Analysing a full chip flat is slow and noisy, so teams analyse blocks first and integrate the results. That only works if each block exposes everything the integrator needs at its boundary. A block that is clean internally but silent about its reset assumptions pushes its risk to the top level, where it is hardest to fix.
Boundary item | Why integration needs it |
|---|---|
Reset ports with polarity and style | The integrator must know which top-level reset each port joins and whether release is already synchronised inside. |
Assumptions on each reset input | Minimum assertion time, required release order, and which clock the release is synchronised to. |
Guarantees on each reset output | What the block drives while in reset and how it behaves at release, so neighbours can rely on it. |
Internal disposition summary | Clean, or clean except for named findings with dispositions the integrator must honour. |
Clock and reset domain map | Which internal domains each boundary signal Belongs to, so inter-block crossings are visible. |
Integration analysis covers what no block can see alone: reset crossings between blocks, the top-level reset tree from its sources to every block port, synchroniser placement across the boundary, and whether each block assumption is met by its actual connection. A block that assumes a synchronised release connected to a raw reset source is an integration defect, even though both sides looked clean alone.
Structural checks at the top level read the boundary models, not the block internals, so the review first audits the models: every block port mapped, every assumption stated, nothing defaulted. Then it reads the inter-block findings the same way as block findings: each one fixed, accepted with proof, or waived.
The block verdict stands for the block as specified, but integration fails: the assumption is violated at the connection. Fix the connection with a synchroniser or change the block assumption and re-analyse.
Because a synchroniser inside one block may protect only that block. A crossing whose source and destination sit in different blocks can still release into an unprotected flop at the top level.