Laser doe dicing

Laser Dicing / Cutting DOE

Operates under the rules in ../../INSTRUCTIONS.md (reference-first, reference-only, interactive mode).

When to use this skill

When developing a recipe for separating material — full-thickness cutting, wafer dicing, PCB routing, or stealth dicing. The decisions here are different from ablation (surface removal): there's a path-selection step at the top, a geometric feasibility limit, polarization considerations, and dissimilar-material trade-offs.

What this skill is NOT

• Not for surface marking / shallow scribing (use laser-doe-ablation).
• Not for diagnosing an existing recipe gone wrong (use laser-defect-diagnose).

References to load at start

• references/laser_dicing.md — primary, full document
• references/laser_ablation.md — for the ablative-cutting branch (overlap, thermal accumulation)
• references/laser_process_calculations.md — fluence, peak intensity
• references/gaussian_beam_theory.md — M², spot size, depth of focus
• references/laser_optics_selection.md §5 — damage threshold check on chosen fluence

Inputs to gather first

Ask the operator:

1. Material + thickness.
2. Through-cut, or partial-cut + cleave? (Dicing-style is partial + cleave — laser_dicing.md §1.)
3. Is the material transparent at the operating wavelength? (Critical for stealth-dicing decision.)
4. Edge-quality requirements — chip-out tolerance, wall-angle requirement, recast tolerance.
5. Throughput target.
6. Available laser — wavelength, max power, pulse-width regime (UFL? ns? CW?), polarization (linear / circular / unpolarized).
7. Optics + scanner — focal-spot diameter, beam expander, working distance.
8. Heterogeneous stack? (e.g., PCB with epoxy + glass fiber + copper.)

→ Pause: confirm inputs.

---

Workflow — staged decisions

Stage 0 — DECISION: which cutting path?

Per laser_dicing.md §1–§2.

Path	When to use
Stealth dicing	Material is transparent at λ; brittle (Si at appropriate λ, sapphire, glass); partial-cut + cleave acceptable
Ablative cutting (UFL or ns)	Standard opaque material; micrometer-level accuracy needed
Fusion cutting	Thick metal; high-power CW available; sub-mm tolerance acceptable

→ Pause: present the three with rationales. Confirm the path. Branch:

• Stealth → jump to Stage 4-S.
• Ablative or Fusion → continue to Stage 1.

Stage 1 — Geometric feasibility check (ablative / fusion path)

Per laser_dicing.md §3.

The wall-angle limit is a right-triangle problem:

max_single_column_depth ≈ focal_spot_diameter / tan(min_wall_angle)

Where min_wall_angle is the steepest slope at which the wall still couples enough laser energy to ablate.

Compute and report

• The triangle's height for the operator's focal-spot diameter at typical wall angles.
• A typical focused beam reaches ~ 200 µm in average materials (per laser_dicing.md §3 — engineer's intuition).
• If material thickness exceeds that limit:
  • Single-column cut will not get through.
  • Plan a multi-pass parallel widening strategy.
  • Number of passes scales with depth². Doubling cut depth quadruples cut time.

→ Pause: present the geometric feasibility report. If multi-pass is needed, confirm the depth-to-passes estimate is acceptable to the throughput budget. If not, recommend a different focal-spot configuration (smaller focus + more passes vs. larger focus + fewer but wider passes).

Stage 2 — Polarization decision

Per laser_dicing.md §4.

Situation	Polarization
First-pass-only on flat surface	Linear is OK
Multi-pass with formed walls	Circular polarization strongly recommended — X and Y cuts then behave identically

If the operator's laser outputs linear polarization and circular is needed, recommend inserting a quarter-wave plate.

→ Pause: confirm polarization configuration. Verify hardware can deliver it.

Stage 3 — Material-class warnings

Per laser_dicing.md §5–§6.

Check the operator's material against known problem classes:

Material class	Specific warnings
Glass / Si at melt	Self-healing reflective surface; UV or mid-IR strongly preferred
Plastics / polymers	Molten material reflows back into kerf; UFL strongly preferred
PCB stack (epoxy + fiber)	Recipe compromise: high speed + cooling gas for epoxy, high pulse energy + correct λ for fiber
Metal-on-plastic stack (blind via)	Shock-delamination regime — pulse-width and PRF tuning is critical
Self-funneling cuts	Some configurations form waveguide-like walls — high aspect ratio possible (§5)

→ Pause: flag any applicable warnings. Operator confirms they understand and how they'll address them.

Stage 4-A — Multi-pass parameter matrix (ablative)

For ablative cutting:

Build the matrix

Three axes:

• Pass count: 1, 2, 4, 8, 16 (logarithmic).
• Power: log-spaced 4–6 levels (1.6× steps from laser_ablation.md §12).
• Scan speed: log-spaced 4–6 levels.

Total cells: typically 5 × 5 × 3 = 75 if all three are explored. Pin one variable (usually pass-count) to a reasonable starting estimate from Stage 1's geometric calculation, to keep the matrix tractable.

What the skill computes per cell

• Fluence (J/cm²).
• Multiple of ablation threshold (if known from a prior laser-doe-ablation run on the same material).
• Pulse-to-pulse pitch.
• Estimated cycle time per cut.
• Optic damage check (laser_optics_selection.md §5).

→ Pause: review the matrix. Operator can edit, drop flagged cells, or restart.

After running

Operator inspects each cell for:

• Did it cut through?
• Wall quality (taper, recast).
• HAZ at edges.
• Re-deposited material.

→ Pause: identify the best cell (cut-through + clean walls + acceptable cycle time). Confirm.

Stage 4-S — Stealth dicing parameter sweep

Per laser_dicing.md §2.

Goal: internal modification only, no surface ablation.

Build the matrix

Three axes:

• Focal Z (in-bulk depth): 4–5 levels stepping into the bulk.
• Pulse energy: 4–5 log-spaced levels.
• Pulse spacing: 3 levels (close, medium, sparse) — defines modification line continuity.

Critical constraint

• Pulse energy must be above the in-bulk modification threshold (non-linear absorption regime).
• Pulse energy must be below the surface-ablation threshold at the entry surface.
• The skill computes both thresholds and flags cells outside this window.

→ Pause: review. Operator commits.

After running

Cleave the wafer along the modification line. Inspect:

• Did it cleave cleanly?
• Edge quality.
• Surface damage on the entry side (must be none — that's the entire point of stealth).

→ Pause: identify the best cell.

Stage 5 — Validate at production scan speed

The matrix was tuned at characterization speed; production speed may differ.

Run the chosen recipe at production scan speed on a real coupon. Iterate if needed.

→ Pause: confirm production run is acceptable.

---

Final output

1. Decision log (markdown) — path-choice rationale, polarization choice, geometric pass count, material warnings.
2. DOE matrix (CSV / xlsx).
3. Final recipe summary (markdown) with reference citations for every decision.

Save to outputs/laser-doe-dicing/<YYYY-MM-DD>-<material>/.

---

Cross-skill links

• Position-specific defects (corners, line ends) during validation → laser-delay-tuner.
• Cell can't cut through but the geometry math says it should → re-check Stage 1, or hand to laser-troubleshoot.
• Power drifts during the experiment → laser-power-measurement.
• General defect classification mid-run → laser-defect-diagnose.