Needle Variation
Needle Variation improves a design by inserting infinitesimally thin “needle” layers at the spot where they help most, then refining the result. To find that spot, it scans every position in the stack with every candidate material and computes how the merit function would change if a needle of a given material were inserted there, a quantity known as the P-function. The most promising position and material win, the needle goes in at a finite thickness, and a refinement pass settles the new stack. The cycle repeats until no insertion improves the design any further.
There are two versions in the ribbon. Both share the same candidate-material pool and the same underlying scan.
Needle Automatic
Section titled “Needle Automatic”Runs the whole scan-insert-refine loop on its own until it reaches needle-optimality. Inside each layer the scan tests 16 positions, and every point where the needle function has a minimum along the layer becomes a candidate. The wavelength grid the band targets are sampled on is sized from the design and grows with it during the run (see Operand Reference).
A generation counts as an improvement only when it lowers the merit function by at least 0.01 %, the same fall at which each refinement inside the run stops. Smaller gains would fill the layer budget with needles that stay about a nanometre thick. The first time no insertion improves the design, the run refines copies of it two, four, eight and sixteen times thicker and goes on from the best of them; the history shows this as a Thicker start row. If the run never beats the design it had before, it ends on that design.
Needle Manual
Section titled “Needle Manual”Shows you the same P-function scan as a plot and lets you click the position and material to insert a single needle by hand. This is useful for testing a topology idea, seeding a layer where you know one belongs, or stepping through synthesis one insertion at a time. With Refine after insert on, each insertion is refined with SQP, the Refinement window’s default method, for the number of Refine iterations set.
Settings
Section titled “Settings”Candidate materials: the pool of materials Needle is allowed to insert. The pool starts with only the materials of the design’s layers ticked, front and back; the substrate and the media are listed under This design but unticked, and no other catalog is ticked. All / Clear select or empty the pool in one click. Once you change the pool, the window keeps your choice for every design, except that each design’s substrate and media stay unticked until you tick them on that design; a pool emptied with Clear starts again from the design’s own materials the next time the window opens.
Max layers: an upper limit on how many layers the design may grow to.
Target MF: stop once the merit function reaches this value (0 means run to convergence).
The advanced section exposes the synthesis tuning:
Needle δ (nm) (Probe δ in Needle Manual): the thickness step used when probing candidate positions.
Min thickness (nm): the minimum inserted thickness and the floor used during refinement (1 nm by default). Keep it at the synthesis default while synthesizing; raise it to your manufacturing minimum only in a later Refinement and Design Cleaner pass, so you don’t hold synthesis back with a manufacturable floor too early. Needle Automatic offers a needle inside a layer only where both parts of the split layer stay at or above this floor; the Needle Manual profile draws the needle function everywhere and leaves the choice to you. A layer that refinement drives down onto the floor stays only while it helps: when a generation picks its best candidate, Needle Automatic also refines that design without its layers on the floor, with half the iterations, and keeps that version when its merit is no higher.
Refine iter / gen: how many refinement steps run after each insertion.
Inner refiner: which method refines the stack after each insert. The default is Trust-region Newton, which ended lower than Conjugate Gradient on most of the benchmark’s cases and grows the stack without collapsing it. See Optimization Methods for the alternatives.
Candidate search: how thoroughly each step explores the improving candidates (fast, balanced, or thorough), trading speed against quality.
Needle Automatic ignores the minimum and maximum thickness limits (MNT and
MXT rows); re-enable them by running Refinement
afterwards. Needle Manual leaves them out of its scan and of the thickness a
needle starts at, and its refine after an insertion holds them. The surface
mode set in the Design Editor is honoured for every mode, not just the front.
A peak-to-peak (PPEF) row is left out of synthesis too, with any math row that refers to it, so the MF shown here does not count it: its gradient comes from only the two wavelengths where the error is highest and lowest, and the needle search stalls on that. Needle fits the curve block instead, a note under Settings says so, and a Refinement run afterwards brings the peak-to-peak error down.
How to read it
Section titled “How to read it”The window shows a live preview of the design as needles go in, a scan plot of the P-function versus position for the top candidate materials (the deepest dip marks the best place to insert), and a per-cycle merit history. When you click Done the final design is committed to the active design.
Needle can stall when the topology change it needs is a thick spacer rather than a thin needle; Gradual Evolution escapes those cases with its forced-insertion step, so a common pattern is a few Needle cycles, then Gradual Evolution, then Needle again.
References
Section titled “References”- Sullivan & Dobrowolski, Appl. Opt. 35, 5484 (1996).
- Tikhonravov et al., Appl. Opt. 35, 5493 (1996).
- Tikhonravov et al., SPIE 4829 (2003).