SOL 102 › Unit 9

Unit 9 — Thin-Film Growth and Deposition Methods

Big idea. Modern cells are stacks of films only nanometres thick: silicon nitride anti-reflection layers (~75 nm), aluminium-oxide passivation (a few nm), tunnel oxides (~1.5 nm), doped poly-Si (~100 nm), a-Si:H (~5–10 nm), transparent conductive oxides (~70–100 nm). How a film nucleates and grows — and which deposition tool makes it — decides its optical, electrical and passivation quality.

Learning objectives

Why this matters

Thin-film deposition — PECVD, PVD/sputtering, ALD, CVD and TCOs — appears in 9 of 31 Tesla/SpaceX solar postings; vacuum/plasma systems in 5 (01-job-requirements.md). TOPCon, HJT and perovskite-on-silicon tandems are, physically, thin-film stacks on a wafer.

Nucleation

Atoms or molecules arriving at a surface (adatoms) diffuse around until they meet others and form clusters. Small clusters are unstable (too much surface energy for their volume) and fall apart; above a critical nucleus size they grow. Nucleation is easier on defects, steps and impurities (heterogeneous nucleation) than in free space. Higher substrate temperature lets adatoms diffuse further — fewer, larger grains; higher arrival rate gives more, smaller nuclei. The same ideas govern crystallisation from solution (Lab 102-A) and perovskite film formation.

Growth modes

Three growth modes (schematic). Layer-by-layer (film wets the substrate, e.g., lattice-matched epitaxy); islands (film atoms prefer each other; e.g., metals on oxides); layer-then-islands (strain builds up after a few layers).
Three growth modes (schematic). Layer-by-layer (film wets the substrate, e.g., lattice-matched epitaxy); islands (film atoms prefer each other; e.g., metals on oxides); layer-then-islands (strain builds up after a few layers).

Which mode occurs depends on surface and interface energies (does the film "wet" the substrate?) and lattice mismatch. Most production films on solar cells are amorphous or polycrystalline and grow far from equilibrium, but the vocabulary is used in process discussions.

Why vacuum?

Many methods need low pressure so atoms travel from source to substrate without colliding with gas, and to keep films clean. The mean free path of a gas molecule is

\[\lambda = \frac{k_B T}{\sqrt{2}\,\pi d^2 p}\]

Worked example 1 — mean free path

Nitrogen (\(d \approx 0.37\) nm), 300 K, \(p = 1\) Pa (0.0075 torr): \(\lambda = \dfrac{1.38\times10^{-23}\times300}{1.414\times\pi\times(3.7\times10^{-10})^2\times1} \approx 6.8\) mm. At atmospheric pressure (10⁵ Pa) it is ~68 nm; at 10⁻³ Pa (high vacuum) ~7 m — longer than the chamber, so evaporated atoms fly straight.

Deposition methods

Method How it works Typical solar uses
Thermal CVD / APCVD / LPCVD gases react on a hot substrate LPCVD intrinsic/doped poly-Si for TOPCon (~600 °C); oxides
PECVD (plasma-enhanced CVD) a plasma breaks up precursor gases (SiH₄, NH₃, N₂O, H₂…) so films form at 150–450 °C SiNₓ:H ARC/passivation; a-Si:H for HJT; doped poly-Si precursor layers
ALD (atomic layer deposition) alternating, self-limiting surface reactions — one sub-nm layer per cycle Al₂O₃ passivation (TMA + H₂O/O₃), ~0.1 nm/cycle
Sputtering (PVD) plasma ions knock atoms off a target onto the substrate ITO/TCO for HJT and tandems; metal seed layers; CdTe-line TCOs
Evaporation (thermal/e-beam) heat a source in high vacuum; vapour condenses on substrate metal contacts in labs; some perovskite layers
Solution processing spin-, blade-, slot-die- or spray-coating of precursor inks, then anneal perovskites, some transport layers
Vapour transport / CSS sublime a source onto a nearby hot substrate CdTe (First Solar uses a vapour-transport process)
Process temperature windows in solar-cell manufacturing (typical ranges; approximate, compiled from general process literature — exact recipes vary). Low-temperature methods matter for HJT (a-Si:H must stay below ~200–250 °C) and for perovskite top cells.
Process temperature windows in solar-cell manufacturing (typical ranges; approximate, compiled from general process literature — exact recipes vary). Low-temperature methods matter for HJT (a-Si:H must stay below ~200–250 °C) and for perovskite top cells.

