Broadband Monitoring Simulator

The Broadband Monitoring Simulator is a 6-page wizard that simulates what happens when your design is actually deposited and watched by an in-chamber broadband spectrophotometer. It grows the coating layer by layer with realistic deposition-rate jitter, per-material index drift, and signal noise, lets the simulated monitor decide when to cut each layer, and then shows you the manufactured spectrum next to the theoretical one so you can see how well the design survives production.

You set up the deposition conditions on the first four pages, run a single computational-manufacturing experiment on page 5 and scrub through it like a movie, and read the resulting performance on page 6.

Settings

The wizard walks through one topic per page.

Page 1: Deposition Rates. For each material, set the mean rate (nm/s), the RMS rate fluctuation, and the correlation time that controls how slowly the rate drifts. The preview shows a sample rate-vs-time trace; press Randomize to draw a new one. The rate wanders within each layer at the scan interval, as the preview shows, and carries on into the next layer of the same material. With a correlation time of 0 the rate noise averages out of the thickness completely.

Page 2: Parameters Deviation. Per material, add a systematic and random shift to the real refractive index, plus a systematic inhomogeneity. The lower table lets you exclude individual layers from monitoring (they are then cut purely on time) and give each one an extra relative thickness error. Shutter delay (mean and RMS, in seconds) models the lag between the cut decision and the shutter actually closing.

Page 3: Monitoring System. Choose the measured quantity and polarization (T or R, s/p/average), the angle of incidence, the scan interval between spectrum readings, and the monitoring band (λ min, λ max, and number of points). Chip glass is the witness chip the monitor watches: it opens on the design substrate, and picking another material moves the monitor signal onto that glass, for a witness that is not the same glass as the part. The preview shows the ideal monitoring signal for the layer selected in the tab strip.

Page 4: Signal Errors. Add random noise (percent of signal) and a slow baseline drift to the monitor signal. The preview shows the noisy signal for the selected layer.

Page 5: Deposition Simulation. Press Start to run one full manufacturing experiment. The coating then plays back layer by layer on an interactive timeline (play/pause, speed, scrub, layer ticks). The bar chart compares the estimated, actual, and target thickness of the current layer; the spectrum shows the theoretical guide curves (end, 80 %, 90 %) against the as-built curve. Once a layer is cut, its estimated bar is what the monitor believes it deposited, which is the target unless the cut ran late.

Page 6: Resulting Performance. Tabs show the manufactured vs. theoretical spectrum, relative and absolute thickness-error bars per layer, and tables of as-built thicknesses and refractive indices.

How the monitor cuts a layer. From 60 % of a layer’s planned time on, the monitor fits the thickness of the growing layer to every scan. Each fit scans the whole range from zero to three times the target at a step finer than the fringes of the monitoring band before it refines, so it needs no starting guess and settles in the best-fitting fringe rather than the nearest one. A tracker follows the layer’s thickness and rate from these fits, weighting each fit by how well its scan pins the thickness, and the shutter closes where the tracked thickness reaches the target, between scans if need be. Without noise every layer ends on target at any scan interval. Each fit is made over the monitor’s own estimate of the layers below, not over what the chamber really deposited, so an error in one layer carries into the fits of the next ones the way it does in a real chamber. An excluded layer enters that model at its target.

The coating side that is deposited, and the way the resulting spectrum is scored, follow the surface mode set in the Design Editor, shown as a badge on the window. The in-chamber monitor signal is read through the whole witness chip: the growing coating on its front face and its bare back face, added incoherently, the way a spectrophotometer aimed through the chamber actually sees it. A bare chip of n = 1.52 glass reads 91.8 %, not the 95.7 % of the coated surface alone. The chip hangs in the chamber, so the signal is read with air above the growing coating whatever medium the design is embedded in.

How to read it

Page 6 is the verdict. If the manufactured curve hugs the theoretical one and the error bars are small, the design is robust to the monitoring conditions you set. Large thickness errors on a particular layer point to a layer that is hard to monitor at the chosen wavelength or strategy, a candidate for a different monitoring wavelength, tighter rate control, or a more tolerant redesign. Because every run uses fresh random draws, run it a few times (or re-run page 5) to see the spread of outcomes rather than trusting a single realization.

References

• Tikhonravov & Trubetskov, Appl. Opt. 44, 6877 (2005), computational manufacturing as a bridge between design and production.
• H. A. Macleod, Thin-Film Optical Filters, 5th ed., Ch. 12.