Yield and RedundancyProblemRelated ResearchWhat is most susceptible to failures ?Possible Approaches ?Open QuestionsYield and RedundancyMarc Riedel, CaltechIris Bahar, Brown U.Etienne Jacobs, MagmaDiana Marculescu, CMUPhillip Stanley-Marbell, CMUEric Rotenberg, NCSU2ProblemThe goal: Achieving reliable computing systems from devices with high defect ratesReliability-Aware SynthesisGiven a technique for improving fault-tolerance, how do we judge the efficacy of it in terms of a combination of performance, reliability, power consumption, etc…Redundancy driven synthesis – what to replicate?Observable nodesDevices with high fanoutInstead of redundancy removal  Redundancy addition for increased reliabilityFault ModelWhere to handle it?  Level of abstraction What to handle?  Types of faults3Related ResearchLogic LevelVon Neumann’56AssumptionsPippenger’94Purely theoretical – not automated!!! RT levelStill open ???Architectural levelSlipstream processors (NCSU)Diva (UMich)System levelCMP-based mainframes do use redundancy for increased fault-tolerance!mmO1O2O34What is most susceptible to failures ? Failures at inputs versus outputsInputs: potentially propagates throughout the circuit… But may be masked by other signalsAt primary output: must be masked for correct I/O behavior!Need a measure of:How susceptible a gate is to fail……Or which devices, when failed, will be most critical to the correct functioning of the systemHere synthesis can play a major role!5Possible Approaches ?What works?…A lot of redundancy!Biologically inspired approachesCan models such as how the brain works, and work on neural nets be used in contrast to traditional logicModels of computation and relation to the ability synthesize fault-free systemsDo we need to have/emulate another type of logic (e.g.,threshold logic?)For analysis: borrow / extend ideas from Information Theory6Open QuestionsNo guarantee of complete reliability, but rather a specifiable probability of correct functioningReduce cost of testing by testing only what really mattersCheck only the checker!What parts of the circuit should be made redundantIdentify what’s important and what’s likely to failHow does the addition of synthesis methods for fault-tolerance increase the complexity of verification?E.g., speculate and then check using redundant logic. Who’s going to verify that? (Or do we need