The router does not create space. A block can meet timing on paper and still be unroutable, and when detailed routing fails the cause almost always sits upstream — in the floorplan, the placement and the utilisation — not in the router that reported it.
| Lever | Stage | What it does |
|---|---|---|
| Floorplan and macro placement | Floorplan | Sets channels, pin sides and aspect ratio — the single biggest influence |
| Target utilisation | Placement | Global headroom; 65–80% is typical but design-dependent |
| Cell padding / density screens | Placement | Spreads cells out of local hotspots |
| Congestion-driven placement | Placement | Trades a little timing for a routable spread |
| Blockages and macro halos | Floorplan/Placement | Keeps cells out of channels and congested regions |
| Pin-access-aware placement | Placement | Avoids unroutable pin patterns at boundaries |
| High-fanout / clock restructuring | RTL/CTS | Removes the nets that create clumps |
| Layer assignment and NDR | Routing | Balances horizontal and vertical resources, widens critical nets |
Utilisation and congestion are not the same measurement, and confusing them is the most common way a block gets to detailed routing before anyone notices it is unroutable. Utilisation is how much of the placement area is occupied by cells. Congestion is whether the router has enough track resource to connect them. A block at 60 percent utilisation can be badly congested in one corner while a block at 85 percent routes cleanly — because congestion is about the local demand for routing tracks, not the average density of cells.
This is why the router cannot rescue a congested design. By the time global and detailed routing run, the cells are placed and the tracks are fixed; the router allocates what exists. If placement has packed a region of high-fanout logic into one gcell, or a timing-driven pass has pulled a critical cone together, the demand for tracks in that region exceeds supply and no amount of routing effort recovers it. The honest fixes are upstream: change the floorplan, relax utilisation locally, pad the cells, or restructure the nets that concentrate demand.
The schedule risk is that congestion is cheap to detect early and expensive to detect late. A global-route congestion map is available minutes after placement; a detailed-route DRC explosion arrives after hours of routing and forces an iteration back through placement. Reading the global-route map early — and treating a hotspot as a placement instruction rather than a routing nuisance — is what keeps a block off the critical path.
Only marginally. The router allocates tracks in the space placement left it; if cells are packed into a region or a macro channel is too narrow, no routing effort recovers it. Congestion is closed in floorplan and placement, and the router's job is to confirm the result rather than create it.
There is no universal number. 65 to 80 percent is a common working range, but a block dense in high-fanout logic or with tight macro channels can congest well below that, while a clean datapath routes above it. Utilisation is an input to congestion, not a measure of it.
It can. Pulling cells together to close a critical path concentrates pin and wire demand, and aggressive timing-driven placement is a common source of local hotspots. The balance is congestion-aware timing-driven placement, not one bought at the expense of the other.
Written for engineers. Share it with one.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.