Grown-Junction Transistor

A grown-junction transistor is a bipolar junction transistor whose p-n junctions are formed inside a single crystal while it is being pulled from a molten semiconductor, by changing which impurities are in the melt as growth proceeds. Morgan Sparks and Gordon Teal made the first working examples at Bell Labs in 1950, and the idea turned Shockley's 1948 junction theory into a physical device: a bar of germanium with an n-type layer, a very thin p-type layer and another n-type layer along its length.

The method begins with a crystal puller and a crucible of germanium doped to one type. A small seed crystal is lowered to touch the molten germanium and then drawn upward slowly while rotating, so that the melt freezes onto the seed as one continuous crystal. The melt initially contains a donor impurity such as antimony, so the first part of the crystal grows n-type. When the crystal has reached a certain length, a small pellet of an acceptor such as gallium is dropped into the melt in enough quantity to outweigh the donor and make the newly freezing material p-type. After a short stretch of growth, a second addition of donor tips the balance back, and the crystal continues n-type. The result is a long ingot in which a thin p-type slab, the future base, lies between two n-type regions, with the two junctions marked by the points where each addition was made.

Each step after the pull is mechanical and unforgiving. The ingot is sliced, then cut into small bars with the layers running across them, and the position of the p layer is located, often by electrical probing. Fine wires are attached: one to each end region as emitter and collector contacts and a third to the thin p-type middle as the base contact, a delicate job since the layer is a small fraction of a millimetre thick. The bar is then etched, cleaned and sealed. Western Electric put these devices into production in the early 1950s, and the technique was passed on to licensees at the April 1952 symposium, one reason early commercial junction transistors from several firms look alike.

The limits of the method are easier to list than to overcome. The base layer was set by how fast the dopants could be changed and how the crystal froze, so it could not easily be made thinner than a few thousandths of an inch, and a thick base means carriers take longer to cross it, so gain falls and the usable frequency stays low, typically a small number of megahertz at most. Yield was low because only a short section of each ingot contained the thin p layer in the right place, and slicing losses wasted much of the expensive germanium as sawdust and rejected bars. Uniformity along the crystal depended on temperature control and stirring, and each ingot differed slightly from the last. Germanium's sensitivity to heat also limited the operating range.

Several misconceptions cling to the method. The first is that the p-n junctions were made by diffusing impurities into the crystal; in a grown junction the dopant is incorporated at the moment the crystal freezes, whereas diffusion, which came later and became central to the planar process, moves impurities into an existing solid. The second is that the technique was abandoned because it did not work; it worked well enough to put junction transistors into the market, and the 1950 demonstration was the first reproducible proof of the transistor action that Shockley had predicted. It was displaced because other methods gave thinner bases, cheaper manufacture and higher frequencies, notably alloy-junction devices, rate-grown variants, surface-barrier types and later diffused silicon.

Its heirs are still around, though not in transistor form. Modern silicon ingots are pulled by the same Czochralski method that Teal and Little adapted for germanium, but they are doped uniformly, and junctions are formed later on the wafer by lithography and ion implantation. The grown junction survives as a lesson: that a device structure can be written into a crystal by controlling what goes in as it forms, an idea that reappears in epitaxial growth and in the layered semiconductors of lasers and LEDs.

Read the grown-junction transistor as a transistor recorded in a crystal like tree rings: every change in the melt during the pull became a layer in the solid, which proved that junction transistors could be built reproducibly, even though its thick base and heavy waste soon made other methods more attractive.