How fast is the martensitic transformation in Ni-Mn-Ga at sub-nanosecond timescales?

The 90 ps figure is a probe-limited upper bound

The headline number is not a stopwatch reading of a single atomic jump. The experiment pumps a Ni-Mn-Ga film with a 270 fs laser pulse and probes the crystal structure with 9 keV X-rays from a synchrotron source whose pulse lasts about 100 ps. Because the probe integrates over that window, a martensite-to-austenite change completed within roughly 90 ps can only be reported as an upper limit: the real transformation may be shorter than the instrument can resolve.

The authors are explicit about this ceiling, and they use it to compare against earlier work. A previous experiment on the same class of material, driven by a 7 ns laser pulse, showed a transformation time near 7 ns that was limited by the laser duration. Replacing the 7 ns pump with a 270 fs pump moved the observed time down to around 100 ps, which is the resolution limit of the present probe rather than a measured intrinsic constant.

The paper also places its result against the only comparable diffraction experiment at similar resolution, which reported a martensite-to-austenite transformation within 200 ps. The present observation is about twice as fast, and the authors read that gap as a sign that they are approaching an intrinsic limit rather than simply improving a setup.

Five design choices compress the full cycle to 5 ns

A transformation time is only half of what applications need; the reverse transformation must also complete. The paper reports a full martensite-to-austenite-to-martensite cycle within 5 ns, which the authors describe as the fastest value reported so far. That cycle time depends on how quickly heat can be added and removed, not only on how quickly the lattice can rearrange.

The authors list the conditions that made it possible. A 270 fs pump delivers heat faster than a 7 ns pulse could. The film is only 50 nm thick, comparable to the 62 nm extinction length of the laser, so it is heated almost uniformly, and the small thermal mass is also easy to cool. Heat leaves only through the substrate, so the team chose silicon for its high thermal conductivity instead of the MgO used in earlier thin-film experiments. A base temperature of 150 K, far below the transformation temperature, creates a large gradient that drives fast cooling. Finally, a laser fluence just sufficient to trigger the transformation limits how much heat must later be removed.

These are experimental design variables, not intrinsic material constants. The authors note that the martensite-to-austenite part is probably near its intrinsic limit, while further improvement is still needed to bring the reverse austenite-to-martensite step to the same speed.

Over-heating alone predicts the wrong trend

If the transformation were purely thermally driven, stronger over-heating above the austenite start temperature should shorten it. The 7 ns experiments showed exactly that decrease. In the 270 fs experiments, the transformation time instead increased slightly with over-heating, which contradicts a simple temperature-driven picture and points to a second influence.

A second anomaly reinforces the point. For the first four data points, the calculated over-heating is negative, meaning the lattice temperature should not reach the austenite start temperature at all, yet a complete transformation to austenite is observed at the minimum fluence of 5 mJ/cm². The authors explain this by noting that at these timescales electrons, spins and the lattice are not in equilibrium: the laser couples to electrons first, the electron temperature rises during the femtosecond pulse, and the lattice takes up to several picoseconds to catch up.

The paper treats this as a qualitative argument rather than a calibrated correction. It notes that the three-temperature picture was established in nickel and that its specific values are not directly transferable to Ni-Mn-Ga, which has soft phonons and premartensite. The claim is that the electronic and spin subsystems raise the driving force toward austenite, so austenite may already be present before the 100 ps probe can see it.

Thermal stress competes with over-heating

The paper's time-resolved strain measurement supplies the competing variable. Using an off-normal austenite reflection, the authors track in-plane and out-of-plane lattice parameters during and after the laser pulse. The in-plane parameter stays nearly constant across fluences and delays, as expected for a film clamped by a thick substrate that remains near base temperature. The out-of-plane parameter rises sharply at the pulse and then decays.

The ratio of the two parameters gives a compressive film strain. Before the pulse, the two lattice parameters differ by a constant 4.6%, attributed to biaxial strain from epitaxial growth and thermal stress during preparation. The maximum strain increase grows with fluence. At 15 mJ/cm², a simple thermal-expansion model with a bulk coefficient of 15 × 10⁻⁵ K⁻¹ predicts 0.89% strain, reasonably close to the measured 0.98%.

Converting strain to stress with a 20 GPa modulus gives about 147 MPa of in-plane compressive stress at that fluence. Using a Clausius-Clapeyron-type coefficient of 2.2 MPa/K for bulk Ni-Mn-Ga, the authors estimate that this stress should raise the transformation temperature by about 67 K, which is 22 K more than the 45 K over-heating calculated at the same fluence. In other words, stress pushes the transformation one way while over-heating pushes it the other, and the observed timing reflects both.

Large lattice distortion demands collective atomic motion

The experiment cannot resolve the atomic pathway, so the authors extend the picture with molecular dynamics simulations using machine-learned force fields trained on density functional theory calculations. These simulations reach the picosecond scale that the diffraction probe cannot resolve.

The simulations compare martensite phases with different distortion magnitudes. For the NM phase, the paper reports that the relevant atomic displacements combine after 3 ps and the transition finishes after 8 ps. For 14M, which has a much lower distortion, the transformation starts and finishes much earlier, indicating that a large tetragonal distortion delays the transformation. When one axis of the simulation box is fixed at a = 5.65 Å to mimic substrate constraint, both the lattice distortion and the transformation time become smallest.

The authors' interpretation is that the large distortion requires many atoms to displace collectively, and that coordination is the key factor limiting speed. This is a simulation-based mechanism, not a direct observation, and it is offered as the explanation for why martensitic transformations are slower than other structural transformations despite being diffusionless.

The speed limit applies to thin films under these conditions

The measurements come from one specific system: a 50 nm epitaxial Ni₅₃Mn₁₈Ga₂₆Cu₃ film grown on a silicon substrate with a chromium buffer and a strontium titanate layer, cooled to 150 K and driven by 1030 nm pulses at fluences from 5 to 15 mJ/cm². The 5 ns cycle and the 90 ps upper bound belong to that configuration, and the authors' own comparison shows how strongly the numbers depend on pump duration, film thickness and substrate choice.

The paper's applied argument follows from that dependence. The authors expect the largest benefit in miniaturized microsystems, which can be heated and cooled quickly; their reduced size lowers total power, but a higher cycle frequency can compensate and yield high power densities from a small amount of martensitic material. For room-temperature operation, they suggest raising the transformation temperature rather than changing the cooling scheme.

The most useful distinction for a reader is therefore not a single speed record but a split between what is intrinsic and what is engineered. The collective atomic motion tied to lattice distortion sets the material-side limit, while pump duration, film thickness, substrate conductivity, base temperature and fluence set how close a device gets to it. The paper's own priority is to close the remaining gap on the reverse austenite-to-martensite step, since the forward step is already near the probe's resolution.

Sources used in this answer