Overview
A pseudorandom generator (PRG) is discussed in the supplied evidence as a stimulus-generation component in simulation-based verification of processors. The verification flow described generates stimuli using PRGs, applies those stimuli to processor inputs, and monitors achieved functional coverage to judge verification completeness. [C1]
Stimulus forms
The evidence states that PRG-generated stimuli for processor verification can take multiple forms, including:
- bit vectors applied to the input ports of the processor; and
- programs loaded directly into program memory. [C2]
This means the PRG is not tied to only one stimulus representation in the described verification setting; it can be part of workflows that exercise either low-level input ports or program-level processor behavior. [C2]
Constraint adaptation with neural feedback
The cited paper proposes a technique that dynamically alters constraints for a PRG using a recurrent neural network. In that technique, the recurrent neural network receives coverage feedback from simulation of the design under verification and uses it to influence the PRG constraints. [C3]
Reported verification effects
For demonstration, the work used processors provided by Codasip, noting that their coverage state space is reasonably large and differs across processor kinds. The authors state that the techniques are widely applicable. [C4]
The reported experimental results are that coverage closure is reached sooner and that a small set of high-coverage stimuli can be isolated for use in regression tests. [C5]
Role in the verification loop
Within the evidence, the PRG sits in a feedback-oriented verification loop:
- generate pseudorandom stimuli;
- apply stimuli to processor inputs or program memory;
- simulate the design under verification;
- measure functional coverage; and
- adjust PRG constraints using recurrent-neural-network processing of coverage feedback. [C1][C2][C3]
This positions the PRG as the controllable source of verification stimuli, while the coverage monitor and recurrent neural network provide feedback for improving stimulus generation toward coverage goals. [C1][C3]