Undamped, Lengh arying TLP Pulses Measuremens and ESD Model Approximaions Gonzalo Andrés Pacheco, Julio Guillermo Zola Elecronic Circuis Laboraory Elecronic Deparmen Faculy of Engineering Universiy of Buenos Aires, Argenina e-mail: pachecogonzalo@gmail.com, jzola@fi.uba.ar Absrac Since he inroducion of he ransmission line pulse TLP- for he sudy of elecrosaic discharge ESD- evens on inegraed circuis IC-, here has been many approaches on how o compare is ess resuls wih hose from oher ESD models. Despie i has a very srong correlaion wih he indusry-sandard human body model HBM-, mos implemenaions of TLP weren' suiable for comparisons wih oher ESD models. This work sudies he differen uses of TLP equipmen in order o achieve differen esing waveforms, and infers resuls o oher ESD models based upon he analysis of heir resuls hrough simulaions and saisic esing. I. INTRODUCTION Along wih he consan evoluion of semiconducor devices SD echnology, also come elecrical oversress problems caused by elecrosaic discharge evens [1, 2]. ESD evens can occur on several sages of an inegraed circui IC producion process, from he firs insances of silicon waffer cuing o he cusomer's board mouning of he finished produc [1, 2]. Since here are differen charged objecs ha can place vasly differen ESD hreas o ICs, several ESD models exis. The indusry broadly acceped mainly hree of hem: The human body model (HBM), he machine model (MM) and he charged device model (CDM) [3, 4, 5, 6, 7, 8, 9]. Proecion circuis are implemened and esed on he ICs inpus and oupus in order o provide robus proecion in he case of an ESD even. Sill, differen ESD esers could provide inconsisen informaion due o differen implemenaions, as well as misleading daa of he proecion circui characerisics his is where TLP becomes a very powerful ool [2, 1, 11]. The TLP implemens a square curren waveform wih conrolled ampliude and ime lengh, which makes esing and informaion gahering of proecion circuis much easier and accurae. Despie no being a sandard, he TLP is nowadays he preferred esing mehod o run ESD ess on ICs, and so i has become a mus for ESD esing [11, 12]. This paper covers alernaive implemenaions of he TLP in order o achieve differen waveforms and ye bring he benefis of he TLP esing over oher esing mehods. II. BASIC WAEFORMS A. Sandard ESD waveforms Mos ESD models sandards define heir waveforms hrough ime and ampliude consrains, such as possible ime lenghs or peak ampliudes. Though, mos of hem sugges simple implemenaions and recommended componen values, which can be used o build basic models ha boh comply wih he sandard and are easy o implemen. Their waveforms are shown in Figure 1, which are ypical of simple 1s or 2nd order circuis. 1ns 15ns 1ns HBM MM CDM Figure 1. Waveforms of some basic ESD models.
B. The TLP: How i works The TLP, whose basic scheme is shown on Fig. 2, discharges a ransmission line on he device under es DUT generaing a square waveform see Fig. 3-. The ransmission line is previously charged by a high impedance high volage supply, and is laer discharged on he es device, delivering a square curren pulse o he device. The high volage supply G conrols he curren waveform ampliude, while he pulse duraion is se by he ransmission line lengh, for example d = 1m [1, 13, 14]. 1 2 3 4 5 18 16 14 12 1 8 6 4 2 2 4 6 8 1 Figure 4. Typical I D- DS race for a 1.5μm GGMOS proecion device. Figure 2. Basic TLP scheme Since he DUT is designed o have low impedance during ESD evens, he sysem will be mismached in erms of ransmission lines. This is he reason why he aenuaor is inroduced: i provides and inpu and oupu characerisic impedance Z, so he sysem is kep mached and independen of he DUT [12, 15]. III. TLP HBM COMPARISON I's well-known ha TLP compares o HBM [4] under cerain condiions for many exising echnologies [1, 16]. The correlaion beween he wo models can be found via heir insananeous peak curren, which can be relaed o heir iniial charge volages hrough: Z HBM TLP RHBM = (1) Where, TLP : iniial charge volage of he TLP model. HBM : iniial charge volage of he HBM model. R HBM : series resisance of he HBM (see Fig. 5). Figure 3. Typical TLP waveform The advanage of his approach is ha he pulse shape, and paricularly is rise-ime, is no alered, because i ravels hrough an adaped sysem. The TLP is used o obain informaion of races of proecion devices. This is achieved hrough a series of shos which increase is iniial volage charge, and measures curren and volage across he device, unil i fails [2, 16]. All ess shown in his work were performed by he auhors on 1.5μm lengh, high widh, high curren and medium volage grounded gae MOS (GGMOS) srucures. Figure 4 shows a ypical proecion race which was obained by he auhors wih an indusry sandard TLP equipmen, under he 1μA leakage curren failure crierion. Figure 5. Proposed HBM (sandard values). To back equaion (1), TLP and HBM ess were performed. The DUT were he GGMOS srucures menioned before, again wih he 1μA leakage curren failure crierion and using 2 samples per es. Figure 6 is buil from he acquired mean and sandard deviaion values of heir peak curren failure, and i can be seen he high correlaion beween he failure peak curren of boh models. Despie boh models have compleely differen waveforms, he resuls of one es can be inferred hrough he oher model s esing daa. Bu since TLP esing offers many
