Crystal Growing

Crystal growing is the process of turning purified semiconductor material into a single, continuous crystal large enough to be sliced into wafers. For silicon, the dominant method is the Czochralski process: polycrystalline silicon is melted in a crucible just above its melting point of about 1414 degrees Celsius, a small seed crystal is dipped into the melt, and the seed is slowly pulled upward while rotating so that atoms freeze onto it in the seed's exact orientation. What emerges is a heavy cylinder called a boule, the raw material from which every wafer, and so nearly every chip, begins.

The method is named for the Polish chemist Jan Czochralski, who described it in 1916 while measuring how fast metals crystallize. He touched a capillary to molten tin and drew out a thin single-crystal thread, a study of crystallization velocity, not a manufacturing plan for electronics. The technique was dormant for decades until semiconductor research needed single crystals. At Bell Labs, Gordon Teal and John Little adapted it around 1948 and reported pulling single-crystal germanium in 1950, in a growth apparatus in which a seed was lowered into molten germanium and withdrawn. Teal's crystals mattered because the point-contact transistor of 1947 had been made on messy polycrystalline germanium; single-crystal material, and the grown pn junctions that Teal and Morgan Sparks produced in 1951, made the junction transistor of Shockley's theory practical.

Germanium came first, and the move to silicon was a second feat of crystal growing. Silicon melts hundreds of degrees hotter than germanium and in its liquid state attacks nearly every container material, so the melt had to be held in a crucible of fused quartz. Teal, who had moved to Texas Instruments, grew silicon crystals and used them for the first grown-junction silicon transistor in 1954. Later in the 1950s, W. C. Dash at General Electric introduced the narrow neck step that lets the growing crystal shed dislocations inherited from the seed, so that boules could be grown free of that defect. Together, these developments gave the industry a way to make large, dislocation-free silicon crystals on demand.

Modern Czochralski growth is a feedback-controlled balance of heat, rotation and pull rate. The crystal and the crucible rotate in opposite directions to even out temperature and dopant distribution, the pull speed, typically on the order of a millimeter per minute, together with the heater power sets the diameter, and cameras and weighing systems adjust both continuously. Boules for 300 millimeter wafers, the standard for leading-edge logic and memory, are typically a couple of meters long and weigh hundreds of kilograms, and the industry once studied 450 millimeter wafers before the cost and tool complexity of the move stalled it. Applying a magnetic field to the melt, in magnetic Czochralski growth, damps convection and lets fabs tune the oxygen content.

Oxygen is Czochralski silicon's characteristic impurity and also the reason an alternative exists. The quartz crucible slowly dissolves into the melt and feeds oxygen into the crystal at levels around 10^18 atoms per cubic centimeter; in modest amounts it strengthens the wafer and, after heat treatment, creates oxygen precipitates that trap metal contaminants, but it also complicates high-resistivity and power devices. The float-zone method, a crucible-free technique first developed for silicon at Bell Labs in the early 1950s, melts a narrow zone of a polysilicon rod and moves it along with an induction coil, giving very low oxygen and very high resistivity, though it is harder to scale beyond 200 millimeters; float-zone wafers therefore serve power semiconductors and detectors, while Czochralski supplies the bulk of digital chips and solar cells.

A frequent misconception is that crystal growing also purifies the silicon to the required level. Growth gives structure, not purity: the chemical purification happens earlier, in the Siemens process and in techniques such as zone refining, and Czochralski growth actually adds oxygen and a little carbon. What it does remarkably well is create a single crystal in which distances between atoms are the same from one end of a 300 millimeter wafer to the other, and, once dopant is added to the melt, a controlled resistivity. Dopant segregation during growth still causes the concentration to vary along the boule, so manufacturers cut it into ranges and sort wafers by resistivity.

Read crystal growing as the step that converts purified material into a single perfect lattice: Czochralski's 1916 observation, adapted by Teal and Little for germanium in 1950 and scaled to the 300 millimeter silicon boules of today, gives every chip a wafer whose atoms line up without a break, which is the precondition for everything that follows in the fab.