Functional Verification Interview Questions With Answers

Design verification interviews check three things: whether you can plan what to verify, whether you know the SystemVerilog and UVM tools for it, and whether you think like someone trying to break the design. Here are fifteen common questions with the answers interviewers look for. Each links to the lesson or project on this site where the idea is run in a simulator.

The verification loopWrite a plan from the specification, build the testbench, run regressions, read the coverage, and update the plan or tests where coverage shows gaps, until the plan is closed.Test planTestbenchRegressionsCoverageclose gaps
Verification is a loop: coverage tells you what the regressions haven't reached yet.

Planning

  • Q1. What goes in a verification plan? Answer: the features from the specification, how each is checked (scoreboard, assertion, directed test), the coverage that shows it was exercised, and what counts as done.
  • Q2. What is the difference between code coverage and functional coverage? Answer: code coverage is measured automatically from the RTL (lines, branches, toggles, FSM states); functional coverage is written by you from the plan (covergroups, cover properties). 100% code coverage can still miss a feature nobody implemented.
  • Q3. When are you done verifying? Answer: when the plan's functional coverage is closed, code coverage holes are explained, regressions pass across many seeds, and the bug rate has flattened. Never just 'when the tests pass'.

Stimulus and checking

  • Q4. Why constrained random instead of directed tests? Answer: random stimulus reaches combinations nobody thought to write; constraints keep it legal and steer it toward interesting cases. Directed tests still cover specific corner cases and bring-up. Lesson 3
  • Q5. What does a scoreboard do? Answer: it compares what the DUT produced, as seen by a monitor, with what a reference model predicts, independent of how the stimulus was generated. The APB project
  • Q6. Why should the monitor be separate from the driver? Answer: the monitor reports what really happened on the pins, so checking works for any stimulus, including another team's driver, and can be reused at the next level up.
  • Q7. A test passes with 100% coverage. Can there still be a bug? Answer: yes. Coverage shows what was exercised, not whether it was checked; the APB project's planted bug passed 100% coverage and was caught by the scoreboard.

Assertions and coverage

  • Q8. Immediate or concurrent assertion? Answer: an immediate assertion checks a condition once, where it is written; a concurrent assertion is clocked and checks a sequence over time on every cycle. Lesson 5
  • Q9. What is the difference between |-> and |=>? Answer: |-> checks the consequent in the same cycle as the antecedent; |=> checks it one cycle later.
  • Q10. What is a cross, and why can it explode? Answer: a cross records combinations of two or more coverpoints. Its size is the product of their bins: 5 operations x 3 operand kinds is 15, but 16 x 16 x 4 is 1,024. Cross only what the plan needs and ignore impossible bins. Lesson 4

UVM

  • Q11. Name the main UVM components. Answer: sequence item, sequence, sequencer, driver, monitor, agent, scoreboard, environment, test. Lesson 6
  • Q12. Why create components with type_id::create instead of new? Answer: the factory can then substitute a derived type from the test, without editing the environment.
  • Q13. What ends the run phase? Answer: when every raised objection has been dropped.

Debugging

  • Q14. A random test fails once in a thousand seeds. What do you do? Answer: rerun that seed to reproduce it, find the first point where the DUT and model disagree, and work backward in the waveform; then decide whether the DUT, the model or the specification is wrong.
  • Q15. How do you know your testbench can find bugs at all? Answer: plant one. Lessons 2, 5 and 7 and the APB project each break the design on purpose and check that a test fails.

How to answer well

Tie each answer to an example you have run: 'in my FIFO testbench, coverage showed the full bin was never hit until I biased the traffic' is far stronger than a definition. Work through the verification course on this site and you will have one for nearly every question here.

Related

Sources

Last verified

Written on 17 November 2026; every example linked was run in Vivado xsim 2025.2.