PECVD SiNₓ is the workhorse: silane and ammonia in a plasma at ~400 °C deposit hydrogen-rich silicon nitride; tuning gas ratios sets the refractive index (~1.9–2.3) and hydrogen content. ALD Al₂O₃ gives a few nm of film with negative fixed charge for passivating p⁺ surfaces; because each cycle self-limits, thickness is controlled by counting cycles, and coverage of textured surfaces is excellent. Industry uses spatial ALD or fast batch ALD for throughput.

Worked example 2 — ALD cycles

Target 6 nm Al₂O₃ at a growth per cycle of about 0.11 nm (typical literature value for TMA/H₂O; estimate): \(6/0.11 \approx 55\) cycles.

Designing an anti-reflection coating

A single transparent film of index \(n_1\) and thickness \(d\) on silicon cancels reflection at wavelength \(\lambda_0\) when it is a quarter-wave thick inside the film, and cancels it completely if its index is the geometric mean of the media on either side:

\[d = \frac{\lambda_0}{4 n_1}, \qquad n_1^{\text{ideal}} = \sqrt{n_0\,n_\text{Si}}\]

In words: the reflections from the top and bottom of the film travel half a wavelength apart and cancel; they cancel fully when both reflections are equally strong.

Worked example 3 — SiNₓ coating

In air (\(n_0 = 1\)) with \(n_\text{Si} \approx 3.94\) at 600 nm: ideal \(n_1 = \sqrt{3.94} = 1.98\) — silicon nitride! Thickness for \(\lambda_0 = 600\) nm: \(d = 600/(4\times2.0) = 75\) nm. Under glass/EVA (\(n_0 \approx 1.5\)), the ideal index rises to \(\sqrt{1.5\times3.94} \approx 2.4\), which is one reason module-optimised cells use slightly higher-index or graded SiNₓ stacks.

Computed reflectance of polished silicon, bare vs with a 75 nm SiNₓ (n = 2.0) quarter-wave film, and with a 103 nm SiO₂ (n = 1.46) film. Thin-film interference at normal incidence, flat surface; n(Si) from approximate literature values (Green 2008), absorption neglected. Combined with texture, real cells reach a few per cent average reflectance.
Computed reflectance of polished silicon, bare vs with a 75 nm SiNₓ (n = 2.0) quarter-wave film, and with a 103 nm SiO₂ (n = 1.46) film. Thin-film interference at normal incidence, flat surface; n(Si) from approximate literature values (Green 2008), absorption neglected. Combined with texture, real cells reach a few per cent average reflectance.

Stress and adhesion

Films deposited hot and cooled down are stressed if their thermal expansion differs from silicon's (Unit 10):

\[\sigma \approx \frac{E_f}{1-\nu_f}\,\Delta\alpha\,\Delta T\]

Large stress causes cracking, delamination or wafer bow. Intrinsic stress from the growth process (ion bombardment in sputtering, hydrogen content in PECVD) adds to it. Adhesion depends on interface chemistry and cleanliness — another reason for pre-deposition cleaning.

Common misconceptions

Misconceptions

  • "Thicker coating = less reflection." Optical coatings work by interference at a specific thickness; too thick or too thin is worse.
  • "Plasma is just for heating." The plasma supplies reactive radicals and ions so chemistry happens at low temperature.
  • "ALD is too slow for factories." Spatial and batch ALD tools are used in high-volume PV lines.
  • "Vacuum is only to keep films clean." It also lets atoms travel in straight lines and lets plasmas exist at useful conditions.

What this means for manufacturing and the job

Review questions

1. Quarter-wave thickness of an n = 2.05 SiNₓ film for 620 nm?

620/(4 × 2.05) = 75.6 nm.

2. Why does PECVD let SiNₓ deposit at ~400 °C instead of ~800 °C?

The plasma dissociates SiH₄ and NH₃ into reactive radicals, supplying the energy that would otherwise come from heat.

3. Why is ALD good for textured wafers?

Self-limiting surface reactions coat every exposed surface uniformly (conformal), independent of line-of-sight.

4. Mean free path at 0.01 Pa (N₂, 300 K)?

6.8 mm × (1/0.01) = 0.68 m.

5. Ideal single-layer ARC index for silicon under EVA (n = 1.48)?

√(1.48 × 3.94) ≈ 2.41.

Sources