AN EFFICIENT SYSTEM-LEVEL TECHNIQUE TO DETECT DATA-DEPENDENT FAILURES IN DRAMSam

AN EFFICIENT SYSTEM-LEVEL TECHNIQUE TO DETECT DATA-DEPENDENT FAILURES IN DRAMSam www.phwiki.com

AN EFFICIENT SYSTEM-LEVEL TECHNIQUE TO DETECT DATA-DEPENDENT FAILURES IN DRAMSam

Justice, Chet, Contributing Writer has reference to this Academic Journal, PHwiki organized this Journal AN EFFICIENT SYSTEM-LEVEL TECHNIQUE TO DETECT DATA-DEPENDENT FAILURES IN DRAMSamira KhanDonghyuk LeeOnur MutluPARBORDRAMMEMORY IN TODAY’S SYSTEMProcessorMemoryStorageDRAM is a critical as long as per as long as mance2MAIN MEMORY CAPACITYGigabytes of DRAMIncreasing dem in addition to as long as high capacity1. More cores2. Data-intensive applications 3How did we get more capacity

South Hills School of Business & Technology PA www.phwiki.com

This Particular University is Related to this Particular Journal

DRAM SCALINGTechnologyScalingDRAM CellsDRAM CellsDRAM scaling enabled high capacity 4DRAM SCALING TRENDTechnologyScalingDRAM CellsDRAM CellsMore interference results in more failures5Scaling places cells in close proximity,increasing cell-to-cell interferenceHow can we enable DRAM scaling without sacrificing reliability6

SYSTEM-LEVEL DETECTION AND MITIGATION7Detect in addition to mitigate failures after the system has become operational Manufacturers can make cells smallerwithout mitigating all failuresSYSTEM-LEVEL DETECTION AND MITIGATION8Enables scalability [SIGMETRICS’14, DSN’14, DSN’15]Lets vendors manufacture smaller, unreliable cellsImproves reliability [ISCA’13, ISCA’14, DSN’14, DSN’15]Can detect failures that escape the manufacturing testsImproves latency [HPCA’15, HPCA’16, SIGMETRICS’16]Reduces latency as long as cells that do not fail at lower latencyEnables refresh optimizations [ASPLOS’11, ISCA’12, DSN’15]Reduces refresh operations by using low refresh rate as long as robust cellsCHALLENGESystem-level detection in addition to mitigationfaces a major challenge due to a specific type of failure:DATA-DEPENDENT FAILURES9

FAILURENO FAILURE11INTERFERENCEDATA-DEPENDENT FAILURESSome cells can fail depending on the data stored in neighboring cells10JSSC’88, MDTD’02Data-dependent failure is a major type of cell-to-cell interference failure CHALLENGE IN DETECTING DATA-DEPENDENT FAILURESDetect failures by writing specific patterns in the neighboring cell addresses11PROBLEM: Scrambled address is notvisible to system (e.g. memory controller) SCRAMBLEDADDRESS X-4XX+2X-1X+1CAN WE DETERMINE THE LOCATION OF PHYSICALLY ADJACENT CELLSNAÏVE SOLUTIONFor a given failure X, test every combination of two bit addresses in the rowNot feasible in a real systemO(n2)81928192 tests, 49 days as long as a row with 8K cells12

OUR APPROACH: PARBOR13Goal: A fast in addition to efficient way to determine the locations of neighboring cells PARBOR: Summary14Reduces test time using two key ideas:Exploits heterogeneity in cell interference to reduce test time by detecting only one neighbor Exploits DRAM regularity in addition to parallelism to detect all neighbor locations by running parallel tests in multiple rowsDetects neighboring locations within 60-99 tests in 144 real DRAM chips, a 745,654X reduction compared to naïve testsA new technique to determine the locations of neighboring DRAM cellsOUTLINE15Data-Dependent FailuresChallenges in System-Level DetectionOur Mechanism: PARBORExperimental Results from Real ChipsUse Cases

