Drive-in

Drive-in is a thermal step that redistributes dopants already introduced into silicon, spreading them deeper and sideways to form the required electrical profile. The heat does not push atoms downward like a piston. Atomic diffusion produces a net flux along concentration gradients, with the rate controlled by temperature, dopant species, and the crystal's defect population.

A fixed dose gets wider, not larger. In the textbook two-step diffusion process, predeposition introduces dopant from a source; the subsequent drive-in redistributes that dose without continuing to supply it. If no dopant escapes or enters, the depth-integrated concentration stays constant. As the profile broadens, its surface concentration falls. This is the physical distinction between a limited-source drive-in and a constant-surface-concentration diffusion, where an external reservoir keeps feeding the wafer.

The Gaussian is a model with boundary conditions. For an initially very thin surface dose Q in a semi-infinite solid, constant diffusivity D, and no surface loss, the ideal profile is C(x,t) = Q / √(πDt) × exp[−x²/(4Dt)]. Here Q is atoms per unit area, C is atoms per unit volume, and x is depth. The integral from the surface to infinite depth is Q. This is not automatically the profile of an implanted wafer: an implant begins with a finite depth distribution, and its damage changes subsequent diffusion.

Temperature history matters more than the hold time alone. A simple equilibrium model uses D(T) = D₀ exp[−Eₐ/(kBT)], with temperature in kelvin. When D depends only on temperature, replace Dt by the accumulated diffusion budget B = ∫D[T(t)]dt. Ramps, the soak, cooldown, and later hot steps all contribute. B has units of length squared; the characteristic spreading length scales as √B. Two recipes with the same number of minutes need not produce the same profile.

Junction depth is a crossing, not the end of the tail. For donors introduced into a uniformly acceptor-doped substrate, the metallurgical junction is where the electrically active donor and acceptor concentrations balance. In a fully active ideal Gaussian with surface concentration Cₛ and background Nᵦ, xⱼ = 2√[B ln(Cₛ/Nᵦ)], provided Cₛ exceeds Nᵦ. The background therefore matters as well as the diffusion budget. The depletion region under bias is a different quantity and extends around that junction.

More diffusion does not mean a simple junction-depth multiplier. At fixed dose, increasing B also reduces Cₛ. Although the profile's width grows as √B, the junction position contains the changing logarithmic concentration ratio as well. Saying that four times the drive-in budget always doubles junction depth confuses profile width with an electrical boundary. A process model should retain dose, starting profile, background, and activation rather than tracking a single length.

The mask opening is not the final doped footprint. A surface mask can block dopant introduction outside a window, but it does not form a vertical wall inside the silicon. During drive-in, dopants spread beneath the mask edge as well as into the depth. This lateral encroachment affects spacing, junction curvature, and the remaining distance between neighboring regions. Its size is profile- and geometry-dependent; a universal lateral-to-vertical ratio is not a substitute for two-dimensional simulation.

The crystal is not a passive background. Implantation produces excess vacancies and self-interstitials. Their evolution can temporarily accelerate dopant redistribution, known as transient enhanced diffusion. Oxidation also changes point-defect populations and the moving surface boundary. Consequently, an oxidizing drive-in cannot be represented simply by adding oxide growth to an otherwise unchanged inert-ambient calculation. Dopant species, concentration, defect chemistry, and surface segregation determine the actual response; broad “enhanced” or “retarded” labels need their process conditions.

Redistribution and activation are different outcomes. Annealing can repair damage and place dopants into electrically active configurations, while clustering or precipitation can leave some atoms inactive. SIMS measures chemical concentration versus depth, not activation by itself. Electrical profiling, such as calibrated spreading-resistance measurements, provides complementary information. Sheet resistance is useful for process monitoring, but one integrated resistance value cannot uniquely establish the junction depth or profile shape.

Qualify the whole thermal sequence. Begin with the measured starting dose and profile, simulate the intended drive-in together with later thermal steps, and compare the predicted result with depth profiling and electrical tests. Check both vertical depth and lateral encroachment. Temperature calibration, wafer loading, ambient, surface films, and dose retention all belong in the qualification. The objective is a specified active profile and device behavior, not simply a completed furnace timer.

Sources: TU Wien IuE, Propaedeutic Example, Equilibrium Diffusion, Transient Enhanced Diffusion, and oxidation-enhanced/retarded diffusion models; EAG Laboratories, SIMS Services; imec, Characterization of Electrically Active Dopant Profiles with the Spreading Resistance Probe. Graphs are explanatory models, not process recipes.