Semiconductors, Crystal Growing and Purification

Crystals of high purity are essential to the manufacture of semiconductor devices, including integrated circuits, and much effort has gone into their refinement. The first single semiconductor crystal was drawn in 1948 and, from 1949 or 1950 onward, monocrystalline material has been used exclusively.

The conditions imposed are extremely stringent and have become progressively more so as manufacturing technology has advanced. From the time of the invention of the transistor in 1947 until the mid-1960s most were made using germanium, although since then silicon has been mainly employed. By the 1980s silicon accounted for 98 percent of all semiconductor devices sold worldwide.

The three basic stages in the production process are extraction, purification, and crystal growing.

Extraction. The extraction of metallurgical grade silicon takes place within an electrode-arc furnace. Quartzite, coal, coke, and wood chips are fed into the furnace and a series of chemical reactions take place, finally resulting in the extraction of silicon of about 98 percent purity. The major contaminant is usually boron, although carbon and other impurities are also present. Only a small fraction (about 1 percent) of silicon produced by these means is used in the semiconductor industry.

Purification. This stage is necessary in order to produce electronic grade silicon, a material polycrystalline in structure with impurity levels of only a few parts per billion. Such low impurity levels are necessary to ensure that contamination does not occur during the following crystal-growing stage.

The purification process currently used is basically as follows:
1. Pulverised metallurgical grade silicon is heated with anhydrous hydrogen chloride at 300°C, yielding trichlorosilane (a gas), together with hydrogen.

2. The trichlorosilane is cooled and becomes liquid (at 32° C).
3. Fractional distillation is employed to remove impurities present (especially phosphorous and boron).

4. The reaction is reversed by passing a mixture of trichlorosilane and hydrogen over a resistance-heated silicon rod at a temperature of between 1000 and 1200°C.
5. Small crystals of electronic grade silicon are deposited on the rod during the reaction. Pure polycrystalline rods so formed are a few meters long and several inches in diameter.

Due to the complexity and cost of the technology, it had become limited to the U.S., Japan and West Germany well before the end of the twentieth century.

An earlier method of material purification, zone refining, was first described by the Russian physicist Petr Kapitza in 1928 and perfected at Bell Laboratories by William G. Pfann in 1951. The impure silicon or germanium polycrystalline rod is supported in a quartz tube in an inert ambient, and a small zone melted by radiofrequency heating. The impurities then collect at the zone edge where the temperature of the melt is lowest.

When the zone is gradually swept from one end of the rod to the other the impurities are swept along the rod and collected at the end. This effect occurs because most impurities have a lower melting point than silicon or germanium, which therefore crystallize out of the melt before the impurities. Although a great advance upon earlier methods, zone refining has the disadvantage that only relatively small amounts of material can be purified and the removal of boron in particular is time consuming.

Crystal Growing. First, it is necessary to ensure that the material is as dislocation-free as possible, since crystal defects affect the rate at which impurities diffuse. Dopant atoms (introduced in highly accurate amounts during subsequent production processes) travel much more quickly at the surface of grain boundaries due to crystal dislocations. Dopant atoms introduced into material with an unacceptable level of dislocations would result in a highly uneven diffusion profile, and consequent device failure. It is also important to control the crystal orientation.

This is because: (1) crystals cleave easily in certain directions; (2) during device manufacture, it is invariably necessary to put down uniform layers of dopants and oxides; and (3) chemical etching is frequently used to cut windows in oxides grown on the surface of wafers. Etching will only proceed uniformly if the crystal surface presents the correct orientation.

The favored crystal-growing technique is the Czochralski process, developed by Jan Czochralski in 1918, but perfected for growing germanium and silicon single crystals by Gordon K. Teal and John E. Little at Bell Laboratories in 1950. By the end of the twentieth century it accounted for 70 to 80 percent of production. Three other methods of crystal growing are also used, float-zone, Bridgman, and epitaxy. Each has its advantages and disadvantages.

In the case of the Czochralski method, the crystal is ‘‘pulled’’ from molten material (melt), which is contained within a crucible. There is therefore a possibility of material contamination. With germanium (melting point 960°C) it is possible to use a graphite crucible. However, this cannot be done with silicon (with a melting point of 1420°C) and therefore quartz is used. The process takes place within an enclosed inert ambient (helium or argon).

The temperature of the melt is controlled externally by radio-frequency heating. The seed crystal, mounted in the desired orientation, is held in a chuck fixed at the end of a shaft, which is driven by a motor mounted externally above the enclosure. The crystal is then lowered to make contact with the melt, which is set at about 15°C above its melting point (about 1435^°C).

The crystal itself will not melt because of its higher resistivity. The seed crystal is then slowly raised and rotated, causing the molten silicon to freeze onto the seed crystal with the same orientation as the seed itself. Finally, the fully grown crystal is cooled to about 300^°C before being exposed to the external atmosphere.

The float-zone method developed by P.H. Keck in 1953 overcomes the difficulty of crucible contamination by vertically suspending a polycrystalline silicon rod in an inert ambient. A radiofrequency coil melts a narrow cross-section of the rod above the seed crystal and is slowly raised. The silicon below the molten zone resolidifies in singlecrystal form with the same orientation as the seed crystal. The molten zone does not flow out, due to the geometry and surface tension. This process is more expensive and difficult to control than the Czochralski method.

The Bridgman method (named after Percy W. Bridgman) involves a boat containing seed and melt being slowly pulled through a horizontal tube furnace, the melt freezing in the required orientation as the crystal enters the cooler zone. Towards the latter years of the twentieth century this was still the preferred technique for growing gallium arsenide (GaAs) crystals. Single crystals of 75 millimeters diameter and low dislocation density were by then being produced on a production basis.

Conclusion. All the above processes involve crystal growth from the melt. Epitaxy is the technique of depositing single-crystal material in successive layers upon an atomically flat single crystal substrate. The substrate most commonly used is silicon, which enables many compound semiconductors to be deposited; for example, crystalline GaAs and cadmium telluride (CdTe).

During the epitaxial process the deposited material takes up the same crystalline orientation as the substrate. Fabricating transistors by impurity doping of epitaxial single crystals grown from the gas phase was first achieved at Bell Laboratories in 1960. In the following year, N. Nielson of RCA described the process of epitaxially growing GaAs and Ge crystals from the liquid phase. Molecular beam epitaxy, although originating in elementary form in the 1960s, only came into production use by the early 1980s. It is however capable of greater precision.