Why Autorouters Don’t Work: The Mindset!

Estimated reading time: 3 minutes

Ask any group of PCB designers what they think of autorouters and the majority will say that they do not use them because they do not work. I have been battling this mindset for over 20 years now and it still persists today, even with the dramatic advances in routing technology. This way of thinking generally comes from those designers who use the entry-level tools that have limited routing capability. But even the most primitive autorouter may have some useful features. It’s all about changing that mindset of the designer and having a crack at it.

I started laying out boards back in the Bishop Graphics days where layout began with a pencil sketch, on graph paper. Then, donuts and fine black tape were stuck to clear film, at twice the actual size, to produce the required connectivity. The 12 mil tape, which we referred to as “spiderweb,” was the thinnest trace width (6 mils finished) manufacturable at that time. It was really a matter of just connecting the dots. Double-sided layouts were sometimes stuck to the same film to improve registration, using red and blue colors to photographically distinguish the layers. But routing has come a long way since then.

The first computer-based PCB design tools that emerged in the late 1970s were grid-based, ran on DOS or UNIX operating systems, and were very basic. Again it was still just connecting the dots, with a graphic trace from point-to-point to build up the layout, and then drawing the circuit on an XY plotter. Basic, but it was effective for the construction on single- and double-sided boards. The next step was to include a netlist for connectivity and then to draw the schematic graphically and extract the netlist to the PCB database. This improved database integrity dramatically.

PCB routers were developed using either the grid-based, gridless, shape-based or geometrical approaches. The first were maze and line searching routers that use an imaginary gridded workspace, while a gridless router uses a workspace with available polygon areas to accommodate the new paths. In a shape-based router, each entity on the board is represented as polygonal geometry with no reference to a specific routing grid. This enables the router to cope easily with boards in which there are SMT devices and fine-pitch BGAs with a variety of pitches and odd shapes. Also, unlike a grid-based router, a shape-based autorouter does not have to work at a particular resolution, so routing of high-density or fine-pitch boards is not significantly slower than for lower density work. Put another way, routing time depends only on the available memory, the number of objects on the boards and on the number of connections to be routed. Later, topology routers allowed designers to plan the strategy for a set of nets with attributes to define routing layers, bias and rules.

The first autorouters were not very capable, limited by computing power and lack of memory. They added too many vias, wasted space due to the strict XY bias, and the quality was poor compared to manual routing. I recall that I used to set up our Advanced Technology Designer Star router to run on the MicroVax mainframe over the weekend, only to find it 50% complete by Monday morning. However, autorouters evolved, like all technology, to include angle routes, reducing vias, push-and-shove algorithms, rip-up and retry, spreading and gloss passes. But so also has interactive routing.

Probably the most popular shape-based router, 20 years ago, was Cooper & Chyan Technology's Specctra router. The Specctra router was used by many PCB layout tools and interfaces to the router still exist today. Design constraints and routing strategies were setup in a “do file” which contained the sequence of commands. The routing was not graphically visible but the routing status was indicated and updated. Cadence’s Specctra for OrCAD is still available today.

To read this entire article, which appeared in the November 2015 issue of The PCB Design Magazine, click here.