Pairing two NV splittings cuts a diamond clock's temperature drift tenfold

The two headline numbers sit at different averaging times, so lining them up shows what the compensation actually buys. The tenfold reduction in temperature-induced drift is quoted at 1,000 seconds of integration ^[2]. The stability figures come from a separate ten-day run ^[7]: below 5e-9 at 200 seconds of averaging, below 1e-8 at 200,000 seconds ^[8]. A reference limited only by white noise improves as the square root of averaging time, so the 200-second value projects to roughly 1.6e-10 after a thousandfold longer average, and the measured long-term number is about sixty times worse than that projection ^[3]. The composite reference is still drift-limited at long times; compensation moved the floor rather than removing it.

Temperature was disqualifying rather than merely inconvenient because of how the sensitivity scales. At 2.87 GHz, 25.3 ppb/mK works out to about 73 hertz of shift per millikelvin ^[1], and the authors describe the zero-field electron splitting on its own as unsuitable for clock use ^[4]. To keep temperature alone from pushing the fractional frequency past the 1e-8 level the finished clock reaches at long times, the crystal would have to be held within about 0.4 millikelvin ^[2]. That is achievable on an optical table with effort, but it is not a spec you would write for a vehicle or an equipment shed.

The second handle is nuclear rather than electronic. The 14N quadrupole splitting sits some 580 times lower in frequency than the electron spin splitting ^[5], and both were prepared and measured on the same ensemble by optical pumping and fluorescence readout, in a 475 gauss field, or 47.5 millitesla, aligned along the NV axis ^[6][4]. Compensation of this kind only works if the two splittings respond to temperature in different proportions, so that a weighted sum of them is largely blind to it; that sum is what the group calls a temperature-compensated composite frequency reference ^[2]. What the published summary does not give is Q's own temperature coefficient, and that is the number setting how much of the 25.3 ppb/mK actually cancels, and how precisely the two readouts must stay weighted for it to keep cancelling.

A rubidium vapor-cell clock ran alongside the prototype for the ten days ^[7], but no stability figures for it appear in the account ^[6], so nothing here ranks the diamond against the reference it was measured against. Read the result as evidence that a solid-state ensemble can be made insensitive to its own temperature, not as a candidate swap. The platform's argument rests on eliminating the trap or vapor cell rather than on short-term stability: it promises a device that can be fabricated and integrated with modern electronics, with emitter counts that could in principle approach Avogadro's number in a kilogram-scale crystal ^[9].