Thermal oxidation in semiconductor fab

In short: Thermal oxidation grows high-quality SiO₂ directly from the silicon wafer by reacting it with oxygen (dry) or steam (wet) at 900–1200 °C, and the Deal–Grove model predicts the resulting thickness from a linear rate (surface-reaction-limited) and a parabolic rate (diffusion-limited) regime.

Thermal oxidation is one of the few fab steps where you consume the wafer to build your film rather than deposit material onto it. It is the backbone of the gate dielectric, field-isolation oxides (LOCOS/STI liners), masking layers, and screen oxides for implantation. Understanding when to run dry versus wet oxidation, and why the growth slows down as the film thickens, is a staple of any process-integration or device-physics interview.

What actually happens at the Si–SiO₂ interface

Oxidation is a reaction, not a coating. The oxidant must diffuse through the already-grown oxide to reach fresh silicon at the buried interface, so the interface moves into the wafer. For every unit of oxide grown, roughly 44% of its thickness comes from consumed silicon. The accepted conversion: growing 100 nm of SiO₂ consumes about 46 nm of silicon (the volume expansion ratio is ~2.17, from the density/molar-volume difference between Si and SiO₂).

Rule of thumb: oxide thickness × 0.46 = silicon consumed. A 20 nm gate oxide eats ~9 nm into the wafer — which is exactly why aggressive scaling eventually pushed the industry off thermal SiO₂ toward deposited high-κ dielectrics like HfO₂.

Dry vs wet oxidation

The oxidant chemistry decides everything about rate and quality:

• Dry oxidation uses pure O₂: Si + O₂ → SiO₂. It is slow but produces dense, low-defect, electrically superior oxide. This is the only acceptable choice for the gate dielectric.
• Wet oxidation uses steam: Si + 2H₂O → SiO₂ + 2H₂. Water has a much higher solubility and diffusivity in SiO₂, so it grows 5–10× faster — ideal for thick field and masking oxides where quality matters less.

Interview favourite: why is dry oxide "better"? Fewer dangling bonds and no residual –OH groups at the interface means lower D_it, better breakdown field (~10 MV/cm), and tighter threshold-voltage control. Fabs even run a short dry step at the start and end of a "dry-wet-dry" cycle to cap a fast wet growth with clean interfaces.

The Deal–Grove model

Deal and Grove (1965) gave the model every process engineer still uses. It treats growth as three fluxes in series — gas-phase transport, diffusion through the oxide, and the interface reaction — and yields the linear–parabolic relation:

xox² + A·xox = B(t + τ)

Two limiting regimes fall out of it, which is the whole point of the model:

• Thin-oxide (linear) regime: growth is limited by the reaction rate at the interface. xox ≈ (B/A)(t + τ). The ratio B/A is the linear rate constant.
• Thick-oxide (parabolic) regime: growth is limited by oxidant diffusion through an ever-thickening film. xox ≈ √(Bt), where B is the parabolic rate constant. This is why a 500 nm oxide takes disproportionately longer than a 50 nm one.

Both B and B/A follow Arrhenius behaviour, so a 100 °C temperature increase can roughly double the rate. The τ term accounts for the fast initial growth Deal–Grove underpredicts for very thin dry oxides — a known limitation that motivated later models (Massoud, reduced-pressure regimes).

Where it fits in the fab flow and modern tooling

Oxidation runs in horizontal or vertical diffusion furnaces (or rapid-thermal / RTO chambers for ultra-thin gate oxides). In the TCAD world, engineers calibrate oxidation and dopant redistribution in process simulators such as Synopsys Sentaurus Process and Silvaco Athena, where the Deal–Grove and Massoud coefficients are literal input parameters. Downstream device and parasitic extraction then flow into tools from Synopsys, Cadence, and Siemens EDA. If you are comparing oxidation with film-building steps, see our companion guide on deposition: CVD, PVD and ALD — and for how the grown oxide is then patterned, our photolithography fundamentals walkthrough.

Want to run real Deal–Grove calculations and furnace recipes with guided labs? Explore the Nano Fab track on our CourseTron courses page and build process intuition that holds up in interviews.

For teams moving from coursework into actual tapeouts, hands-on training through ChipXpert's VLSI programs pairs well with production-grade process and PD exposure.

Common pitfalls (and interview traps)

• Forgetting silicon consumption when computing final step height — it matters for LOCOS "bird's beak" and planarity budgets.
• Using wet oxide for the gate — the higher D_it wrecks V_th control.
• Assuming linear growth forever — thick oxides are firmly parabolic; doubling time does not double thickness.
• Ignoring orientation and doping — (111) silicon oxidizes faster than (100); heavily doped silicon grows oxide faster too.

Frequently asked questions

Why is dry oxidation preferred for gate oxides?

Dry O₂ produces a denser film with far lower interface trap density and no residual hydroxyl groups, giving a higher breakdown field and tighter threshold-voltage control — exactly what a thin, electrically critical gate dielectric needs.

What is the physical meaning of the linear and parabolic rate constants?

The linear constant (B/A) dominates when growth is limited by the chemical reaction at the Si interface (thin oxides); the parabolic constant (B) dominates when it is limited by oxidant diffusion through the thickening film (thick oxides).

How much silicon is consumed during oxidation?

About 46% of the final oxide thickness. Growing 100 nm of SiO₂ consumes roughly 46 nm of silicon because SiO₂ occupies ~2.17× the volume of the silicon it replaces.

Where does the Deal–Grove model break down?

It underpredicts the fast initial growth of very thin dry oxides (below ~20 nm). The τ offset partly corrects this, and models like Massoud's or reduced-pressure formulations are used for ultra-thin regimes.