An Effective Guidance Strategy for Abstraction-Guided Simulation

Abstract

Despite major advances in formal verification, simulation continues to be the dominant workhorse for functional verification. Abstraction-guided simulation has long been a promising framework for leveraging the power of formal techniques to help simulation reach difficult target states (assertion violations or coverage targets): model checking a smaller, abstracted version of the design avoids complexity blowup, yet computes approximate distances from any state of the actual design to the target; these approximate distances are used during random simulation to guide the simulator. Unfortunately, the promise has.yet to be realized, as the performance of previous work has been unreliable-sometimes great, sometimes poor. The problem is the guidance strategy. In this thesis, we first develop a platform to enable flexible exploration of abstraction-guided simulation-different guidance heuristics and formal tools are easily inserted-while providing the capacity, speed, and Verilog compatibility of a leading industry-standard (logic-simulation) tool, Synopsys VCS. Then, we start by exploring some greedy heuristics and find that they tend to perform poorly, adding too much search overhead for limited ability to escape dead ends (local optima). Based on these experiments, we propose a new guidance strategy, which pursues a more global search and is better able to avoid getting stuck. Experimental results show that our new guidance strategy is highly effective in most cases that are hard for random simulation and beyond the capacity of formal verification.

Path Constraint Solving Based Test Generation for Hard-to-Reach States

“…However, the computational cost grows exponentially with each refinement. In [3], a more advanced guidance strategy is proposed to alleviate dead-ends. A group of formerly visited states are retained in "buckets" for each abstract distance (i.e., each onion ring).…”

“…Among them, the abstraction-guided simulation has revealed itself as a promising direction, in which formal analysis is applied to an abstract model to get the approximate distances, and the distances are used to direct the simulation to reach a concrete target state [1][2][3][4][5][6].…”

A novel abstraction-guided simulation approach using posterior probabilities for verification

“…The working principle of abstraction-guided simulation is demonstrated in FIGURE 1 [2]. The distances between the target state and all states of the abstract model are calculated by formally analyzing the abstract model.…”

“…Abstract distance rings and corresponding concrete state sequence in simulation. Dead-ends are usually introduced due to the discrepancy between abstract and concrete designs [2].…”

“…The abstraction-guided simulation [1][2][3][4][5] is a very promising semiformal verification method, in which the information of a partial design is calculated on an abstract model, and then is used to guide the simulation towards target states. The abstract information which is used to guide the simulation is acquired by formally analyzing a simplified abstract model, which breaks through the limited scalability of formal method.…”