SIPI

Package & board / 03

BGA Escape Routing and Breakout Congestion

Getting a few hundred signals out from under a BGA is a geometry problem before it is an electrical one — and its answer sets your layer count, your board thickness, and therefore your via stubs. It is also the part of the channel with the worst impedance control and the most damaged return path.

Pitch decides everything downstream
0.8 mm pitchone trace fits between balls 0.65 mm pitchnothing fits — via-in-pad rings escape roughly one per signal layer — which is why ball count drives board layer count and why the first three millimetres contain most of a channel's discontinuities
The question is how many traces fit between two adjacent balls — pitch minus antipad diameter, divided by the line-and-space rule the fabricator can hold. At 0.8 mm there is usually room for one, so each layer escapes roughly one more ring. At 0.65 mm and below there is room for none, the dogbone fanout stops fitting, and via-in-pad becomes mandatory — filled and plated vias, a real cost adder decided by a pitch chosen for mechanical reasons. Rings escape about one per signal layer, so a large array needs several layers just to break out before any routing has happened. That is why package ball count drives board layer count more directly than the routing does. It is also where the discontinuities are. Within the first few millimetres a signal passes a pad, a via, an antipad in every plane it crosses, a trace at whatever impedance fits, and usually a layer transition — packed close enough to interact. A 3D solve of that region is worth more than a finer model of the long uniform trace that follows.

A modern System-on-Chip (SoC), high-density FPGA, or server processor packs thousands of electrical pins into a Ball Grid Array (BGA) soldered beneath a single monolithic package. On the outermost perimeter rows of the array, routing is straightforward: traces simply fan out across the top surface layer directly into open board real estate. But once you move into the interior rows of a 1,500-pin or 2,500-pin package, hundreds of high-speed signals become physically trapped behind a dense forest of surrounding solder balls.

To escape the center of the array, every interior signal must plunge vertically into the board through a breakout via, transition to an inner routing layer, and squeeze through narrow corridors between adjacent solder balls and via antipads. This high-density routing puzzle — the BGA escape routing or "fanout" — represents the most congested geometric bottleneck on any printed circuit board.

Electrically, the BGA breakout is also the single most hostile discontinuity in the entire interconnect path. Within a span of only two to three millimetres, a delicate multi-gigabit signal encounters an SMD pad, a neck-down trace impedance drop, a vertical via transition, reference plane perforations, and dangling via stubs. How you architect this escape — balancing ball pitch, layer count, dogbone vs. via-in-pad technology, and ground return via placement — directly decides whether your board requires 10 layers or 20, and whether your high-speed channels survive the first fractional inch of routing.

The geometry that sets everything else

Under a ball grid array, signals have to travel from balls in the middle of the array out to the edge. Each one needs a path, and the paths are constrained by a simple question: how many traces fit between two adjacent balls?

That number comes from the ball pitch, minus the antipad diameter the vias need, divided by the line-and-space rule the fabricator can hold. At 0.8 mm pitch with conventional rules there is usually room for one trace between balls, so each routing layer can escape roughly one more ring of the array. At 0.65 mm and below there is often room for none.

Everything downstream follows. Rings divided by traces-per-channel gives layers; layers gives thickness; thickness gives via stub length. A pitch chosen for mechanical reasons decides whether your channel has a resonance in band.

Pitch decides everything downstream

The question is how many traces fit between two adjacent balls. That comes down to the ball pitch minus the antipad diameter, divided by the line-and-space rule the fabricator can hold.

At 0.8 mm pitch with conventional rules there is usually room for one trace between balls, so each layer can escape roughly one additional ring of the array. At 0.65 mm and below there is often room for none, and the dogbone fanout — a short trace from the pad to a via in the space between balls — stops fitting at all. That forces via-in-pad, which means filled and plated vias, which is a real cost adder decided by a pitch chosen for mechanical reasons.

How this sets layer count Rings escape roughly one per signal layer, so a large array needs several signal layers just to break out, before any routing between components has happened. This is why package ball count drives board layer count more directly than the routing does — and why a ball map change late in a programme can add layers to a board.

Why the escape is the worst part of the channel

Within the first few millimetres a high-speed signal passes through a pad, a via, an antipad in every plane it crosses, a short trace at whatever impedance the escape geometry allows, and usually a layer transition. Each is a discontinuity, and they are packed close enough together to interact.

The practical consequence is about where to spend simulation effort. A 3D field solve of the escape region — including the vias, the antipads and the surrounding ground vias — is usually worth more than a more refined model of the long uniform trace that follows it, because the trace is well described by a 2D solve and the escape is not.

Why this decides how your board behaves

What to do about it

Count the escape before committing the stackup. Rings, traces per channel, layers — the arithmetic takes minutes and it is the input the stackup needs.

Budget the breakout explicitly. It is the row people forget in a channel budget, and it is a decibel or two, which is more than an inch of reach.

Keep differential pairs together and symmetric through the field. This is where pairs get split around obstacles and where one half acquires a neighbour the other does not. Asymmetry here converts to common mode that nothing downstream removes.

Provide ground vias inside the field, not just around it. The return current for signals escaping the middle has to get to a plane somewhere, and the middle is where the perforation is worst.

Using this to find a fault
  • Crosstalk worst on the innermost signals. Expected — they run furthest through the densest part of the field. If it is worst on the outer ones instead, look at the plane edge.
  • Mode conversion appearing on pairs that are length-matched. Check whether the pair stayed together and symmetric through the breakout.
  • A channel worse than its trace length explains, by a fixed amount. Discrete structures — the escape and the launch are the usual candidates.
  • Return-path problems that a plane-layer view does not show. View the antipads alone. What looks like a solid plane in the layer view can be mostly holes where the field is.
Go deeper — the escape arithmetic, and via-in-pad

The usable width between two adjacent balls is the pitch minus the antipad diameter. Divide by the minimum line-plus-space and you get traces per channel:

n = ( pitch − antipad ) / ( line + space ) and layers ≈ rings to escape ÷ n, which is the number the stackup needs

Notice that the antipad appears in that expression, which sets up a direct conflict. A smaller antipad frees routing width and raises the via's capacitance, which lowers its impedance and makes it a worse discontinuity. A larger one improves the via and leaves less room to route, forcing more layers. The escape and the via design are the same decision.

Via-in-pad. At fine pitch the dogbone fanout — a short trace from the pad to a via in the gap between balls — stops fitting. The remedy is to put the via directly in the pad, which requires the via to be filled and plated over so solder does not wick down it. That is a real cost adder, and it is triggered by a pitch chosen on mechanical grounds.

It does have an electrical benefit worth noting: via-in-pad removes the dogbone stub, which at fine pitch and high frequency is itself a discontinuity. So the cost buys something, which is a useful argument to have available.

And the reason the escape is short. Everything above is bad, and it is all confined to a few hundred mils. Loss scales with length, so a short bad section costs much less than the same impedance error over inches. The things that do not scale with length — the return path damage and the mode conversion — are the ones worth the attention.

In the real world

The escape is where mechanical, cost and electrical constraints collide most directly, and it is usually resolved by whoever is holding the schedule. Ball pitch comes from the package, layer count comes from cost, line and space come from the fabricator, and the signal integrity consequences are what is left over.

Which makes the escape arithmetic worth doing early and sharing widely. It converts a qualitative worry into a layer count, and a layer count is a number the rest of the project already knows how to argue about.

Related