NoC/AXI policy choices trade fairness, throughput, and determinism.
+------------------+----------------+----------------+----------------+
| Approach | Strength | Weakness | Best when |
+------------------+----------------+----------------+----------------+
| Strict | predictable | slower early | new program |
| Balanced | practical | needs rigor | stable teams |
| Aggressive | fast | escape risk | late schedule |
| Re-architect | durable | expensive | recurring failures |
+------------------+----------------+----------------+----------------+Pick the approach based on residual silicon risk, not only schedule pressure.
Policy-only arguments without artifact evidence
Unbounded waivers
INTEGRATION EVIDENCE MATRIX — Coherency & Ordering Integration
+-------------------+------------------------+--------------------------+-------------------------+
| Evidence | Tells you | Does not prove | Next action |
+-------------------+------------------------+--------------------------+-------------------------+
| Manifest diff | baseline changed | root cause itself | isolate first delta |
| Domain trace | where behavior diverged| ownership accountability | map to contract owner |
| Signoff dashboard | closure posture | reproducibility quality | verify source artifacts |
| Bring-up logs | software-visible symptom| physical/timing cause | correlate with HW trace |
| Waiver register | accepted residual risk | technical closure | review expiry criteria |
+-------------------+------------------------+--------------------------+-------------------------+Fabric correctness and contention behavior must be proven together.
FABRIC FLOW
masters -> routers/VCs -> slaves + memoryLATENCY TAIL
p50 ███
p95 ██████
p99 ██████████NoC contention heatmap
ordering violation report
QoS fairness summary
protocol trace
Bandwidth looked fine at average load, but p99 tail violated SLA due to arbitration starvation.
Isolate traffic class
Check ordering assumptions
Audit arbitration policy
Ask: what baseline, owner, and artifact prove this topic is truly closed?
State baseline manifest and owner with every closure metric.
Run cross-domain regression after every top-level fix.
Comparing results across different manifests.
Unowned issues slipping through review cycles.
Waiving risks without expiry and validation plan.
Coherency correctness requires consistent shareability, home-node behavior, and ordering barriers across CPU, DMA, and accelerator masters.
Metric: stale-data escapes, coherency retry rate, fence latency
This content block can't be displayed here.
Block type: unknown
This content block can't be displayed here.
Block type: unknown
This content block can't be displayed here.
Block type: unknown