Why photonic layout is hard
Photonic layout is not electronic layout. Where you place a component decides whether it can be routed at acceptable loss, which makes physical arrangement and connectivity a single, tightly coupled problem.
If you come to photonic design from the electronic world, the layout feels familiar at first: components on a plane, connections between them, a netlist to satisfy. But the physics underneath is different enough that most of your intuitions about what is cheap and what is expensive no longer hold. In a photonic integrated circuit, the wire is not a wire — it is a waveguide that confines light, and every centimeter, every bend, and every crossing it travels through removes optical power that you never get back.
This page explains why laying out a PIC is genuinely hard, why placement and routing cannot be separated the way they often are in electronics, and why Qfactr treats the design as a physically grounded, routable model rather than an abstract schematic.
Photonic layout is not electronic layout
In a digital chip, metal interconnect is plentiful and cheap. Wires can cross on different metal layers, be lengthened to balance timing, and fanned out almost arbitrarily; routing is mostly a connectivity problem solved over a forgiving substrate. Photonics gives you none of that slack. Light is an analog signal carried in the geometry of the waveguide itself, and the geometry is the circuit.
A few properties of light-in-a-waveguide drive nearly every layout decision:
| Constraint | What it means | Why it makes layout hard |
|---|---|---|
| Crossings cost loss | Waveguides cannot cross on another layer the way metal does; a crossing is a physical device that scatters and reflects light. | You cannot route your way out of a placement that forces many crossings — each one is paid for in optical power. |
| Bends cost loss | Light radiates out of a waveguide as it turns, and loss grows sharply as the bend tightens below the minimum bend radius. | Routes want gentle S-bends, which take space; tight turns to save area leak power. |
| Components are large | Devices like MMI couplers, ring resonators, and grating couplers occupy real micrometers, often large relative to the links between them. | Footprints dominate the floorplan, leaving little room to absorb a bad arrangement with clever routing. |
| Routing is physics-constrained | A connection is real S-bend or Manhattan waveguide geometry between physical pins, not an abstract net. | Whether two pins can be joined at acceptable loss depends on their exact positions and orientations. |
Placement and routing are entangled
In a typical electronic flow you place first and route second, and a router can usually connect whatever the placer produced. In photonics that separation breaks down. Because connections are real geometry that must turn gently, avoid crossings, and respect each device's pin orientation, the feasibility of routing is decided at placement time.
Put a component down in the wrong place and there may be no waveguide path to its neighbor that stays within an acceptable loss budget — not because the router is weak, but because the physics forbids it. Move the same component a few micrometers, or rotate it so its pins face the right way, and a clean route appears. Connectivity (what must connect) and physical feasibility (whether it can connect cheaply) are a single, coupled problem.
Qfactr makes this coupling visible instead of hiding it. Components live at real micron coordinates with physical pins, and routes are drawn as the actual S-bend or Manhattan waveguide path between those pins. Because transmission and loss are derived from that real path geometry, the cost of a layout decision shows up immediately — pins read red until connected and green once joined, and a loss readout reflects the design you actually drew.
Why manual layout is slow
Faced with an entangled placement-and-routing problem, a designer falls back on iteration: place some parts, attempt the routes, discover a crossing or an impossible bend, nudge things, and try again. Each loop touches physical geometry and has to be re-checked against loss. For a handful of components this is tractable. For a dense system it is not.
- The space of arrangements grows combinatorially — every component has a position and an orientation, and every choice changes which routes are feasible.
- Feasibility is non-local: a single misplaced part can force crossings or tight bends several hops away.
- Each candidate arrangement must be evaluated against real routing geometry, not a quick connectivity check, so iterations are expensive.
- Good human layout depends on hard-won intuition for how light moves through space — knowledge that is scarce and slow to build.
The result is that getting from idea to a clean, low-loss layout is measured in weeks of manual iteration. The premise of Qfactr is that with a physically grounded model that already understands routing and loss, you can shorten that loop and iterate in days, not weeks.
Toward routing-aware placement
If the hard part is that placement decides routability, then the leverage is in placing components with routing in mind from the start — exploring arrangements that respect the physical difficulty of threading waveguides through a dense system, rather than treating placement as a purely geometric packing problem and hoping a router can finish the job.
This is exactly where a physics-informed, generative approach matters: proposals should be constrained by real routing and loss physics, not by abstract notions of placement. Because Qfactr already holds the design as a routable, simulation-aware model, the same physical signals that tell you a layout is good can guide a model toward better ones.