Column: Silicon systems design
Figure 1: Local coverage closure can be achieved while system-level behaviours remain unvalidated
Coverage metrics tend to assume that
if each part behaves correctly in isolation, the integrated system will behave correctly as a whole. In practice, this assumption breaks down. Timing relationships’ span dies. Power and thermal eff ects interact dynamically. Soſt ware scheduling exposes ordering dependencies that were invisible during RTL-centric verifi cation. As a result, programmes can report excellent verifi cation health while still harbouring latent integration risks. T ese risks oſt en surface late, during system bring-up, workload validation, or even post-silicon operation. When they do, teams are forced to confront an uncomfortable reality: the metrics were not wrong, but they were incomplete.
The illusion of metric-driven assurance One of the most persistent challenges in large programmes is the false sense of security that stable metrics can create. Regression pass rates improve, coverage reaches a limit and defect discovery slows. On paper, everything looks under control. Yet, confi dence does not rise in parallel. T is phenomenon is not psychological; it is structural. Verifi cation metrics largely
measure what was asked rather than what emerged. T ey confi rm that the verifi cation plan was executed, not that the system behaved robustly under realistic, cross- domain conditions. In complex systems, the most
consequential failures are rarely those that violate explicit assertions. T ey are emergent behaviours that arise from interactions between correct components operating under stress, concurrency, or a typical sequencing. Figure 2 shows how defects or mismatches that are benign at the component level can propagate across interfaces and only become observable at the system boundary.
System-level confi dence is hard to quantify System-level behaviour resists simple measurement because it is shaped by interactions, not features. Traditional metrics struggle to express properties such as resilience, stability under load, or tolerance to unexpected sequencing. T ese qualities are critical to system confi dence, yet they sit outside the scope of most verifi cation scorecards. Observability also degrades as systems scale. Signals are abstracted, compressed,
Figure 2: Local correctness doesn’t prevent system- level failure propagation
www.electronicsworld.co.uk April 2026 13
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48
