HMM single site testing: Can we reproduce componentfailure level with the HMM document?Mirko Scholz (1), Robert Ashton (2), Theo Smedes, Richard Derikx (3), Marcel Dekker (4),Jon Barth (5)(1) imec, Leuven, Belgium; tel.: 321616288950, e-mail: [email protected](2) ON Semiconductor, Phoenix, AZ, USA(3) NXP Semiconductors, Nijmegen, The Netherlands(4) MASER Engineering, Enschede, The Netherlands(5) Barth Electronics, Boulder City, NV, USAAbstract – The ESDA working group 5.6 has conducted single site testing to evaluate the repeatability of passfail results when using the setups in the standard practice 5.6 document. A ten times lower standard deviation isobtained in comparison to the 2011 round robin.I. IntroductionToday, many manufacturers of ICs face the difficultythat they need to provide products which are alsorobust to system-level ESD stress, often withoutknowing the final application of their designs. Thisrequires additional effort during the ESD protectiondesign to prevent any IC failure during system-levelESD testing on the application board. To enable theIC manufacturer to validate the ESD performance oftheir products under system-level stress, the ESDAworking group (WG) 5.6 published the Human MetalModel (HMM) standard practice (SP) 5.6 [1]. Itdescribes test setups for the application of the IEC61000-4-2 stress to components. Both ESD gun andHMM pulser based setups are available. In 2011 theWG conducted a round robin study which involvedeight test sites with stress sources as defined inSP 5.6 [2]. The round robin results showed largevariations with an overall standard deviation of 1.6 kVfor one device under test (DUT) and 3.5 kV for theother one. These results cast doubt on the repeatabilityof failure levels obtained with SP 5.6.To investigate why the 2011 round robin producedsuch large variation WG 5.6 designed a new testinground. The focus of this single site (laboratory) testingis first on the repeatability of data captured withdifferent stress sources at a single site. Both currentthrough the DUT and voltage across the DUT were tobe captured during all measurements so that thesource of differing failure levels could be determined.Only in a second step the data of all participating testsites is compared. In this abstract we introduce firstthe measurement setups and DUT. Pass-fail data oftwo testing sites are discussed and compared. Theresults of this single site testing show a much smallervariation and thus smaller standard deviation. Fromthis data we have more confidence on the repeatabilityof failure levels obtained with SP 5.6.II. ExperimentsEach test site ran two experiments. First the referencewaveforms are captured for each stress source. This isfollowed by pass-fail testing of two DUT. Threesamples of the same DUT are tested. Following theSP 5.6 document for each stress level 10 dischargesare applied to the DUT if an ESD gun is a stresssource. One discharge for each stress level is appliedif an HMM pulser is the stress source. The current andvoltage waveforms of the last passed and the failingstress are recorded with an oscilloscope. Before andafter each stress level the DUT leakage is measured todefine the failure level.TestingFor all stress sources the DUT are soldered to a testfixture board (Figure 1a). Transmission lines connectthe DUT to SMA connectors to enable the applicationof stress and the capturing of voltage waveformsduring stress. The outer contacts of the SMAconnectors connect the test board ground to the setupground plane.Stress with the ESD guns is applied with setup A ofSP 5.6 (Figure 1b). HMM pulser connect directly tothe SMA connectors. Voltage waveforms are capturedthrough a voltage divider and attenuators directlyconnected to the oscilloscope input (Figure 1c). Foreach test site the exact measurement conditions are

noted. This includes also information like the lengthand type of the ground wire of the ESD guns.a)a)b)HMM450 ΩAttn.IHMMDUT50 Ω(scope)c)b)Figure 1: Test setup for ESD gun testing: a) Photo of test fixtureboard with DUT and SMA connectors for connection of stresssource (right) and connection of voltage probe (left), b) testsetup A in the SP 5.6 document and c) simplified schematic of testfixture board.Figure 2a shows the implementation of themeasurement setup for testing with the ESD guns inSite1. A copper plate of 1 x 1 m is used as groundplane. A wire connects the test board ground with theground plane. An additional weight is added to thetest board to increase the downforce. Both provisionsimprove the contact of the test board ground with theground plane of the setup.The ground wire of the ESD guns is connected to oneedge of the plane. It should be noted that the groundwires of all three guns have a similar length (Figure2b). The stress current is applied to a small dischargepoint which is connected directly to the center pin ofthe SMA connector of the test board (Figure 2c).c)Figure 2: Site 1: a) Measurements setup during pass-fail testingwith ESD gun; b) ESD gun ground wires and c) discharge pointin Site1.DUTTwo DUT are selected for this study. DUT1 is atransient-voltage suppressor (TVS) device which turnson around 50 V during 100 ns TLP stress (Figure 3a).DUT2 is a diode-based input ESD protection of acommercial RF buffer amplifier (Figure 3b). DUT1representssmart-power/automotivetypeof

