Functional ATPGs based on simulation are fast, but generally, they are unable to cover corner cases, and they cannot prove untestability. On the contrary, functional ATPGs exploiting formal methods, being exhaustive, cover corner cases, but they tend to suffer of the state explosion problem when adopted for verifying large designs. In this context, we have defined a functional ATPG that relies on the joint use of pseudo-deterministic simulation and Constraint Logic Programming (CLP), to generate high-quality test sequences for solving complex problems. Thus, the advantages of both simulation-based and static-based verification techniques are preserved, while their respective drawbacks are limited. In particular, CLP, a form of constraint programming in which logic programming is extended to include concepts from constraint satisfaction, is well-suited to be jointly used with simulation. In fact, information learned during design exploration by simulation can be effectively exploited for guiding the search of a CLP solver towards DUV areas not covered yet. The test generation procedure relies on constraint logic programming (CLP) techniques in different phases of the test generation procedure. The ATPG framework is composed of three functional ATPG engines working on three different models of the same DUV: the hardware description language (HDL) model of the DUV, a set of concurrent EFSMs extracted from the HDL description, and a set of logic constraints modeling the EFSMs. The EFSM paradigm has been selected since it allows a compact representation of the DUV state space that limits the state explosion problem typical of more traditional FSMs. The first engine is randombased, the second is transition-oriented, while the last is fault-oriented. The test generation is guided by means of transition coverage and fault coverage. In particular, 100% transition coverage is desired as a necessary condition for fault detection, while the bit coverage functional fault model is used to evaluate the effectiveness of the generated test patterns by measuring the related fault coverage. A random engine is first used to explore the DUV state space by performing a simulation-based random walk. This allows us to quickly fire easy-to-traverse (ETT) transitions and, consequently, to quickly cover easy-to-detect (ETD) faults. However, the majority of hard-to-traverse (HTT) transitions remain, generally, uncovered. Thus, a transition-oriented engine is applied to cover the remaining HTT transitions by exploiting a learning/backjumping-based strategy. The ATPG works on a special kind of EFSM, called SSEFSM, whose transitions present the most uniformly distributed probability of being activated and can be effectively integrated to CLP, since it allows the ATPG to invoke the constraint solver when moving between EFSM states. A constraint logic programming-based (CLP) strategy is adopted to deterministically generate test vectors that satisfy the guard of the EFSM transitions selected to be traversed. Given a transition of the SSEFSM, the solver is required to generate opportune values for PIs that enable the SSEFSM to move across such a transition. Moreover, backjumping, also known as nonchronological backtracking, is a special kind of backtracking strategy which rollbacks from an unsuccessful situation directly to the cause of the failure. Thus, the transition-oriented engine deterministically backjumps to the source of failure when a transition, whose guard depends on previously set registers, cannot be traversed. Next it modifies the EFSM configuration to satisfy the condition on registers and successfully comes back to the target state to activate the transition. The transition-oriented engine generally allows us to achieve 100% transition coverage. However, 100% transition coverage does not guarantee to explore all DUV corner cases, thus some hard-to-detect (HTD) faults can escape detection preventing the achievement of 100% fault coverage. Therefore, the CLP-based fault-oriented engine is finally applied to focus on the remaining HTD faults. The CLP solver is used to deterministically search for sequences that propagate the HTD faults observed, but not detected, by the random and the transition-oriented engine. The fault-oriented engine needs a CLP-based representation of the DUV, and some searching functions to generate test sequences. The CLP-based representation is automatically derived from the S2EFSM models according to the defined rules, which follow the syntax of the ECLiPSe CLP solver. This is not a trivial task, since modeling the evolution in time of an EFSM by using logic constraints is really different with respect to model the same behavior by means of a traditional HW description language. At first, the concept of time steps is introduced, required to model the SSEFSM evolution through the time via CLP. Then, this study deals with modeling of logical variables and constraints to represent enabling functions and update functions of the SSEFSM. Formal tools that exhaustively search for a solution frequently run out of resources when the state space to be analyzed is too large. The same happens for the CLP solver, when it is asked to find a propagation sequence on large sequential designs. Therefore we have defined a set of strategies that allow to prune the search space and to manage the complexity problem for the solver.

