Solid-State Diffusion
Solid-state diffusion is the slow migration of atoms through the interior of a crystalline solid, driven by differences in concentration and made much faster by heat. In semiconductor manufacturing it is the classic way of introducing dopant atoms such as boron or phosphorus into silicon: the wafer is held in a furnace at about 900 to 1100 degrees Celsius while a dopant source provides atoms at the surface, which then creep inward to form a layer of opposite type and, where it meets the original material, a p-n junction. This entry concerns dopant diffusion in solids as a device-fabrication tool, as distinct from the general physical idea of diffusion in gases and liquids.
Fick's laws describe the process in simple terms: atoms flow from where they are crowded to where they are scarce, and the flow is proportional to how steep the concentration gradient is. The first law says the flux of dopant atoms equals the diffusivity multiplied by the concentration gradient; the second law, which follows from conserving atoms, says the concentration at a point changes at a rate set by how the gradient itself changes in space. The practical consequence is that a profile spreads out roughly in proportion to the square root of the diffusivity multiplied by time, so doubling the depth takes about four times as long. Calvin Fuller and others at Bell Labs in the early 1950s showed that donor and acceptor impurities could be diffused into silicon and germanium in this way, and the Bell diffused-base transistors and Chapin, Fuller and Pearson's 1954 solar cell, with its thin diffused junction layer, were products of that work.
Temperature is the master control, because diffusivity rises exponentially with it. Atoms move by hopping between lattice positions, or with the help of vacancies and interstitial atoms, and each hop requires overcoming an energy barrier, so diffusivity follows an Arrhenius form: a constant multiplied by the exponential of minus an activation energy divided by Boltzmann's constant times absolute temperature. A rise of only a hundred degrees can multiply the diffusivity several times over, which is why furnaces are held to within a degree or so across the load. Different dopants diffuse at quite different rates: boron and phosphorus are relatively fast, arsenic and antimony slower, so a slow n-type dopant can sit in a buried layer while later high-temperature steps barely move it. At the temperatures involved, a fraction of a micrometre to a few micrometres is a typical junction depth reached in an hour or so.
The two idealized cases correspond to the two stages of the standard recipe. In a constant-source diffusion, the surface concentration is held at a fixed level, often near the solid solubility limit of the dopant in silicon, and the profile has the shape of a complementary error function that deepens over time. In a limited-source diffusion, a fixed amount of dopant is already in the crystal and nothing more is supplied, so the profile spreads into a Gaussian whose peak falls as it widens. Real processes combine them: a short predeposition fixes the dose at the surface, then a longer drive-in, often in an oxidizing atmosphere, pushes the dopant deeper. The junction depth is the point at which the diffused concentration falls to equal the background doping of the starting wafer, and controlling it to a small fraction of a micrometre was essential to making thin transistor bases with useful current gain.
Diffusion made modern transistors possible, but it has real limits, which is why ion implantation largely replaced it for doping. Its great strengths were batch processing, the ability to form junctions at controlled, buried depths, and compatibility with the oxide masking discovered at Bell Labs in 1957. Its weaknesses are that the profile shape is dictated by physics rather than chosen, that lateral spreading under mask edges blurs fine features, and that high-temperature steps move earlier dopants unintentionally. Ion implantation, in which dopant ions are accelerated into the wafer, was patented by Shockley in 1954 and became a production method in the 1970s, offering precise dose and depth control at lower temperature. Diffusion has not disappeared: implants still require a diffusion-like anneal to repair damage and activate dopants, thermal budgets are still tracked, and solar cells and some power devices still use diffused junctions. A common misconception is that diffusion simply means dopants flow downhill like a liquid, when in solids it proceeds by atoms hopping via defects, and the behaviour can therefore change at high concentrations or when the lattice is disturbed by oxidation or implant damage.
Read solid-state diffusion as the furnace-driven spreading of dopant atoms through a crystal, where time and temperature are traded against one another to place a junction at exactly the depth a device needs.