Closing the Gap Between Simulation and Measurement

A simulation that disagrees with a measurement is a common and useful situation. The disagreement contains information about the model, and the work is to find which assumption is responsible rather than to adjust parameters until the curves match.

Why They Disagree

The model uses a stack-up with nominal dimensions, and the fabricated board has dimensions within a tolerance. A dielectric thickness that is ten per cent thinner than nominal changes the impedance and the propagation velocity.

The model uses material properties measured at one frequency, and the actual property varies with frequency. The dissipation factor in particular is often specified at one point and used across a range.

The model treats the copper as smooth, and the fabricated trace has a surface roughness that adds loss at high frequency. This is one of the largest sources of disagreement in a long high speed channel. Our high speed design rules notes describe the parameters that matter.

Establishing What the Board Actually Is

The first step is to measure the fabricated geometry rather than to assume it. A cross section of a coupon gives the dielectric thickness, the copper thickness and the trace width at the actual location.

A coupon built on the same panel carries the same material and the same process, which makes it the right sample for the measurement. Our coupon notes describe the structures that make this possible.

The impedance measurement and the cross section together give the geometry that produced the measured impedance, which is the input the model should have used.

Separating the Contributors

Copper loss and dielectric loss have different frequency dependence, so a measurement over a range of frequencies can separate them. The dielectric loss rises linearly with frequency, while the conductor loss rises approximately with the square root for a smooth conductor.

Comparing two coupons that differ in one parameter is the cleanest experiment. Two lengths of the same line isolate the loss per unit length, and two coupons with different roughness isolate the roughness contribution.

The insertion loss measurement should be made on lines of at least two lengths, because the connectors and launches contribute a fixed loss that is otherwise attributed to the line.

Model Calibration

Where the model is fitted to a measurement, the fitting parameter should correspond to a physical quantity that can be checked. Fitting a dielectric constant alone is defensible; fitting several parameters at once produces a model that fits one measurement and predicts nothing.

The roughness correction is the parameter most often fitted, and it should be constrained to the range of the actual foil. A fitted roughness that is physically impossible indicates that something else is wrong.

The calibrated model should then be validated against a different structure on the same board, not against the one it was fitted to. Our quality notes describe how the measurements are recorded.

Tolerances and the Design Margin

The simulation should be run at the corners of the fabrication tolerance rather than at nominal, because the design must work at the worst case.

The dielectric thickness, the trace width and the dielectric constant all vary, and their combination determines the impedance range. A design that works only at nominal is a design that will fail on some panels.

The measurement data from a coupon supports the tolerance assumption, since the actual spread across the panel can be measured and used rather than guessed. Our design release checklist notes where the tolerance assumption is recorded.

When the Measurement Is Wrong

The measurement setup is a source of error as well. The launch, the connectors and the calibration of the network analyzer all affect the result, and a poorly calibrated measurement can disagree with a good model.

The verification is to measure a known standard, such as a length of coaxial cable with a known loss, on the same setup. The measured value should match the specification.

Time domain reflectometry shows the launch as an impedance discontinuity, and its size is an indication of how much the measurement is affected by the fixture.

Using the Result

The purpose of the correlation is to make the next design predictable. A model calibrated on one product and confirmed on another allows the simulation to be trusted for the third.

The record should include the model parameters, the measurement conditions and the structure used for validation, so that the calibration can be repeated or revised later.

Where the correlation cannot be achieved, the honest conclusion is that the design needs more margin, and the margin is cheaper than the investigation once the schedule is short.

Process Control and Verification

On a design of this kind, measurement is the item that decides how the rest of the board is arranged. The measurements that matter are the repeatable ones: conductor width and spacing, annular ring, finished hole size, plating thickness and surface finish are all verifiable on a coupon that travels with the panel. Running a first article through the same checks as the production panel confirms that the two agree, and that comparison is the cheapest form of process control available at prototype stage.

Keeping a sample from the panel turns a dispute into a measurement, because the same coupon can be re-examined by both parties without rebuilding the batch. Reviewing the design before the data is released is cheaper than correcting it after the panel is in the tank, because every step downstream inherits the decision made at the front end.

Documenting the assumption is part of the design work, and a short note on the drawing prevents a question that would otherwise arrive a day later and cost a day of schedule.

Process Control and Verification

On a design of this kind, measurement is the item that decides how the rest of the board is arranged. The measurements that matter are the repeatable ones: conductor width and spacing, annular ring, finished hole size, plating thickness and surface finish are all verifiable on a coupon that travels with the panel. Running a first article through the same checks as the production panel confirms that the two agree, and that comparison is the cheapest form of process control available at prototype stage.

Keeping a sample from the panel turns a dispute into a measurement, because the same coupon can be re-examined by both parties without rebuilding the batch.

Process Control and Verification

On a design of this kind, measurement is the item that decides how the rest of the board is arranged. The measurements that matter are the repeatable ones: conductor width and spacing, annular ring, finished hole size, plating thickness and surface finish are all verifiable on a coupon that travels with the panel. Running a first article through the same checks as the production panel confirms that the two agree, and that comparison is the cheapest form of process control available at prototype stage.

FAQ

Should the model be adjusted to match the measurement? Only where the adjusted parameter corresponds to a physical quantity that can be verified on the board.

What is the largest source of disagreement? Cupper roughness and the fabricated dielectric thickness, both of which can be measured and both of which are often assumed.

What does gopcb provide for correlation? We provide coupons on the production panel, cross sections of the measured structures, impedance and insertion loss measurements over frequency, and the data in a form that can be compared with the simulation. Where the design has margin concerns, we measure before the product is committed.