Waveguides & routing
How waveguides guide light, where optical power is lost along a path, and why in photonics geometry is loss — the principle that shapes every route you draw in Qfactr.
A waveguide is the wire of a photonic integrated circuit: a patterned region of high-index material that confines light and carries it from one component to another. Unlike a metal wire, a waveguide is a continuous optical structure, and its exact geometry — width, length, every bend and crossing — changes how much light arrives at the far end.
That makes routing a physical act, not a bookkeeping one. This page covers how a waveguide guides light, the mechanisms that cost optical power along a route, and how those ideas show up directly in the way Qfactr lets you draw and evaluate connections.
How a waveguide guides light
A waveguide guides light by index contrast: a core of high refractive index surrounded by lower-index cladding. Light traveling in the core is kept in place by total internal reflection at the core–cladding boundary — rays that strike the interface beyond the critical angle are reflected back inward rather than escaping. The larger the index contrast, the more tightly the light is confined, which is why a high-contrast platform like silicon-on-insulator can route light through bends far tighter than a low-contrast platform such as silicon nitride.
A waveguide does not carry light as a free-space beam but as one or more modes — stable transverse field patterns set by the cross-section and the wavelength. A single-mode waveguide supports exactly one mode per polarization; a multi-mode waveguide supports several. Most signal routing is done single-mode, because multiple modes travel at different speeds and interfere, distorting the signal. The cross-section is therefore designed so that, at the operating wavelength, only the fundamental mode propagates.
Where light is lost
Every route loses some optical power. Insertion loss — the total power lost between two points — is the sum of several mechanisms, each tied to the physical shape of the path. Understanding them is what lets you read a layout and predict where it will hurt.
| Mechanism | Cause | How geometry drives it |
|---|---|---|
| Propagation loss | Scattering off sidewall roughness and material absorption as light travels. | Scales with route length — longer detours cost more, so the shortest acceptable path wins. |
| Bend / radiation loss | At a bend, the mode is pushed toward the outer edge and part of it radiates away into the cladding. | Rises sharply as bend radius shrinks below the platform's minimum; every tight turn leaks power. |
| Crossing loss | Where two waveguides intersect, the mode is locally disturbed, scattering light and coupling to the crossing arm. | Each crossing adds a fixed penalty (plus crosstalk), so routes that avoid crossings stay cleaner. |
| Junction loss | Mode mismatch at transitions — width tapers, component pins, or splits — where the field profile changes. | Abrupt transitions radiate; well-tapered junctions recover most of the power. |
Two consequences fall out of this table. First, length is loss: propagation loss accumulates over distance, so a meandering route is quietly more lossy than a direct one. Second, and more dramatically, bends and crossings are loss: each one is a discrete event where power can radiate away. A layout that looks tidy on a schematic can be expensive once it is drawn as real geometry.
Routing styles and bend radius
Because light cannot turn a sharp corner without radiating, photonic routes are built from smooth curves rather than right angles. Two styles dominate:
- S-bends smoothly offset a path from one pin to another that is laterally displaced — the workhorse for connecting components whose ports do not line up.
- Manhattan routing runs segments along orthogonal axes, joined by rounded corners (never true 90° angles), keeping a layout structured and predictable in dense regions.
Both styles are bounded by the minimum bend radius — the tightest curve a given platform can take before bend loss becomes unacceptable. It is a property of the material platform and waveguide cross-section, not a free choice: high-contrast silicon photonics tolerates micron-scale radii, while lower-contrast platforms need much gentler curves and therefore more room. Respecting the minimum bend radius is the single most reliable way to keep a route low-loss.
In photonics, geometry is loss
The throughline of everything above is one principle: geometry is loss. In electronics a wire is essentially free — its length and the corners it turns barely matter to the signal. In photonics the opposite is true. Every bend, every crossing, every extra micron of detour costs optical power, and that cost is set entirely by the physical shape of the path. There is no abstract layer where connectivity exists independently of layout; how you draw a route *is* how much light survives it.
This is also why photonic layout cannot be treated as a routing afterthought. Where components sit determines whether they can be connected at acceptable loss at all, which is the heart of why photonic layout is hard.
Routing in Qfactr
Qfactr treats this principle as a first-class part of the design model. Waveguides are not abstract nets between symbols — they are explicit routes, drawn as real S-bend and Manhattan geometry between real component pins, in true micrometers on the canvas. A connection has a physical shape from the moment you make it.
Because the route is real geometry, the physics follows directly. Transmission and loss are derived from the actual path rather than estimated from a schematic, so the consequences of a long detour or a tight bend are visible in the design itself. Running Simulate animates a power-flow gradient along the waveguides (teal → green → yellow → red) and reports a loss readout in the status bar, turning the abstract idea that "geometry is loss" into something you can see on the path you just drew. For the step-by-step mechanics of making a connection, see Routing waveguides; for how that loss feedback is computed and where its limits are, see Simulation-aware design.