The sharpness of the transition is what makes this experiment work. (d7-DMe-DCNQI)2Cu is metallic above 79 K and insulating below 78 K [9], so the entire metal-insulator transition sits inside roughly a one-kelvin window [18]. That matters for a thermal question. If a transition is broad, a temperature gradient inside a current-carrying sample smears the phase boundary into a gradient of mixed states and you cannot say where the metal ends. With a one-kelvin window, a small internal temperature difference produces a boundary sharp enough to reason about. Broad transitions are one of the two things the authors say obscured the nonlinear thermal response in inorganic thin films [5].
The current dependence is where the stabilization shows itself. At 0.3 and 0.5 mA, resistance climbed gradually as ambient temperature fell below the transition, and did not lock in; at 2.0 mA it held [12][13]. That is four times the 0.5 mA current [19], and since Joule power goes as the square of current at fixed resistance, roughly sixteen times the heat delivered into the same crystal [20]. The switched state marks a threshold in self-heating rather than in voltage, which is exactly the coupling that was hard to isolate before [4].
The proton NMR does two jobs. Relaxation curves identify what phases are present, and signal intensity acts as a thermometer for the sample itself rather than the gas around it. Under equilibrium conditions the two temperatures agreed, as they should, and under 2.0 mA above the transition they still agreed [15]. That agreement is the control: it establishes the thermometer reads true when there is no self-heating to find, so a divergence below the transition is a measurement of the crystal warming itself, not an artifact.
What this leaves open is how any of it behaves in a device. 79 K is about minus 194 C [21], and the geometry deliberately removes the thermal contact a real memory cell is built to have, suspending the needle in a Teflon tube where heat can only escape through helium gas and the two gold wires [10]. Itou frames the goal as a microscopic basis for the thermal self-organization linking phase coexistence, heat flow and electrical transport [17], and that is the right size of claim. It is a mechanism result in a model material, published in Physical Review Applied [6], with no claim made beyond that.
One honest caveat about the reporting: the account available here breaks off mid-sentence at the moment it begins describing the NMR intensity below the transition under 2.0 mA [22], which is the number the whole argument points toward. The coexistence of metallic and insulating phases in the intermediate-resistance state is stated plainly [14]; the quantitative gap between sample and ambient temperature is not, in the text I have. Anyone building a thermal design rule for volatile switching in memory or neuromorphic hardware [3] will want that figure from the paper rather than from a summary.