### Test generation based on CLP

#####
*DI GUGLIELMO, Giuseppe;FUMMI, Franco;MARCONCINI, Cristina;PRAVADELLI, Graziano*

##### 2009-01-01

#### Abstract

Functional ATPGs based on simulation are fast, but generally, they are unable to cover corner cases, and they cannot prove untestability. On the contrary, functional ATPGs exploiting formal methods, being exhaustive, cover corner cases, but they tend to suffer of the state explosion problem when adopted for verifying large designs. In this context, we have defined a functional ATPG that relies on the joint use of pseudo-deterministic simulation and Constraint Logic Programming (CLP), to generate high-quality test sequences for solving complex problems. Thus, the advantages of both simulation-based and static-based verification techniques are preserved, while their respective drawbacks are limited. In particular, CLP, a form of constraint programming in which logic programming is extended to include concepts from constraint satisfaction, is well-suited to be jointly used with simulation. In fact, information learned during design exploration by simulation can be effectively exploited for guiding the search of a CLP solver towards DUV areas not covered yet. The test generation procedure relies on constraint logic programming (CLP) techniques in different phases of the test generation procedure. The ATPG framework is composed of three functional ATPG engines working on three different models of the same DUV: the hardware description language (HDL) model of the DUV, a set of concurrent EFSMs extracted from the HDL description, and a set of logic constraints modeling the EFSMs. The EFSM paradigm has been selected since it allows a compact representation of the DUV state space that limits the state explosion problem typical of more traditional FSMs. The first engine is randombased, the second is transition-oriented, while the last is fault-oriented. The test generation is guided by means of transition coverage and fault coverage. In particular, 100% transition coverage is desired as a necessary condition for fault detection, while the bit coverage functional fault model is used to evaluate the effectiveness of the generated test patterns by measuring the related fault coverage. A random engine is first used to explore the DUV state space by performing a simulation-based random walk. This allows us to quickly fire easy-to-traverse (ETT) transitions and, consequently, to quickly cover easy-to-detect (ETD) faults. However, the majority of hard-to-traverse (HTT) transitions remain, generally, uncovered. Thus, a transition-oriented engine is applied to cover the remaining HTT transitions by exploiting a learning/backjumping-based strategy. The ATPG works on a special kind of EFSM, called SSEFSM, whose transitions present the most uniformly distributed probability of being activated and can be effectively integrated to CLP, since it allows the ATPG to invoke the constraint solver when moving between EFSM states. A constraint logic programming-based (CLP) strategy is adopted to deterministically generate test vectors that satisfy the guard of the EFSM transitions selected to be traversed. Given a transition of the SSEFSM, the solver is required to generate opportune values for PIs that enable the SSEFSM to move across such a transition. Moreover, backjumping, also known as nonchronological backtracking, is a special kind of backtracking strategy which rollbacks from an unsuccessful situation directly to the cause of the failure. Thus, the transition-oriented engine deterministically backjumps to the source of failure when a transition, whose guard depends on previously set registers, cannot be traversed. Next it modifies the EFSM configuration to satisfy the condition on registers and successfully comes back to the target state to activate the transition. The transition-oriented engine generally allows us to achieve 100% transition coverage. However, 100% transition coverage does not guarantee to explore all DUV corner cases, thus some hard-to-detect (HTD) faults can escape detection preventing the achievement of 100% fault coverage. Therefore, the CLP-based fault-oriented engine is finally applied to focus on the remaining HTD faults. The CLP solver is used to deterministically search for sequences that propagate the HTD faults observed, but not detected, by the random and the transition-oriented engine. The fault-oriented engine needs a CLP-based representation of the DUV, and some searching functions to generate test sequences. The CLP-based representation is automatically derived from the S2EFSM models according to the defined rules, which follow the syntax of the ECLiPSe CLP solver. This is not a trivial task, since modeling the evolution in time of an EFSM by using logic constraints is really different with respect to model the same behavior by means of a traditional HW description language. At first, the concept of time steps is introduced, required to model the SSEFSM evolution through the time via CLP. Then, this study deals with modeling of logical variables and constraints to represent enabling functions and update functions of the SSEFSM. Formal tools that exhaustively search for a solution frequently run out of resources when the state space to be analyzed is too large. The same happens for the CLP solver, when it is asked to find a propagation sequence on large sequential designs. Therefore we have defined a set of strategies that allow to prune the search space and to manage the complexity problem for the solver.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.