How Do You Make Something Purer Than Chemistry Can? Melt It Slowly, Over and Over

When a mixture freezes, the solid that forms is generally purer than the liquid it came from, because most impurities fit badly into an ordered crystal and are rejected into the remaining melt. That is the same effect that makes sea ice nearly fresh. The technique turns that one-step advantage into a repeatable process by arranging for freezing and melting to happen continuously along a bar, so the rejection occurs thousands of times over rather than once, and the impurities are pushed steadily in one direction.

The apparatus is simple and the control required is not:

• •A solid bar of the material is held in a boat or hangs free
• •A ring heater melts a narrow band across it
• •The heater travels slowly from one end to the other
• •At the trailing edge the melt freezes, rejecting impurities
• •At the leading edge fresh material melts into the band
• •Impurities ride along in the band and end up at the far end
• •The contaminated end is cut off and the pass is repeated

The method was developed by William Pfann at Bell Laboratories around 1952 and it made the transistor industry possible rather than merely assisting it. Semiconductor behaviour depends on deliberately added impurities at concentrations of a few parts in a billion, which means the starting material has to be purer than that by a wide margin or the intended additions are swamped by accidental ones. No chemical purification reaches those levels. Silicon and germanium refined this way reached purities of roughly one impurity atom in ten billion, and the technique remains in use for that and for producing reference materials where extreme purity is the whole point.

The mechanism runs both ways and the reverse is deliberately exploited. Where an impurity is wanted rather than removed, a single pass in the opposite direction can be used to distribute an added element evenly along a bar, which matters when a uniform concentration is needed and simple mixing gives a gradient. The related process of growing a single crystal from a melt by slowly withdrawing a seed relies on the same rejection of impurities at the freezing front, which is why a pulled crystal is purer than the melt it came from and why the last portion of the melt is the dirtiest.

The method has clear limits and knowing them explains why it is not used for everything. It only works where the impurity is more soluble in the liquid than in the solid, and an impurity that prefers the solid is driven the wrong way and concentrated in the purified portion. It is slow, since a pass takes hours and dozens of passes may be needed. It requires a container that does not itself contaminate the melt, which is why some materials are floated without touching anything. And it consumes material, because the dirty end is discarded each time, so the yield falls with every pass.

Freezing rejects most impurities into the remaining liquid, and moving a molten band along a bar repeats that rejection continuously, sweeping contamination to one end to be cut off. Developed at Bell Laboratories around 1952, it reached roughly one impurity atom in ten billion and made semiconductors possible. It fails for impurities that prefer the solid phase.