Compliance
What the bench controls before measuring
Every compliance number is a product of its measurement conditions, and the first act of bench literacy is naming them.
What the bench must control first
A compliance measurement does not begin with the instrument. It begins with the fixture, the cables, the probe points and the conditions the measurement claims to describe, because at multi-gigabit rates the bench is part of the channel. This page lists what has to be controlled before a number means anything.
The channel being measured includes everything between the transmitter and the decision point, and the bench adds to it. Test fixtures contribute their own traces, connectors contribute their own discontinuities, and the oscilloscope sees the sum, not the device under test alone. The discipline of the bench is to make the additions known and small: fixtures with characterized loss, cables with documented behavior, de-embedding that removes the fixture's own signature from the result. A measurement without that bookkeeping is a photograph of the bench, not of the device.
The same discipline exists wherever faint things are measured against a background. The amateur astronomer's rule that the instrument is the last thing to check after the site, the sky and the observer's own adaptation is written down in the observing fieldbook, whose checklists for a night's equipment read like a fixture log: every part of the chain named before the looking starts. At a compliance bench the telescope is the least of it; the sky conditions are the channel.
The fixture as part of the channel
Every test point is an interruption. Breaking a link to reach a probe point adds a connector pair and a stub; measuring at the far end of a fixture measures the fixture's loss on top of the device's. The published compliance methodology acknowledges this by defining measurement points precisely: the transition between what is the device and what is the bench is specified, not assumed. Where the specification cannot reach inside a package, it defines the point at which the claim is made and lets the bench show its work.
The eye that results from this chain is read on the page about reading a receiver eye, and the program that formalizes the whole procedure is described in how a compliance program works. Between them sits this page's claim: the figure is only as good as the control the bench exercised before it.
What does the instrument itself contribute?
Bandwidth, noise and jitter of its own. An oscilloscope that is too slow closes the eye it is measuring; a probing arrangement with its own jitter adds to the jitter it reports. The rule of the bench is that the instrument's contribution must be small against the quantity being measured, and that what cannot be made small must be characterized and removed. The literature on the measured object, summarized in the signal integrity article and the jitter article, read on September 6, 2026, names the contributors: deterministic and random components at the device, deterministic and random components at the instrument, and a result that separates them only by bookkeeping.
| Element | What it adds | How the bench controls it |
|---|---|---|
| Fixture traces | Loss and discontinuity | Characterized, de-embedded |
| Cables | Loss and skew | Documented, kept short |
| Probe point | Stub and load | Defined by the methodology |
| Instrument | Bandwidth limit, own jitter | Rated well above the signal |
| Environment | Temperature, supply noise | Recorded, not assumed |
Why do two benches disagree?
Because they control different things. Two benches measuring the same device can produce different eyes without either being wrong: different fixtures, different probe points, different de-embedding, and the residual difference is the sum of the things each bench did not control. This is why a compliance program exists at all rather than a single good measurement. The program standardizes the bench so the figures can be compared, which is the entire content of the word compliance.
Corrections propagate the same way. A BER figure measured at the wrong probe point is a measurement of a different channel, and a margin read after careless de-embedding belongs to the fixture, not the part. The habit of asking where the probe sat is the habit of reading the measurement rather than the number.
Checks before trusting a bench figure
- Name the probe point: the claim is made at a defined place in the chain, and the place is part of the claim.
- Ask what was de-embedded: the fixture's own loss must be removed, not absorbed into the device's result.
- Rate the instrument against the signal: bandwidth and jitter overhead belong to the bench, not to the part.
- Record the environment: temperature and supply noise are variables, not constants.
Common mistakes
- Treating the instrument as transparent. At these rates the bench is part of the channel.
- Comparing figures from different probe points as if they were the same measurement.
- De-embedding a fixture that was never characterized: removal without measurement is invention.
- Reading a closed eye as a failed device before asking what the fixture closed.
The bench controls first, measures second, and reports third, and a figure that arrives without the first step has skipped the only step that makes it a measurement. The number is the residue of the control, and where the control is silent, the number is a photograph of the bench.
Before any measurement is trusted, the bench fixes what the link is allowed to do: the negotiated width and rate, the equalization phase reached, the state of the receiver detect and polling handshakes, and the error counters that accumulate once the link is up. Those settings define the baseline against which later captures are compared, so a change in the numbers can be traced to the link rather than to the instrument. When a link refuses to leave the training states, the same baseline becomes the starting point for diagnosis, and the counters that matter are read alongside the LTSSM transitions recorded at each step.
The page behind these facts
The nature of jitter components, signal margin and the eye as a measurement object are summarized in the jitter article, the signal integrity article and the eye pattern article, all read on September 6, 2026. The compliance context in which these controls are exercised is the program described by this desk on the pages linked above.