Skip to content
STIMSMITH

Coverage Metrics

Concept

Coverage metrics are measures used to assess how thoroughly a system, model, or verification target has been exercised. In RISC-V hardware verification, functional coverage is collected by tools such as RISCV-DV and supported by riscvISACOV as a reusable coverage component. In machine-learning validation, combinatorial and neuron coverage metrics have been studied for anticipating classifier errors, identifying out-of-distribution risk, and guiding DNN test generation, though their effectiveness depends on the metric and domain.

First seen 6/12/2026
Last seen 6/14/2026
Evidence 15 chunks
Wiki v2

WIKI

Definition

Coverage metrics quantify how much of a verification or testing target has been exercised. In hardware verification, the evidence focuses on functional coverage: for example, RISCV-DV can collect functional coverage from an Instruction Set Simulator (ISS), and riscvISACOV provides a reusable functional-coverage infrastructure for RISC-V cores [1].

Coverage metrics are also studied outside hardware. In machine-learning validation, combinatorial coverage metrics have been explored as alternatives to distribution-based measures for identifying data dissimilarities that affect model performance. In deep-neural-network testing, neuron coverage metrics have been proposed by analogy with code coverage in conventional software testing.

READ FULL ARTICLE →

NEIGHBORHOOD

9 nodes · 15 edges
graph · coverage metrics · depth=1

RELATIONSHIPS

10 connections
condition coverage ← part of 100% 2e
Condition coverage is one of the coverage metrics used.
expression coverage ← part of 100% 2e
Expression coverage is one of the coverage metrics used.
toggle coverage ← part of 100% 2e
Toggle coverage is one of the coverage metrics used.
FSM coverage ← part of 100% 2e
FSM coverage is one of the coverage metrics used.
hardware fuzzing ← uses 100% 2e
Hardware fuzzing uses coverage metrics to guide input generation.
TheHuzz ← uses 100% 2e
TheHuzz uses multiple coverage metrics to guide fuzzing and detect bugs.
statement coverage ← part of 100% 2e
Statement coverage is one of the coverage metrics used.
Branch Coverage ← part of 100% 2e
Branch coverage is one of the coverage metrics used.
DeepVerifier ← uses 100% 1e
DeepVerifier is guided by coverage metrics to optimize test sequences.
DeepVerifier: Learning to Update Test Sequences for Coverage-Guided Verification ← uses 100% 1e
The paper uses coverage metrics as quantitative indicators to guide the verification process.

LINKED ENTITIES

4 links
Reinforcement Learning USES Extracted graph relationship
riscvISACOV IMPLEMENTS Extracted graph relationship
Closed-Loop Instruction Generation USES Extracted graph relationship
State Vector USES Extracted graph relationship

CITATIONS

5 sources
5 citations — click to expand
[1] RISCV-DV can collect functional coverage directly from an ISS, and its generated instructions are run on both an ISS and the DUT. [PDF] Reinforcement Learning Framework for RISC-V Functional Verification
[2] riscvISACOV provides a common functional-coverage infrastructure for RISC-V cores and is a coverage component rather than a complete verification environment. [PDF] Reinforcement Learning Framework for RISC-V Functional Verification
[3] The RISC-V RL verification thesis includes collection of coverage values, RL agent/environment implementations, reward calculation, state-vector components, instruction generation, and coverage results. [PDF] Reinforcement Learning Framework for RISC-V Functional Verification
[4] Combinatorial coverage metrics have been studied as alternatives to distribution-based metrics for OOD data, and metric learning improved SDCCMs' ability to anticipate classifier error across six open-source datasets. Metric Learning Improves the Ability of Combinatorial Coverage Metrics to Anticipate Classification Error
[5] Neuron coverage metrics and coverage-driven DNN testing methods have been proposed by analogy to code coverage, but a replication study found coverage-driven methods less effective than gradient-based methods for uncovering defects and improving robustness. Revisiting Neuron Coverage Metrics and Quality of Deep Neural Networks