The signal in most magnetic resonance measurements comes from polarizing the nuclear spins in a sample with a magnetic field. However, for nanoscale samples, the imperfect cancellation of up and down spins — or 'statistical' polarization — can dominate the usual polarization.

Rather than run from this statistical noise, a group from IBM Almaden Research Center and Stanford University, both in the US, is trying to harness it to image spins in small samples. The team, including John Mamin and Dan Rugar, are able to image the statistical polarization arising from 19F nuclear spins in samples of calcium fluoride with 100 nm resolution using magnetic resonance force microscopy1. To do so, however, they first had to eliminate other sources of noise that emerge in the measurement of small volumes of spins.

The key to doing this turns out to be sampling a large number of independent spin configurations in a single measurement. The team therefore applies a radio-frequency pulse that randomizes the orientation of the spins. This strategy allows the group to improve the (statistical) signal-to-noise ratio by a factor of six and increase the imaging speed by almost a factor of forty.