applications whereas DUT2 is typical for analog lowvoltage interfaces. Both devices fail thermally andbelow an HMM stress level of 8 kV.87Current [A]654321001020a)3040Voltage [V]506070Site 1 uses a Tektronix TDS 7404 oscilloscope with4 GHz analog bandwidth for waveform capturing. Thecurrent waveforms are captured with a Tektronix CT1 current probe or with the built-in current probe ofthe pulser. The functional test during pass-fail testingis done with an Agilent B1500 parameter analyzerwhen a GUN was used and a Keithley 2400 sourcemeter when a PULSER was used. First, the stresssources are verified against the requirements of SP5.6. Figure 4 shows the currents measured with the 1GHz target for the three ESD guns used in site 1. Allthree guns are fulfilling the requirements of SP 5.6.Two observations are made: the second peak of theHMM current occurs earlier for GUN3. Also after 60ns GUN3 provides a higher stress current than GUN1and GUN2.307Current [A]65Current [A]GUN1GUN2GUN325420I1530 ns10I360 ns5200012b)345Voltage [V]678Figure 3: 100 ns TLP I-V curves of a) DUT1 and b) DUT2, datanot until failure.III. Results Site 1Site 1 used three ESD guns and two HMM pulsers asstress sources (Table 1). PULSER1 is a 50 Ω HMMpulser. PULSER2 is an HMM pulser with a sourceimpedance of 330 Ω.Table 1: Stress sources test site 1 used in this studyStress sourceManufacturerModelGUN 1TESEQNSG 438GUN 2TESEQNSG 438GUN 3SchaffnerNSG 435PULSER 1HPPITLP 3010CPULSER 2HANWAHED-W5000M406080Time [ns]Figure 4: Reference waveforms into 3 GHz target; stress level:8 kV; I30ns and I60ns: specification in SP 5.6.020Figure 5 compares the reference waveforms of GUN1with the two pulsers for a stress level of 4 kV. Notethat results from the 50 Ω HMM pulser are shown asequivalent stress level based on the 30 ns current witha ratio of 2 A/kV.15Current [A]1GUN1PULSER1PULSER210I30 ns5I60 ns002040Time [ns]6080Figure 5: Comparison of the reference waveforms of the stresssources of site 1; stress level: 4 kV; I30ns and I60ns: specification inSP 5.6.

standard deviations are 0.12 kV and 0.35 kV,respectively.15PULSER3PULSER4Current [A]Both HMM pulsers are in specification with therequirements of SP 5.6. PULSER2 provides a slightlyhigher current after 30 ns. After waveformverification pass-fail testing on the DUT is carriedout. The failure level obtained from DUT1 rangedfrom 5.2 to 6 kV (Figure 6a) with an overall standarddeviation of 0.31 kV. The failure level from DUT2ranged from 5 to 5.8 kV (Figure 6b) with an overallstandard deviation of 0.28 kV. These values are muchlower than the standard deviation of 1.7 kV obtainedfor a TVS device in a single site in the round robin.10I30 ns5I60 ns6Failure level [kV]002056GUN1GUN2a)Failure level [kV]3GUN3PULSER2PULSER154326PULSER3 PULSER45Failure level [kV]Failure level [kV]a)432b)80Figure 7: Site 2: Reference waveforms of PULSER3 andPULSER4; stress level: 4 kV.424060Time [ns]GUN1GUN3PULSER2GUN2PULSER1Figure 6: Site1: Failure level of DUT1 (a) and DUT2 (b)IV. Results Site2Two 50 Ω HMM pulsers were used by Site2.PULSER3 is of the same type as PULSER1 (HPPITLP 3010C). PULSER4 is a Barth Electronics 4702.Both HMM pulsers provide stress waveformsaccording to the requirements in SP 5.6 (Figure 7).Thus similar failure levels are expected for the passfail testing. The data obtained from DUT1 lies in arange from 5.3 to 5.6 kV. (Figure 8a). DUT2 fails in arange from 5.3 to 6.2 kV (Figure 8b). The overall65432b)PULSER3 PULSER4Figure 8: Comparison of the obtained failure level during pass-failtesting in Site2 for DUT1 (a) and DUT2 (b).V. Results Site1 and Site2Figure 9 summarizes the failure level of Site1 andSite2. Overall DUT1 fails within a range of 5.2 to6 kV with a standard deviation of 0.28 kV. DUT2fails within a range of 5 to 6 kV with a standarddeviation of 0.33 kV. These results clearly show animprovement over the round robin where standarddeviations up to 3.5 kV were obtained.

