,3,4\}$5

where $j\in\{1,2,3,4\}$6 is the mode-shape amplitude at the wall.

As $j\in\{1,2,3,4\}$7 varies, $j\in\{1,2,3,4\}$8 crosses through zero, the frequency becomes imaginary, and the mode becomes unstable. The system then snaps, and the wall hops by exactly one cell, defined here as one main beam plus one coupling beam. Encountering that instability enforces a $j\in\{1,2,3,4\}$9-advance of the pump phase, $n$0. As a result, each wall ratchets forward by one site per half-cycle, and over one full cycle of $n$1 the soliton shifts exactly one unit cell in a quantized manner. The critical Euler buckling load of a slender beam,

$n$2

enters the continuum-to-lattice mapping through $n$3 with $n$4, thereby connecting the lattice description to beam mechanics (Omidvar et al., 1 Sep 2025).

4. Control parameters: direction, speed, and operating window

Directionality is set by the static compressions. Experimentally one chooses $n$5 so that the “main” sites are bistable. Swapping $n$6 and $n$7 reverses the racetrack direction because it mirrors the unit cell through $n$8. This makes the direction of information propagation controllable through adjustable mechanical constraints on the buckling beams.

The AC amplitude $n$9 determines whether the trajectory in parameter space passes through the instability lobes. Too small $\lambda$0 gives no hopping; too large $\lambda$1 gives multiple hops per cycle. Within the intended operating regime, the propagation speed is one cell per cycle of $\lambda$2, so driving $\lambda$3 at frequency $\lambda$4 yields

$\lambda$5

with $\lambda$6 the cell length. In the macroscale demonstration, adiabatic operation requires $\lambda$7 below the first linear eigenfrequency, approximately $\lambda$8 Hz. At microscale, $\lambda$9 can be kHz–MHz. Local bias forces, including tilted end-supports and micrometer-adjusted preload, can tune each beam’s preferred buckling direction and offset systematic setup asymmetries. These parameters collectively define a narrow but explicit design window in which the transport remains quantized and single-step (Omidvar et al., 1 Sep 2025).

5. Logic functionality in branched racetrack networks

Branching the racetrack network and introducing negative couplings between selected sites yields logic gates. A minimal NAND uses two input racetracks, $\omega_0$0 and $\omega_0$1, meeting the first site of an output racetrack through a negative spring of strength $\omega_0$2, together with a small positive bias $\omega_0$3 on the output site. The low-order potential is

$\omega_0$4

Here $\omega_0$5, and $\omega_0$6 is reduced by a factor $\omega_0$7 so that the output is “softer.” Choosing $\omega_0$8 makes the output prefer $\omega_0$9 (“1”) unless both inputs buckle to $c$0, and that competition yields the NAND truth table.

More complex networks, including buffers, inverters, and half-adders, are realized by tapering stiffness along a track and arranging branch couplings. The underlying tight-binding structure with low-order nonlinearities is presented as a general pathway toward racetrack memories in neutral systems. This suggests that the racetrack is not limited to storage and shift-register behavior; it can also serve as a substrate for domain-wall logic in which transport and gate operation are implemented within the same buckling-based architecture (Omidvar et al., 1 Sep 2025).

6. Relation to magnetic racetrack memory, quantitative contrasts, and limitations

Mechanical racetrack memory should be distinguished from magnetic racetrack memory at the level of both actuation and failure modes. In magnetic systems, ferromagnetic domain walls are an essential ingredient for racetrack memory, and their motion can be driven by magnetic field, spin-transfer torque, and spin-orbit torque. Recent work on spin inertia shows that inertial dynamics of the individual magnetic moments induce massive dynamics of the domain wall; in the absence of Gilbert damping the domain-wall dynamics become chaotic, while for finite damping field-like driving can significantly increase the wall velocity compared to conventional massless dynamics (Bassant et al., 11 Mar 2026). By contrast, the mechanical platform uses topological ratchet instabilities of boundary modes to push buckled-beam domain walls, and it is explicitly framed as a route to racetrack memories in neutral systems (Omidvar et al., 1 Sep 2025).

Aspect	Magnetic racetrack	Mechanical racetrack
Operating principle	spin-transfer torque or spin-orbit torques push domain walls along a nanowire	topological ratchet instabilities of boundary modes push buckled-beam domain walls
Scalability	nanowires are $c$1 nm thick, domain-wall widths $c$2 nm, cell sizes $c$3 nm	planar MEMS fabrication; cells can be $c$4 nm if designed for high stiffness
Energy cost	$c$5 J/bit; current densities $c$6 A/m$c$7	at microscale, $c$8 J/bit from the buckling estimate
Robustness	stable against mechanical shock but sensitive to stray fields and Joule heating	immune to electromagnetic noise, but viscoelastic damping and fabrication inhomogeneity must be controlled
Read/write	tunnel magnetoresistance; spin-torque pulses	optical readout or integrated capacitive/piezo sensors; global cyclic AC compression or local piezo actuators

For the mechanical device, the energy per hop is estimated as the area under the $c$9 curve,

$$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$0

Using $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$1 N/m and $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$2 m at microscale gives $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$3 J/bit, approaching the thermal limit $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$4 J. The principal limitations stated for the mechanical platform are that viscoelastic damping and fabrication inhomogeneity must be controlled; low-loss silicones or single-crystal flexures can push $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$5-factors above $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$6. A second misconception is that topological language by itself guarantees arbitrary drive amplitudes or frequencies. The reported operating regime is stricter: the AC modulation must cross the instability lobes, yet remain below the regime where multiple hops per cycle occur, and adiabaticity constrains the drive frequency relative to the first linear eigenfrequency (Omidvar et al., 1 Sep 2025).

7. Significance of the mechanical approach

The mechanical implementation combines a patterned chain of bistable buckling beams, a cyclic global actuator, and carefully chosen static preloads $$a_{0,1}=a_{0,3}=+\gamma\beta_1,\quad a_{0,2}=a_{0,4}=+\gamma\beta_2$$7 together with local biases to realize an in-plane, elastically powered racetrack memory. Its central novelty is that robust transport is achieved in a neutral system by exploiting topological boundary modes and their instability under cyclic loading, rather than by applying currents or external magnetic fields to a magnetic nanowire. The transport is quantized at one unit cell per full cycle, the direction is mechanically reversible through the unit-cell asymmetry, and the same platform extends to NAND, buffers, inverters, and half-adders through branching and negative couplings (Omidvar et al., 1 Sep 2025).

A plausible implication is that mechanical racetrack memory occupies a distinct design space within domain-wall information processing. The comparison with spin-inertia-driven ferromagnetic walls highlights a sharp contrast: in magnetic racetracks, inertia can introduce high-speed motion but also chaos in the zero-damping limit, whereas in the elastic ratchet the sought effect is a controlled instability of a localized boundary mode that advances the wall by a single lattice step (Bassant et al., 11 Mar 2026). Within the scope of the reported results, the mechanical platform is therefore best understood as a domain-wall racetrack for neutral systems whose quantization comes from a boundary-ratchet mechanism, not from a fully adiabatic bulk topological pump (Omidvar et al., 1 Sep 2025).