DTI Crosstalk Benchmark

Generated on 2026-06-11 from PixelStack geometry sweeps and a representative scalar FDTD snapshot.

This report separates cheap, reproducible DTI geometry evidence from the more expensive localized-source crosstalk benchmark. It is intended as a design-space gate before running longer vector FDTD or high-order RCWA jobs.

Executive summary

• A 100 nm full-depth DTI occupies 0.19 of the silicon volume in the generic 1.0 um 2x2 BSI PixelStack geometry.
• A 100 nm, 0.6 um BDTI occupies 0.0393 of the silicon volume because only the backside/top portion is trenched.
• The representative 44x44x118, 3500-step scalar FDTD snapshot reports max neighbor crosstalk 0.2582 for FDTI and 0.2609 for BDTI, a gap of 0.0027 absolute.
• The periodic trench RCWA/FDTD alignment snapshot remains within about 3 percentage points in R/T/A for both FDTI and BDTI.

Scope

The crosstalk matrix is a scalar FDTD visual benchmark, not a production full-vector FDTD solve. Use it to compare geometry and normalization paths; use longer vector runs for final isolation claims.

Geometry Sweep

FDTI width sweep

width nm	max XY DTI area	effective DTI volume	open Si volume
0	0	0	1
40	0.0784	0.0784	0.9216
60	0.1164	0.1164	0.8836
80	0.1536	0.1536	0.8464
100	0.19	0.19	0.81
120	0.2256	0.2256	0.7744
150	0.2775	0.2775	0.7225

BDTI depth sweep

BDTI depth um	active depth um	effective DTI volume	open Si volume
0	0	0	1
0.3	0.3	0.0197	0.9803
0.6	0.6	0.0393	0.9607
1.2	1.2	0.0786	0.9214
1.8	1.8	0.1179	0.8821
2.4	2.4	0.1572	0.8428
2.9	2.9	0.19	0.81

Silicon DTI Masks

Representative Crosstalk Snapshot

mode	mean self collection	max neighbor crosstalk	mean PD signal	energy tail change
fdti	0.5831	0.2582	15320.3545	0.0471
bdti_0p6um	0.5827	0.2609	15249.1798	0.047

Periodic Trench Alignment Snapshot

mode	max abs dR	max abs dT	max abs dA	Si absorption proxy	trench field leakage
fdti	0.0266	0.027	0.0278	0.6087	0.1305
bdti	0.0183	0.0088	0.019	0.6105	0.096

Interpretation

• FDTI and BDTI have similar crosstalk in the representative coarse scalar snapshot because the run is primarily a path and normalization check.
• The geometry sweep still shows the intended monotonic controls: wider FDTI increases silicon trench volume, and deeper BDTI approaches FDTI.
• A production DTI report should extend this with wavelength-resolved, localized-source vector FDTD and a true width/depth/material crosstalk sweep.

Regeneration

uv run python scripts\generate_dti_crosstalk_report.py

Generated metrics are stored at docs/public/reports/dti-crosstalk/dti_crosstalk_metrics.json.