However, setup properties like the size of the groundplane, the resistance between test board and groundplane ground and the length and diameter of theground wires can be different. It is also possible thatsmall differences in the interpretation of theprocedures contributed to the variation. These issuescan have significant impact on the rise time,amplitude and shape of the current applied to a DUT.Using the same setup for all three ESD guns, like inthis single site testing, removes this uncertainty.Failure level [kV]65432a)VI. Transient AnalysisGUN1GUN3PULSER2 PULSER4GUN2PULSER1 PULSER3Failure level [kV]65The measurements setup in this new single site testinground allows the capture of voltage and currentwaveforms during device stress. This enables thetransient analysis of the DUT behavior during stressand the calculation of the power or energy at failure.Figure 11 shows the voltage waveforms for DUT1and DUT2 in Site1 at a stress level below the failurelevel. The data shown is taken with three differentstress sources.4200GUN13b)Voltage [V]2GUN1GUN3PULSER2 PULSER4GUN2PULSER1 PULSER3PULSER2100Figure 9: Comparison of the obtained failure level during pass-failtesting in Site 1 and 2 for DUT1 (a) and DUT2 (b).50When combining the failure data of Site1 and Site2similar standard deviations are obtained for ESD gunsand the pulsers (Figure 10).0200.84060Time [ns]80100200overallGUN1PULSER1PULSER2ESD guns150HMM pulsers0.6Voltage [V]standard deviation [kV]0a)10.4100500.20PULSER11500DUT1DUT2Figure 10: Overall standard deviation and the standard deviationsfor each type of stress source and DUT1 and DUT2In the round robin much larger standard deviationswere obtained. This can be because the different testsites used different setups complying with SP 5.6.0b)204060Time [ns]80100Figure 11: Voltage waveforms of a) DUT1 and b) DUT2 capturedin Site1; stress level: 4 kV.

(1)Figure 13 compares the energy over time for DUT1and DUT2 for Site1. The maximum energy correlateswell with the DUT failure level: for lower failure levela higher maximum energy is obtained. In Figure 12bthe failure of DUT2 occurs faster when stressed withPULSER2. This is also reflected in the energy curvewhere an energy of 4 µJ is reached after only 10 ns.When stressed with GUN1 and PULSER1 this energyis reached after 40 ns.2520Energy [ J]No waveform calibration or de-embedding of setupparasitic [3] is applied to the data. Thus, the initialpeak can vary between the different measurementsetups. But still, after the initial peak a similar voltagewaveform is measured independent of the stresssource or setup. Figure 12 shows the voltagewaveforms during failure for DUT1 and DUT2.Different time to failure are observed. DUT1 failsfaster during the first 15 ns if stressed with PULSER2.When stressed with Gun1 and PULSER1 the devicefails after 60 ns. DUT2 fails during the first 15 nswhen stressed with the pulser based setups. Whenstressed with the ESD gun the failure occurs after 30ns. Both DUT are expected to fail thermally. Becauseof this the voltage waveforms can be used to analyzethe energy that leads to device failure. In case of aDUT failure due to voltage overshoot the energyanalysis is misleading. In such failure mode theamplitude and rise time of the initial current peakbecomes important and different failure level ER2Voltage [V]80080120Time [ns]16074065020a)4060Time [ns]80Energy [ )4004080120Time [ns]160Figure 13: Energy over time during the last passing stress level: a)DUT1 and b) DUT220042100Voltage [V]40a)6020b)0Comparison with Site 20204060Time [ns]80100Figure 12: Voltage waveforms of a) DUT1 and b) DUT2 capturedin Site1 during failure.Equation (1) calculates the energy over time:In this section data from the Site 2 pulser of is addedto the analysis. PULSER1 and PULSER3 are 50 Ωpulser with the same model number and made by thesame manufacturer. PULSER4 is a 50 Ω system froma different manufacturer. The pass-fail testing (Figure9) results in similar failure level for all four pulsers.

