Electromagnetic Simulation: Design Rules and Process Limits
Electromagnetic simulation entered the PCB flow for one practical reason: the behaviour that breaks a board, coupling, reflections and resonances, is invisible on a schematic and expensive to discover on fabricated copper. A simulator lets a designer ask what the fields will do before the panel is etched, and it answers in minutes rather than in a respin. That is the whole value proposition, and it explains why teams that once treated simulation as a luxury now treat it as part of layout.
What Simulation Actually Answers
Simulation is not a verdict on a design. It is an answer to a specific question: where does this trace couple into that neighbour, how far does the return current spread under a plane split, what does the input impedance of the power distribution network look like at 200 MHz. Each question has a model behind it, and the model is only as trustworthy as the geometry, the stackup and the material data that were fed into it.
Teams therefore get the most from electromagnetic simulation when they write the question down before opening the tool. A study with no question produces colourful field plots and no decision, and a decision that is deferred leaves design margin to guesswork. Two hours spent narrowing the question usually saves a day of model building that answers nothing in particular.
Signal Integrity Before the Layout Is Finished
The earliest useful application is at the interface level. A fast bus, a differential pair or a clock net can be modelled as a transmission line with a driver, a load and a set of discontinuities, and the simulator will show how much of the eye remains at the receiver. Our notes on microstrip and stripline routing explain why the same geometry behaves differently on an outer layer and an inner layer, which is exactly the distinction the model has to capture.
What simulation adds beyond rules of thumb is the interaction between effects. Crosstalk, reflections and loss do not simply add; they can reinforce one another at particular lengths and data patterns. A model that includes the neighbouring traces and the real stackup can reveal a resonant length that no single rule would have predicted.
Power Integrity and the Plane Question
Power integrity follows the same logic on a different network. The question is whether the impedance between the supply pins and the reference stays low across the band where the devices draw transient current. Decoupling capacitors, their mounting inductance, plane spreading inductance and plane resonances all contribute, and the result is rarely obvious from a bill of materials.
Simulation here is often cheaper than measurement, because the failure mode, a plane resonance excited by a switching current, is difficult to reproduce on a bench without a great deal of probing. Our notes on EMI suppression design principles describe the loop-area reasoning that underlies the same problem in the radiated domain.
Where Prototype Iteration Hurts
The alternative to simulation is prototype iteration: build a board, measure it, guess at the cause, cut the guess and build again. Each cycle costs a fabrication and assembly round, a schedule slot and an engineer’s attention. The method works, and it remains the right method for problems that no model describes well, but it is a poor fit for effects that depend on geometry in a way that is tedious to iterate.
The economics change with layer count and material. A two-layer prototype that can be reworked by hand tolerates iteration; a fourteen-layer board with controlled impedance and blind vias does not, because a change to the stackup invalidates every impedance target on it. That is the point at which simulation stops being optional and becomes the cheaper path.
Keeping Models Honest
A model that has never been checked against a measurement is a hypothesis. The discipline is to correlate once, on a representative structure, and then trust the model for the variants. A test coupon on the same panel as the product is the cheapest way to do this, and the correlation data stays useful for the next project. Our notes on layout verification methods cover where measurement and simulation each do the work.
Material data is the usual source of disagreement. Permittivity and loss tangent vary with frequency, with resin content and with the weave, and a nominal value from a datasheet can be a few percent away from the local value under a given trace. When the model and the board disagree, the material is the first place to look and the last place engineers search.
Choosing What to Simulate
Simulating everything is not a strategy. The interfaces worth modelling are the ones with the least margin: the fastest link, the most constrained routing channel, the power rail with the largest transient, and the analogue front end that has to resolve small signals next to switching logic. Those four studies typically cover most of the risk on a mixed design.
It also helps to decide in advance what a result would change. If a simulation outcome cannot change a decision, it is documentation rather than analysis. Framing the study that way keeps the effort proportional to the risk and keeps the schedule honest when a deadline is close.
Working With the Fabricator
Simulation and fabrication meet at the stackup. Line width, dielectric thickness and copper weight are the same variables the fabricator controls, and their tolerances are the same ones the model assumes. A model built on ideal values will disagree with a board built to the edge of a process window, so the process capability belongs in the simulation input rather than in a footnote.
That is where a fabricator who can explain capability is worth more than one who simply quotes a price. At gopcb the stackup discussion happens before the impedance study is frozen, so the targets on the drawing match what the process can hold rather than what a nominal calculator produced.
Where Simulation Time and Cost Actually Go
The cost of simulation is usually misjudged. The licence is rarely the largest item; the expensive parts are building a model that matches the real geometry, obtaining credible material data and validating the result once. After that first study, the marginal cost of the second question on the same board is small, which is why teams that simulate one interface tend to end up simulating four before the project closes.
Turnaround matters as much as accuracy. A coarse model that answers in an hour will be used while layout is still open, while a very detailed model that takes a week will be read after the design is frozen, when nothing can be changed. Choosing the mesh and the solver to fit the schedule is a legitimate engineering decision rather than a compromise of principle.
The habit that pays is to record what the model assumed: the stackup revision, the material data source, the frequency range and the boundary conditions. Six months later, when a board behaves differently from the study, that record is what tells a team whether the physics changed or the design did.
FAQ
Does simulation replace measurement? No. It reduces the number of prototypes needed to find a problem, and measurement confirms that the model and the board agree. A flow that uses only one of the two will eventually be surprised by the other.
Can a small team afford it? Modern tools run on ordinary workstations, and a single interface study takes hours rather than weeks. The barrier is usually the model library and the discipline to keep geometry current, not the licence.
What if the model cannot be trusted? Model the structure you can measure, correlate on a test coupon, and use rules of thumb with generous margin elsewhere. Partial confidence, clearly labelled, is more useful than none.
