A DDR link that runs on the bench proves the easy case. The subsystem has to hold timing and signal integrity across voltage, temperature and process, through training and calibration, into a board Faststream did not draw — and signoff is all the conditions the bench never showed.
A DDR subsystem couples a memory controller, a physical interface and a channel that leaves the die for a board and a memory device. On the bench, at room temperature, a working link says little; the requirement is that it holds across every process, voltage and temperature corner, and over a channel with real loss and reflections.
The subsystem trains and calibrates itself at start-up to centre the data eye, but the design has to guarantee that training converges and the margin survives drift once running. Signal integrity across the channel, timing across corners and the controller's protocol correctness all have to close together.
So the work is signoff across the whole envelope: timing and signal integrity across corners, robust training and calibration, and verification of the controller against the memory protocol rather than a single tested condition.
Stated before any of them had an answer.
A link that runs at room temperature on a clean board proves the easiest case; the risk lives in the corners and the real channel.
Loss, reflections and crosstalk on the channel to the memory degrade the eye; the PHY has to close margin over a channel, not a probe.
The subsystem calibrates itself at start-up; the design has to guarantee that converges across conditions and holds as the system drifts.
A timed link that violates the memory protocol still fails; controller verification against the standard is its own closure.
The link runs on the bench; signoff asks whether it runs at every corner, over a real channel, after training, drifting — which is the whole job.
The scope of the work, rather than a capability list.
Rarely the subsystem that sounds difficult.
The eye that looks open on an ideal channel closes over a lossy one; getting margin over the real board is the signal-integrity crux.
Most of the risk is in the conditions the bench never shows; signing off across every corner is where the effort goes.
A calibration that converges usually is a field failure; guaranteeing it converges across conditions is harder than making it work once.
A link can be timed and still protocol-wrong, or protocol-correct and mistimed; closing both simultaneously is the discipline.
Timing and signal integrity held across every corner, not just on the bench.
Signal integrity signed over a real board with loss and reflections.
Calibration that reliably centres the eye and survives drift.
A controller verified against the memory standard, not a single condition.
Customer projects are presented at property, capability, outcome and integration level. Customer names, internal architecture, confidential deliverables and commercial terms are not disclosed. Where a detail would identify a customer it is omitted rather than approximated. More is available under a non-disclosure agreement, within the limits the customer has agreed.
Every item links to its own page.
Because the bench is the easiest condition: room temperature, a clean board, a typical part. A memory subsystem has to hold timing and signal integrity across every process, voltage and temperature corner, over a real channel with loss and reflections, after self-training, as the system drifts in operation. Signoff is precisely the set of conditions the bench never exercises, which is why a link that runs there proves relatively little.
At start-up the subsystem calibrates its PHY to centre the data eye — compensating for the channel and the specific devices. The risk is that training must converge reliably across all conditions and the resulting margin must survive drift; a calibration that centres the eye on the bench but fails to converge at a cold corner, or whose margin erodes as the system warms, is a field failure. Guaranteeing robust convergence is part of the design, not an assumption.
Because they are different failure modes. A link can be perfectly timed at the electrical level and still violate the memory protocol — mis-sequencing commands or mishandling refresh — and it will fail. Conversely a protocol-correct controller over a mistimed channel fails electrically. The controller is verified against the memory standard as its own closure, alongside the timing and signal-integrity signoff, because passing one does not imply the other.
Written for engineers. Share it with one.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.