Hence, similar voltage waveforms are obtainedindependent of the pulser for both DUT (Figure 14).2520020Voltage [V]150Energy [ 2PULSER3PULSER455000020a)4060Time 05Energy [ J]Voltage [V]804043PULSER1PULSER2PULSER3PULSER42201000b)60 80 100 120 140 160Time [ns]204060Time [ns]80100Figure 14: Voltage waveforms of a) DUT1 and b) DUT2 capturedwith pulsers in Site1 and Site2; stress level: 4 kV.Figure 15 plots the energy over time during the lastpassing stress level. Similar trends like during passfail testing are observed. When stressing DUT1 thereis only little variation between the four pulsers. Whenstressing DUT2 with PULSER1, PULSER3 andPULSER4 a similar energy over time curve isobtained. The highest energy level is obtained whenstressing with PULSER2. This corresponds to thelowest obtained failure level.b)0204060 80 100 120 140 160Time [ns]Figure 15: Energy over time during the last passing stress level: a)DUT1 and b) DUT2VII. ConclusionsSingle site testing has been carried out to evaluate ifreproducible failure level can be obtained with SP 5.6.The results from two measurement sites show a muchlower variation and standard deviation in comparisonto the round robin. The overall standard deviation is0.33 kV which is ten times lower than before. Thenew results give higher confidence that SP 5.6 canprovide repeatable results within one lab and betweendifferent labs. In the transient analysis we observe asimilar device behavior when stressed with thedifferent stress sources and setups. Consequently theobtained energy over time curves show the sametrend: the lower the failure level the higher ismaximum energy during the stress. This also provesthat calculating the energy does not reveal anyadditional information than what is obtained with thefailure level.What will be the future of SP 5.6? Tightening thewaveform specifications, as proposed after the round

robin [2], can also result in a lower variation of theDUT failure level. All stress sources in this studyprovide a stress current with waveform parameterswithin 10 %. The current version of SP 5.6 and therelated IEC and ISO standards allow a tolerance of 30 % which results a larger variation betweendifferent users and testing labs.Currently more data is being captured in a third testsite. If this data also results in low variation andstandard deviation, the working group will considerthe steps necessary to elevate the SP document to astandard test method (STM). The elevation to a STMwill incorporate the learning we obtained from thissingle lab testing w.r.t. the definition of the stresswaveform and testing setups. It will also include thetesting of devices which fail due to the initial peak ofthe HMM current. It will guarantee reproducibleresults between different test sites when using theSTM.References1.ANSI/ESD SP 5.6-2009. “Electrostatic DischargeSensitivity Testing - Human Metal Model (HMM)”,ESDA2.Kathy Muhonen, et al, “HMM Round Robin Study:What to Expect When Testing Components to the IEC61000-4-2 Waveform”, Proceedings of EOS/ESDSymposium, 2012.3.Mirko Scholz, et al, “Calibrated wafer-level HBMmeasurements for quasi-static and transient deviceanalysis,” Proceedings of EOS/ESD Symposium, 2007.

GUN 1 TESEQ NSG 438 GUN 2 TESEQ NSG 438 GUN 3 Schaffner NSG 435 PULSER 1 HPPI TLP 3010C PULSER 2 HANWA HED-W5000M Site 1 uses a Tektronix TDS 7404 oscilloscope with 4 GHz analog bandwidth for waveform capturing. The current waveforms are captured with a Tektronix CT-