BGA Escape Routing Design Guide

BGA escape routing is the work that a ball grid array forces on the designer before any functional trace can be drawn. The package hides its connections underneath, so every signal has to be routed out from the inside of the array to the rest of the board. That fanout is the escape routing, and it usually decides the layer count of the whole design. A ball grid array with a fine pitch can consume four or six layers before a single functional trace is routed.

What Escape Routing Means

The escape routing is the set of traces and vias that connect each ball to the routing channels outside the package footprint. It is a routing problem with a fixed number of exits and a limited space.

The number of balls on the perimeter of the array sets how many signals can leave on the outer layer. A row of balls in the middle of the array has to travel further, and the space for that travel has to come from the layers below.

The pattern is therefore a function of the ball pitch, the ball count and the number of layers available. Reducing the pitch does not reduce the difficulty, because the pads and the channels shrink with it.

A useful starting point is to count the rows rather than the balls, since the routing difficulty is set by how deep a ball sits inside the array. A device with twenty rows is far harder to escape than one with ten rows and the same ball count.

The result appears in the layer count decision, which is often made by the fanout rather than by the number of signal nets.

Dog Bone Fanout

A dog bone fanout places a via beside each ball pad and connects the two with a short trace, so that the shape resembles a dog bone. The via takes the signal to an inner layer where there is room to route.

The method is cheap because the vias are ordinary through holes, and it is the standard choice for a pitch of about 0.8 mm and above. The trace between the pad and the via is short and its impedance is not controlled.

The cost is space. Each via needs a pad and a clearance, so the dog bone needs an area beside every ball, which limits the density that can be achieved.

The trace from the pad to the via adds a small stub, and on a fast signal that stub becomes part of the transition. At a very high data rate, a via in pad becomes the better choice.

Via in Pad Fanout

A via in pad places the via directly under the ball, which removes the short trace and the space that it needed. The ball is soldered over the top of the filled via.

The via has to be filled and plated over so that the solder does not wick into the barrel during reflow. The process is more expensive and it has to be done by a shop that is experienced with it.

The benefit is a much denser fanout, which can save layers on a large array. The electrical transition is also cleaner, because there is no stub.

The construction and its limits are described in the via construction guide, and the choice between the two methods is usually made on the layer count and on the assembly capability.

Layer Count and Routing Channels

The routing channels between the balls limit how many traces can pass through a row. The channel width is the distance between two pads minus the clearance, and the number of traces that fit is the integer part of that division.

A pitch of 0.8 mm typically allows one trace between two balls on the outer layers, while a pitch of 0.5 mm allows none. That single fact drives the fanout to the inner layers and the layer count upwards.

The usual approach is to route the outermost rows on the surface, the next rows on the first inner layer, and so on, working inward. The number of layers required is then the number of rows divided by the number that each layer can serve.

The calculation should be done before the placement is fixed, because a small change in the ball count or in the package can change the layer count and the cost of the whole board.

Power and Ground in the Array

A modern ball grid array has a large number of power and ground balls, and they have to be connected to planes rather than routed as signals. The planes provide the return path and the current capacity.

The ground balls should be connected with a via directly to the ground plane, and the power balls to the appropriate supply plane. A power ball that is routed as a signal trace adds inductance to the supply that the device cannot tolerate.

The plane split under the array is a common source of noise, because a signal that crosses a split has no return path beneath it. The planes should be arranged so that the region under the array is continuous for the supplies that the device uses.

The decoupling capacitors belong on the opposite side of the board directly under the array where the space allows, and the vias that connect them should be as short as possible.

Design Rules and Fabrication

The pad size for the balls comes from the package drawing, and the mask and the paste aperture have to be designed with the assembly process rather than with the artwork alone.

The via size in a dog bone fanout is limited by the drill rules and by the space between the pads. A via that is too large cannot fit, and one that is too small has a high aspect ratio in a thick board.

The fabrication notes should state the via construction, the fill requirement for a via in pad and the surface finish, because those choices affect the assembly yield more than any other part of the design.

The inspection of a ball grid array is done by X ray, and the comparison of the methods is in the X ray and AOI guide. A reworked array is covered by the reballing material.

Practical Rules

Calculate the fanout before the placement, and choose the layer count from the ball count and the pitch rather than from the signal count.

Use a dog bone fanout where the pitch allows it, and a via in pad where the density demands it and the shop can build it.

Connect every power and ground ball with a via to a plane, and keep the planes continuous under the array.

Document the via construction and the fill requirement, and inspect the first article with X ray before the panel is released.

FAQ

How many layers does a BGA need? It depends on the ball count and the pitch. A small array can be fanned out on two layers, while a large fine pitch device can consume six or more.

What is a dog bone fanout? A via placed beside each ball pad and joined with a short trace, which is cheap but needs space beside every ball.

Why use a via in pad for a BGA? Because it removes the short trace and the space it needs, which allows a denser fanout and a cleaner electrical transition.