Test Dlirective Generation forFunctional Coverage Closure UsingInductive Log,ic Programming
Hsiou-Wen Hsueh, Kerstin Eder · 2006
Functional verification isa complexandtime- consuming taskinthedesign process. Recently, various ap- proaches havebeendeveloped toimprove verification efficiency, including advanced coverage analysis techniques, coverage-driven verification methodologies andcoverage-directed stimulus gen- eration techniques. Oneremaininug challenge istofully auto- matefunctional coverage closure. Thispaperpresents anovel approach forcoverage-directed stimulus generation basedon inductive learning fromexamples. Testsequences andtheir related coverage areexamined toinduce general rules which describe thecharacteristics ofthese tests. Coverage closure can beautomated byapplying therulelearning toclusters similar tothetarget coverage holeandcombining theresulting rules to obtain directives fortest generation. Thevalidity oftheapproach isdemonstrated onapilot casestudy. I.INTRODUCTION Generating stimulus toincrease functional coverage isa keychallenge insimulation-based verification. Closing afunc- tional coverage modelisbynomeanstrivial incomplex industrial processor designs. E.g., theeffects ofaggressive pipelining inmodernprocessors often makeitnotobvious toseewhatinput stimulus todrive totarget a specific combination ofsignal values (coverage task), especially when these signals arerelated tomicro-architectural features ofthe design. Inpractice typically upto90%ofcoverage tasks can bereached viabiased pseudo-random tests. Theusercontrols test generation byproviding asetofparameters orconLstraints, called directives, whichbiastests towards certain conditions orcorner cases. However, evensupplying thedirectives requires significant engineering skill andisoften onlyaccomplished through many trial-and-error runs. Attheendengineers resort towriting directed tests byhandaiming tocoverthemissing cases. As aconsequence asmuchas90%ofaverification team's time canbespent onclosing theremaining 10%coverage manually. Moreautomation ofthefunctional coverage closure process isnecessary toaccomplish verification tasks faster, andwith lessengineering effort. Recent advances haveestablished coverage-driven verifica- tionmethodologies whicharefeedback loops that automate mostofthesimulation-based verification process. Coverage results areusedtohelp construct thedirectives forapproaching coverage holes. Themostdemanding aspect regarding full automation ofthese feedback loops istheautomatic generation ofthedirectives forfunctional coverage closure. Coverage- directed stimulus generation (CDG)techniques (16) address