A DRAM cellCapacitorTransistorContactTransistorBitlineCapacitorBitlineLOGICAL VIEWVERTICAL CROSS SECTIONA DRAM CELL16DATA-DEPENDENT FAILURESFailures depend on the data content in neighboring cellsIndirect path17Indirect pathCoupled CellsDETECTING DATA-DEPENDENT FAILURES18Need to write specific data patterns in neighboring addresses To test cell at address X, write 1 at address X in addition to 0s at address X+1 in addition to X-1

OUTLINE19Data-Dependent FailuresChallenges in System-Level DetectionOur Mechanism: PARBORExperimental Results from Real ChipsUse CasesCHALLENGE: SCRAMBLED ADDRESS SPACE 20SCRAMBLEDADDRESS X-4XX+2X-1X+1SCRAMBLEDADDRESS X-XX+Scrambled address not visible to systemCannot detect failures without the address mapping in as long as mationDifferent as long as each generation in addition to vendorNeed a dynamic way to detect address mapping in as long as mation in the system CHALLENGE: SCRAMBLED ADDRESS SPACE 21SCRAMBLEDADDRESS X-2XX+5SCRAMBLEDADDRESS X-4XX+2Vendor AVendor B

NAIVE SOLUTION Determine the location of neighboring cellsNAÏVE SOLUTION: O(n2)For a given failure X, test every combination of two bit addresses in the rowAddress bits: (0, 0), (0, 1), (X-1, X), (X, X+1) (n-1, n)For vendor AX will fail only when X-4, X+2 tested 81928192 tests, 49 days as long as a row with 8K cellsNot feasible in a real system22A fast in addition to efficient way to determine the locations of neighboring cells GOAL23OUTLINE24Data-Dependent FailuresChallenges in System-Level DetectionOur Mechanism: PARBORExperimental Results from Real ChipsUse Cases

Justice, Chet Orange County Bride Contributing Writer www.phwiki.com

PARBOR: KEY OBSERVATIONSKey observation 1:Data-dependent failures depend on the heterogeneity in coupled cellsSome cells are strongly coupled in addition to fail based on the data content in just one neighborReduce test time by detecting only one neighborCHALLENGE: Detecting failures with only one neighbor in as long as mation cannot find all failures25Reduces test time based on two key observations:PARBOR: KEY OBSERVATIONSKey observation 2:DRAM exhibits regularity in addition to parallelismNeighbors are located at the same distance in different rows of DRAMDetect all neighbor locations by running parallel tests in multiple rows26Reduces test time based on two key observations:STRONGLY COUPLED CELLFails even if only one neighbor’s data changes WEAKLY COUPLED CELLFails if both neighbors’ data changeKEY OBSERVATION 1: STRONGLY VS. WEAKLY COUPLED CELLS27

Instead of detecting both neighbors, reduce test time by detecting only one neighbor location in strongly coupled cells Does not need to detect every two bit addressesLinearly tests every bit address 0, 1, , X, X+1, X+2, nADVANTAGESReduces test time to linear O(n)Can reduce test time further by applying recursive tests to linear testsKEY IDEA 1: EXPLOITING STRONGLY COUPLED CELLS2829RECURSIVE TESTRECURSIVETESTING 0, 1, 2, 3014, 52, 30, 1SCRAMBLEDADDRESS X-44, 5, 6, 76, 7X26Recursive test reduces test time compared to linear testing30CHALLENGE: Detecting failures with only one neighbor in as long as mation cannot find all data-dependent failures

USE CASE: PHYSICAL NEIGHBOR-AWARE TESTABC52A significant fraction of failures can be detected only by PARBOR (20-30%)

Justice, Chet Contributing Writer

Justice, Chet is from United States and they belong to Orange County Bride and they are from  Mission Viejo, United States got related to this Particular Journal. and Justice, Chet deal with the subjects like Marriage/Weddings; Regional Interest

Journal Ratings by South Hills School of Business & Technology

This Particular Journal got reviewed and rated by South Hills School of Business & Technology and short form of this particular Institution is PA and gave this Journal an Excellent Rating.