BARL Optimization Benchmark

Generated on 2026-06-11 with the COMPASS TMM solver on the generic 1.0 um BSI stack.

This report turns the BARL cookbook recipe into a reproducible benchmark. It sweeps single-layer Si3N4 thickness, a two-layer SiO2/HfO2 grid, and then compares representative designs on the same 400-700 nm wavelength grid.

Executive summary

• The 550 nm ideal ARC index between cf_green and silicon is 2.5148; Si3N4 gives a quarter-wave thickness of 67.9612 nm.
• The best single-layer Si3N4 sweep point is 70 nm with mean R=0.0604.
• The best two-layer SiO2/HfO2 grid point is SiO2 5 nm / HfO2 65 nm with mean R=0.0627.
• Best design in this candidate set reduces mean reflection by 0.029 absolute versus no BARL.

Planar proxy

This is a TMM planar green-stack benchmark. It is appropriate for BARL screening and reflection trends, but it does not include lateral Bayer geometry, microlens focusing, metal-grid diffraction, or crosstalk.

Candidate Spectra

Single-layer Sweep

Two-layer Sweep

Design Scorecard

Design	role	layers	total nm	mean R	max R	R@550	mean A	energy residual
No BARL	baseline	none	0	0.0895	0.3656	0.0091	0.6492	1.11e-16
Default 4-layer	sample default	sio2 10 nm / hfo2 25 nm / sio2 15 nm / si3n4 30 nm	80	0.0698	0.2213	0.2213	0.6345	1.11e-16
Si3N4 quarter-wave	analytic	si3n4 68 nm	67.9612	0.0605	0.1788	0.1788	0.6274	1.11e-16
Best single Si3N4	sweep best	si3n4 70 nm	70	0.0604	0.1766	0.1766	0.6272	0
Best SiO2/HfO2	sweep best	sio2 5 nm / hfo2 65 nm	70	0.0627	0.195	0.195	0.6294	1.11e-16

Interpretation

• BARL tuning should optimize broadband reflection, not only R at 550 nm. A quarter-wave layer is a useful seed, but it is not automatically the broadband optimum once color-filter and planarization phases are included.
• The default sample BARL is an illustrative process stack, not a guaranteed optimum. This report makes that explicit by comparing it against simple swept alternatives.
• The two-layer optimum sits on the lower SiO2 sweep boundary, so the next local search should extend toward thinner SiO2 or test HfO2-only variants.
• After a TMM BARL candidate is selected, run a patterned RCWA check because the metal grid and microlens can move the apparent optimum.

Regeneration

uv run python scripts\generate_barl_optimization_report.py

Generated metrics are stored at docs/public/reports/barl-optimization/barl_optimization_metrics.json.