A diffusion simulation can look credible while being wrong in the quantity that matters most: junction depth, sheet resistance, peak concentration, or lateral encroachment beneath a gate edge. To understand how to simulate dopant diffusion, treat it as a calibrated process-modeling problem rather than a contour-plot exercise. The required inputs are the implant or deposited dopant profile, the thermal history, the material state, and a numerical mesh capable of resolving the gradients produced by each step.
Start with the physical question
The simulation setup should follow the engineering decision it must support. For a CMOS well drive, the concern may be final vertical junction depth and concentration under the active region. For a source/drain process, lateral diffusion and overlap with the gate can dominate device behavior. For a discrete power device, the relevant target may be a deep, lightly doped drift region with a specified resistivity profile.
These cases do not require the same domain size, mesh density, or diffusion model. Defining acceptance quantities before constructing the model prevents a common failure mode: spending time refining plots that do not answer a process or device question.
At minimum, identify the dopant species, substrate orientation and background concentration, initial profile source, anneal sequence, ambient, and temperature ramp. Also decide which outputs will be compared with measurements or process targets. Typical choices are the concentration profile at selected locations, metallurgical junction depth, sheet resistance, active dose, and lateral diffusion distance.
Build a defensible initial dopant profile
Dopant diffusion begins with the concentration distribution present before the thermal cycle. That distribution may come from ion implantation, constant-source predeposition, limited-source deposition, epitaxial growth, or an imported measured profile. Its accuracy sets the ceiling for the later diffusion result.
For implanted boron, phosphorus, arsenic, or antimony, an analytic or tabulated as-implanted profile should represent projected range, straggle, dose, tilt effects where relevant, and any masking geometry. A simple Gaussian approximation can be useful for early design studies, but it may not represent channeling tails or the near-surface behavior of a real implant. If the implant profile has been characterized experimentally, use that information instead of compensating for a poor starting profile with arbitrary diffusivity adjustments.
For furnace diffusion, distinguish between predeposition and drive-in. During a constant-source predeposition, the surface concentration can approach the solid solubility limit for the process temperature. During a limited-source drive-in, the total dose is conserved while the profile broadens and the peak concentration falls. Confusing these boundary conditions changes both the profile shape and the junction depth.
In two-dimensional process simulation, geometry belongs in the initial setup as well. Oxide thickness, openings in masks, silicon topography, and neighboring regions influence where dopants enter and how they diffuse laterally. A one-dimensional calculation remains valuable for profile calibration, but it cannot predict edge effects.
Select the diffusion physics to match the regime
The simplest diffusion model is based on Fick’s law:
[ frac{partial C}{partial t} = nabla cdot left(Dnabla Cright) ]
where (C) is dopant concentration and (D) is diffusivity. With a constant diffusivity, this model offers useful intuition and can be adequate for lightly doped, limited cases. Semiconductor process conditions frequently require more.
Diffusivity is strongly temperature dependent and is commonly represented by an Arrhenius relationship, (D = D_0exp(-E_a/kT)). A small temperature error at high anneal temperature can therefore produce a meaningful junction-depth error. Use the actual thermal schedule, including ramp-up, soak, and cool-down periods when their thermal budget is significant.
At higher concentrations, diffusion can become concentration dependent. Point defects, Fermi-level effects, dopant clustering, and transient enhanced diffusion after implantation can alter the effective rate and the electrically active dopant concentration. Boron diffusion after an implant anneal is a familiar example where interstitial-mediated transient effects can matter. Arsenic and phosphorus can require attention to clustering and activation at high concentrations.
The correct level of physics depends on the process window and the required accuracy. A compact model may be appropriate for establishing a preliminary process profile. If the result will be used to set implant energy, anneal time, or a critical device dimension, use a calibrated model that includes the mechanisms demonstrated to affect that process. More physics is not automatically better if model parameters are unavailable or cannot be validated.
Use a mesh that resolves junctions and interfaces
Numerical diffusion solvers compute concentration on a mesh, so mesh placement has direct physical consequences. Refine the mesh where concentration changes rapidly: near the silicon surface, at implant peaks, around pn junctions, beneath mask edges, and at material interfaces. The mesh can be coarser in regions where the concentration is nearly uniform and far from the area of interest.
An excessively coarse mesh smears the profile numerically, making a junction appear deeper or a peak lower than the process predicts. Excessive refinement everywhere increases solve time without adding useful accuracy. The practical test is mesh convergence: reduce local element size, rerun the case, and confirm that the reported junction depth, sheet resistance, and peak concentration change by less than the tolerance needed for the decision.
Mesh quality matters especially in two dimensions. Narrow windows, curved corners, and thin oxide layers can create poorly shaped elements or abrupt size transitions. These can reduce solver efficiency and obscure whether an observed feature is physical or numerical. Refine locally and transition gradually into the surrounding domain.
Run the thermal sequence as process steps
A diffusion model should represent the actual sequence, not a single equivalent anneal unless equivalence has been established. Oxidation may inject point defects and consume silicon. An intervening etch can alter boundary conditions. A rapid thermal anneal and a long furnace drive-in can deliver similar nominal thermal budgets while producing different activation and diffusion behavior.
Set each step with its duration, temperature, ambient, and applicable surface or interface conditions. Track dose conservation where a closed-system drive-in is expected. For constant-source diffusion, verify that the imposed surface condition is physically plausible at the selected temperature. In implanted processes, evaluate activation separately from total chemical concentration when the device simulation requires electrically active dopants.
A useful workflow is to first calibrate a one-dimensional vertical profile against a known process monitor, then transfer the validated thermal and diffusion parameters to the two-dimensional geometry. This separates core diffusion calibration from lateral effects and reduces the number of uncertain variables adjusted at once.
Validate against measurements, not only expectations
A simulation is calibrated when it reproduces independent observables within an agreed tolerance. Secondary ion mass spectrometry can provide concentration-versus-depth data, although interpretation near the surface and at low concentrations requires care. Spreading-resistance profiling, sheet-resistance measurements, electrochemical capacitance-voltage profiling, and junction-stain measurements can provide complementary constraints.
Do not calibrate only to junction depth. Different profiles can cross the background concentration at the same depth while having different peak concentration, dose, and electrical activation. A stronger comparison uses several quantities: profile shape, sheet resistance, junction position, and, where possible, lateral encroachment. If one parameter set fits a predeposition monitor but not a drive-in monitor, the issue may be the selected physics or boundary condition rather than a single diffusivity value.
Document the parameter set, source data, mesh settings, and thermal schedule used for each calibrated case. That record is what turns an individual run into a reusable process model for design teams, researchers, and future process revisions.
Choose the simulator for the dimensionality of the problem
A one-dimensional solver is efficient for rapid studies of vertical profiles and thermal-budget sensitivity. Once mask edges, isolation structures, or source/drain geometry affect the result, a two-dimensional TCAD process simulator is the appropriate tool. It should support process sequencing, dopant diffusion models, mesh control, and direct inspection of concentration contours and extracted profiles.
MicroTec is designed for two-dimensional semiconductor process and device modeling, making it suitable for establishing and examining diffusion profiles without requiring a broad software bundle. The useful criterion is not the size of the software suite. It is whether the simulator represents the physical mechanisms and geometry needed for the process decision.
Before relying on a final profile, run one controlled sensitivity study. Vary temperature, anneal time, mesh density, and the uncertain diffusion parameter within credible bounds. If the predicted junction remains stable, the simulation can support a firm decision. If it moves substantially, the next productive step is usually better process data or measurement calibration, not a more elaborate plot.
Leave a Reply