Gain-Based Technology MappingforMinimumRuntime LeakageunderInput Vector Uncertainty

dependent onthelibrary gates, technology mapping (TM)which The gain-based technology mappingparadigm hasbeen bindsacircuit toasetofspecific gates isanespecially potent successfully employed forfinding minimumdelay andminimum technique. areamappings. However, existing gain-based technology mappers Technology mappingforminimumdelayandareaarewell fail tofind circuits withminimal leakage power. Inthis paper, we developed intheformofload-independent models(3)(15)(16), introduce algorithms andmodeling strategies that enable efficient load-binning techniques (10)andgain-based strategies (4)(2). gain-based technology mapping forminimum leakage power. The Gain-based TM wasintroduced in(4)andhasbeenessential for proposed algorithm isprobability-aware andcanrigorously take solving theproblem oftiming closure (12). Gain-based techniques intoaccount input state probability distribution togenerate a areattractive because oftheir superior computational efficiency circuit mapping withminimumleakage ata given percentile. forrichlibraries. Technology mappingforleakage hasbeen Minimizing leakage athighpercentiles isessential forminimizinginvestigated forthefirst timein(7). However, thesolution of(7) peakleakage, whichstrongly influences thecooling limits and relies onaheuristic treatment ofloaddependency, whichleads to packaging costs. sub-optimal results inthecase ofacontinuously sized gate library. Thealgorithms havebeentested ontheISCAS85 benchmark suite.Furthermore, itisalsolikely toincur ahighrun-time penalty Results indicate that themappings produced bythenewalgorithmbecause oftheneedtodiscretize gates insuchalibrary. consume, onaverage 14%lesser leakage poweratthe99% Theproblem ofTM forleakage inagainbasedsetting hasnot percentile with1% delaypenalty whencompared withthe beenstudied previously. Thefailure ofexisting gain-based approaches usedinprevious gain-based mappers(2).Also, algorithms whicharebased onimplicit area minimization tofind a compared toa dominant-state mapper, ourapproach producesmapping withminimumleakage necessitates thedevelopment of mappings with15%lesser meanvalueofleakage. Thenew suchatechnique. algorithm alsoreduces leakage athighquantiles by12.8%on Thefirst novel contribution ofthis paper isthedevelopment ofthe average, compared toadominant state leakage minimizing mapper technology mapping algorithm forleakage minimization using an andthemaximumsavings canbeashighas21.49%across the efficient gain-based setting. Thesecondessential contribution is benchmarks. Compared tothebinbased mapper (10), theruntimethat thealgorithm isformulated tofind amapping that minimizes ofthealgorithm is15Xfaster.