Diffusion Coefficient

The diffusion coefficient, D, describes how readily atoms redistribute through a material under specified conditions. In a simple diffusion model it connects a concentration gradient to a net atomic flux. It is not an atom's velocity, and it is not a junction depth. Its units are area per time, commonly cm²/s or m²/s.

For semiconductor dopants, a useful value of D must identify the dopant, host material, temperature, and model assumptions. Concentration, point defects, and process history can also matter. A number copied from a table without that context can be precise-looking but inappropriate for the wafer being modeled.

A gradient creates net transport. For one-dimensional diffusion with a scalar diffusivity, Fick's first law is J = −D ∂C/∂x. Here C is concentration and J is the number of atoms crossing unit area per unit time. The minus sign means that the simple diffusive contribution points down the concentration gradient. Atoms still move randomly in both directions; the imbalance produces the net flux.

With C in atoms/cm³ and a gradient in atoms/cm⁴, multiplying by D in cm²/s gives J in atoms/(cm²·s). At a uniform concentration, this simple model gives zero net flux even though individual atoms continue to move. In more general systems, chemical-potential gradients and electrical forces require a more complete transport description.

Combining flux with conservation gives ∂C/∂t = ∂/∂x(D ∂C/∂x). Only when D is spatially constant does this reduce to the familiar D ∂²C/∂x² form. This distinction matters when diffusivity depends on the local concentration or defect population.

Diffusion spreads a profile; it does not move every atom the same distance. For constant D in an unbounded one-dimensional medium, the variance of a Gaussian distribution grows by 2Dt. Its standard deviation therefore obeys σ² = σ₀² + 2Dt. For a negligibly narrow starting distribution, the characteristic width grows as the square root of time, not linearly with time.

For that narrow-source approximation, four times the diffusion time gives twice the width. A real initial implant profile already has a finite spread, so its total width does not generally double when time is quadrupled. A maintained surface concentration also has a different solution from a fixed total dose. The coefficient alone cannot specify a profile without initial and boundary conditions.

Temperature enters exponentially in an Arrhenius model. A common approximation is D = D₀ exp(−Eₐ/kBT), where D₀ is a prefactor, Eₐ an activation energy, kB Boltzmann's constant, and T absolute temperature in kelvin. Use consistent energy units for Eₐ and kB. A plot of ln D against 1/T is a straight line only over a range where one effective Arrhenius description is appropriate.

There is no universal multiplier for a 25 °C temperature increase. For fixed D₀ and Eₐ, the ratio between two temperatures is exp[(Eₐ/kB)(1/T₁ − 1/T₂)]. It depends on both temperatures and the activation energy. Likewise, a fixed ranking of dopants from “fastest” to “slowest” can be misleading when concentration, defects, or temperature regimes change.

Point defects help determine the effective diffusivity. Dopant motion in silicon can involve interactions with vacancies and self-interstitials, including mobile dopant-defect pairs. The relative contributions depend on species and conditions. Implantation or other processing can disturb the equilibrium defect population, producing transient diffusion behavior that a temperature-only coefficient does not capture.

High doping can also alter charge-state populations and transport. Clustering and electrical activation introduce further distinctions: total chemical dopant concentration is not necessarily the electrically active concentration. A broader chemical profile and a particular electrical junction depth are therefore not interchangeable measurements.

Account for the entire relevant thermal history. In the restricted model where D varies only with time through temperature and is uniform in space, the variance increment becomes 2∫D[T(t)]dt. Heating ramps, holds, and cooling can all contribute. This integral is a useful simplified thermal-budget measure, but it does not replace coupled defect and dopant modeling when diffusivity depends on local state.

A junction depth additionally depends on the dopant profile, background doping, activation, and the definition used for the junction. It is not simply √Dt. Boundaries between silicon, oxides, and other materials can introduce segregation and different transport mechanisms; one material's coefficient cannot be reused indiscriminately in another.

The key idea: D is a transport parameter inside a model, not a complete process prediction. State its units, material system, temperature range, and assumptions; then combine it with conservation, the starting profile, boundary conditions, and the relevant thermal history.

Sources: MIT OpenCourseWare solid-state diffusion and dopant-diffusion teaching materials; BYU Cleanroom diffusion equations and source-profile models. Graphs illustrate specified mathematical models, not fabrication recipes.