Pixel Structure Realism

Prerequisites

Pixel Anatomy introduces the layer stack and Pixel Optical Effects explains how each component moves QE and crosstalk. This page is about the structural details that separate a textbook stack from a real fabricated pixel.

A first-order pixel model — flat silicon, a rectangular oxide-filled trench, a clean microlens cap — is enough to reproduce the dominant QE trends. But the last few percent of QE, the near-infrared (NIR) response, and the crosstalk floor are all set by structural details that a flat model misses. This page collects the realism features that matter and maps each to its COMPASS config knob.

Realism gap map

Real-pixel feature	Why it matters optically	COMPASS config
Backside inverted-pyramid texture	Graded-index moth-eye AR + light trapping; large NIR QE gain	silicon.surface_texture
DTI conformal high-k liner	Thin high-index ring; passivation + negative fixed charge	silicon.dti.liner
Tapered DTI sidewall	Etched trenches narrow with depth; changes deep isolation fill	silicon.dti.taper_angle
Microlens residual base	Reflow/etch-back leaves a flat polymer slab under the cap	microlens.base_thickness
Per-color CF relief + contact angle	Each resist has its own height and sidewall slope	color_filter.<color>
Metal grid with rounded corners	Real grids are not perfect squares	color_filter.grid.corner_radius

Backside inverted-pyramid texture (light trapping)

Modern NIR-enhanced BSI sensors etch an inverted-pyramid array (IPA) into the silicon backside that faces the incoming light. Two mechanisms boost QE:

1. Graded-index anti-reflection. Going from the trench fill into bulk silicon, the area-averaged silicon fraction rises smoothly from ~0 at the surface to 1 at the pyramid apex. The effective index therefore ramps gradually rather than stepping abruptly at a flat Si interface, suppressing the front-surface reflection in the same way a moth-eye coating does.
2. Light trapping. The faceted surface refracts long-wavelength light to oblique angles, lengthening its path in the (weakly absorbing) silicon and increasing the chance of absorption before it escapes.

Published device simulations and process reports show that combining an IPA with deep DTI raises near-infrared QE substantially — on the order of ~3x at 850 nm and ~5x at 940 nm relative to a planar backside — because silicon's absorption length grows to tens of microns in that band.

COMPASS models the IPA as a staircase of pyramidal pits carved from the top of the silicon layer and back-filled with fill_material. The pit half-width shrinks linearly from period/2 at the surface to zero at the apex, producing the graded silicon fill fraction:

silicon:
  surface_texture:
    enabled: true
    type: inverted_pyramid
    height: 0.35          # texture depth into silicon (um)
    period: null          # defaults to the pixel pitch
    fill_material: sio2
    n_slices: 8           # staircase resolution for RCWA

DTI: liner, fill, and taper

A real backside deep trench is not a clean oxide rectangle:

• Conformal high-k liner. The etched silicon sidewall is lined with a thin high-k film (Al2O3, HfO2, or Ta2O5, typically tens of nm) that both passivates dangling bonds and carries a negative fixed charge to repel electrons from the damaged surface, lowering dark current. Optically it is a thin high-index ring sitting between the silicon and the lower-index core fill, so it changes the trench's reflection and its crosstalk contribution.
• Core fill. Inside the liner the trench is filled with oxide (most common), or in some designs poly-silicon, tungsten (metal DTI), or even an air gap.
• Tapered profile. Plasma-etched trenches narrow with depth. A vertical-wall idealisation over-counts the isolation material deep in the substrate and mislocates the trench/silicon boundary for ray and field calculations.

silicon:
  dti:
    enabled: true
    mode: fdti
    width: 0.12           # trench width at the backside opening (um)
    depth: 4.0
    material: sio2        # core fill
    taper_angle: 82.0     # sidewall angle from substrate plane (90 = vertical)
    n_slices: 6           # staircase resolution for the taper
    liner:
      enabled: true
      material: al2o3     # high-k passivation liner
      thickness: 0.015

When the liner is disabled, the taper angle is 90°, and no texture is present, COMPASS uses its original fast single-slice (FDTI) or two-slice (BDTI) path, so existing configs are unchanged. Enabling any of these features switches the silicon layer to a z-resolved staircase that the RCWA and FDTD back-ends consume transparently.

Microlens residual base

Polymer microlenses are formed by reflowing or etching back a patterned resist. The process always leaves a flat residual slab of the same polymer beneath the curved cap — the lens is never zero-thickness at its edges. Ignoring it slightly under-estimates the optical path in the lens material and the height of the air gap above the color filter.

microlens:
  height: 0.67            # curved cap sag
  base_thickness: 0.10    # flat residual polymer slab under the cap

Putting it together

The sample_p1p12um_nir preset turns all of these on at once and is the subject of the Pixel Structure Realism report, which renders the actual solver permittivity Re(eps)(x, z) so you can see the graded texture, the lined tapered trench, and the residual lens base in the geometry the solver integrates.

python scripts/run_simulation.py pixel=sample_p1p12um_nir

Scope

These features make the geometry more faithful. Whether they change QE for your stack is an optical question — run an RCWA order-converged sweep (and, for the texture, a wavelength sweep into the NIR) before quoting deltas.