Float-Zone Process
The float-zone process grows or purifies a silicon crystal without a crucible: a short molten zone, held between two solid sections of a rod purely by surface tension, is moved slowly along the rod by a radio-frequency heating coil, and the silicon refreezes behind it as a crystal. Because nothing touches the molten silicon except the surrounding gas, the crystal avoids the contamination that a crucible adds, which gives float-zone silicon its very high purity and very high resistivity. It was developed at Bell Labs in the early 1950s by Henry Theuerer, building on William Pfann's zone-melting ideas, and it is the principal alternative to the Czochralski method.
The process works by the melting-freezing cycle moving along a vertical rod, sweeping impurities in one direction. A polycrystalline silicon rod stands upright in a chamber filled with inert gas or held in vacuum, with a small seed crystal at the bottom. A single-turn induction coil heats a thin slice of the rod to the melting point of silicon, and the molten zone hangs between the rods, held by its own surface tension and the electromagnetic forces from the coil. Moving the coil upward melts new material at the top of the zone while crystal freezes at the bottom in the seed's orientation. A wider section is grown first from the seed, then the diameter is brought up, so one pass converts the rod into a single crystal and also sweeps impurities along the rod in the way Pfann had described.
The method's big advantage is purity: no crucible means no oxygen from quartz and far fewer contaminant atoms. Czochralski silicon contains a substantial oxygen content from the fused-silica crucible, whereas float-zone silicon contains very little. This allows resistivities in the thousands of ohm-centimetres and beyond, with long lifetimes for minority carriers. High-voltage power devices, such as thyristors and high-power rectifiers and transistors, need thick, lightly doped silicon to block large voltages, and detectors for particles and X-rays need low leakage and full depletion of thick silicon. Both ask for material that Czochralski growth cannot easily supply.
Doping a float-zone crystal needs special routes, because there is no melt reservoir to dissolve dopant in. One approach is to blow a dopant-bearing gas such as a phosphine or diborane mixture across the molten zone. Another, widely used for power-device silicon, is neutron transmutation doping: the rod is irradiated in a nuclear reactor, where neutrons convert a small fraction of silicon-30 atoms into phosphorus. Because the conversion is spread uniformly through the volume, the resulting n-type resistivity is more uniform than gas doping could give. The crystal must be annealed afterward to repair the radiation damage.
The cost of this purity is size and price. Surface tension must support the whole molten zone, and the larger the diameter the harder that becomes, so float-zone wafers are generally limited to smaller diameters than the largest Czochralski wafers, and the process needs more careful feedstock preparation, since the feed rod must be of suitable shape and quality. As a result, most silicon, including nearly all of the silicon used for ordinary logic and memory chips, comes from Czochralski growth, which scales to the largest diameters at lower cost. A common misconception is that float-zone silicon is simply "better": it is better for particular jobs, but Czochralski wafers have mechanical strength and gettering benefits from their oxygen that many device makers want.
The family resemblance to zone refining is worth noting. Pfann's zone refining passed a molten zone along a rod several times to concentrate impurities at one end. The float-zone method uses the same zone movement, but with the aim of producing one clean crystal rather than concentrating impurities in the tail, and with the molten zone unsupported rather than in a boat. Both rely on the fact that most impurities are more soluble in the liquid than in the solid.
Read the float-zone process as crystal growth with nothing touching the melt: a moving ring of molten silicon that leaves purity behind it, at the price of diameter and cost, and the reason power devices and detectors still choose it over the Czochralski method.