benefis in curve racing simpliciy, i makes i preferable over he HBM. As for oher sandard ESD models like MM or CDM, since hey have damped-oscillaing waveforms his direc comparison canno be done, or performs badly [13, 14]..5.4.3.2.1. 8 1 12 14 16 18 Figure 6. Saisic disribuions of TLP and HBM es failures. So, a new es model is proposed: a non-adaped TLP, o observe how i performs when comparing o dampedoscillaing waveforms. I. THE NON-ADAPTED TLP The non-adaped TLP NA-TLP removes he aenuaor from he TLP previously shown, and becomes he TLP shown in Figure 7. BD : breakdown volage of he DUT. R dyn : dynamic resisance of he DUT a breakdown. -,i : forward volage-waveform reflecion. +,i : reverse volage-waveform reflecion. The final waveform depends on many facors; among hem are he breakdown volage and dynamic resisance of he DUT, and he ampliude of each raveling refleced volagewave in he line. This leads o he relaive ampliudes of he series of pulses o vary on each paricular es. 15 1 5-5 -1-15 -1 1 2 3 4 5 6 7 8 9 1 Figure 8. Typical NA-TLP waveform. Figure 8 exposes a NA-TLP waveform for a 1.5μm echnology DUT, where i can be seen he oscillaing naure of he waveform when he aenuaor is removed. I is worh menioning how simple are he modificaions on some TLP equipmens o implemen his waveform, given he fac ha he aenuaor can be easily removed from he signal pah on some sandard TLP equipmens [19]. Figure 7. The NA-TLP circui. Removing his aenuaor causes he curren pulse refleced a he DUT o ravel back o he power supply, where i finds a high impedance source. Then he curren pulse is refleced back again o he ransmission line and back DUT, bu wih is polariy invered. This leads o a series of refleced pulses ha inver heir polariy when hey reflec on he power supply. Ye, he waveform can be described in erms of pulse reflecions for a single sho, hrough he reflecion coefficien for each lobe ampliude: Γ L, i Where, = Z Z, i BD dyn = + (2), + i +, i Rdyn + Z Rdyn + Z R. NOT ADAPTED-TLP ADAPTED TLP COMPARISON The basic modificaion ha can be performed in order o ge closer o he MM sandard values is o shoren he ransmission line lengh, d, from 1m o 5m, which halves he waveform oscillaion period o 1ns, and is referred o as a 5ns NA-TLP, which is he ime lengh of is pulse lobes. Figure 9 show he es resuls of he proposed NA-TLP, compared o he sandard TLP. These resuls show ha he peak curren is no much affeced, and so he curren failure is independen of wheher he sysem is adaped or no..5.4.3.2.1. 9 1 11 12 13 14 15 16 17 18 Figure 9. Peak curren disribuions for 5ns NA-TLP and 1ns TLP.
Figure 1 shows he differen secondary pulse failures. These are failures ha do no occur in he firs lobe of he waveform, and hus implies ha he DUT fails on a pulse wih an inferior peak curren. 1A (a) 9% 8% 7% 6% 5% 4% 3% 2% 1% % secundario TLP 5ns TLP 1ns TLP 2ns TLP 4ns Figure 1. Poin of failure: Secondary failures. This failure mode can be srongly relaed o he MM, in which he alernaing curren can cause failures on many of is lobes [1, 13], as opposed o he HBM. So, i can be seen ha he 1ns NA-TLP can have a closer correlaion o he MM han he sandard TLP, and oher NA-TLP models. 1A -25 25 5 75 1 125 (b) -25 25 5 75 1 125 ime [ns] Figure 12. NA-TLP es waveforms (a) 5ns (b) 1ns. I. WHY 1 NS TLP Wih furher changes on he line lengh d, he NA-TLP pulse lengh was se from 5ns o 4ns. The resuls are shown in Figure 11, where can be seen ha he line lengh does no much aler he curren failure, keeping i beween 13 and 15 ampere..6.5-5 5 1 15 2 25.4.3.2.1. 1 11 12 13 14 15 16 17 18 Figure 11. Failure curren disribuion vs. NA-TLP pulse widh. Even hough, here is a qualiaive difference when looking a which lobe causes he failure. I was found ha he 1ns NA-TLP, in 77% of he cases, causes he DUT o fail on secondary pulses, whereas he oher esed TLPs cause more failures in primary pulses. Figures 12 and 13 show he NA-TLP waveforms used, acquired wih a 1Gb/s digial sorage oscilloscope. In Figure 13(b) i can seen ha he curren decreases 8% due o losses in he ransmission line, bu he waveform sill proved useful for measuremens. -1 1 2 3 4 5 Figure 13. NA-TLP es waveforms (a) 2ns (b) 4ns. II. CONCLUSION The empirical approach o obain a TLP alernaive o achieve a beer MM measuremen was found. Again, i's worh menioning ha he proposed NA-TLP eser can be obained from a sandard TLP eser wih minimal modificaions. The obained waveforms can comply wih many sandards, given sligh modificaions o conrol heir period. Alhough
he waveform is quie basic in shape when compared o he MM, i provides relevan resuls in he failure characerisics of he DUT. Figure 14. Filer beween he NA-TLP equipmen and he DUT o achieve a MM waveform. Also, o achieve a beer waveform approximaion o a MM, a simple filer can be included beween he NA-TLP equipmen and he DUT, o parially smooh he high frequency conen of he waveform. Figure 14 shows an implemenaion example of his filer. Finally, his mehod proved o provide reliable informaion of he DUT characerisics, as well as a new esing mehod ha can be done wih exising sandard ESD esing equipmen. ACKNOWLEDGMENTS The auhors would like o hanks o Cima Ingeniería and TLP Soluions for heir collaboraion and for leing us use heir TLP 5 eser, and for heir echnical suppor during he ess. REFERENCES [1] J. inson and J. Liou, Elecrosaic Discharge in Semiconducor Devices: An Overview, Proceedings of he IEEE, ol. 86, No. 2 pp. 399-42, Feb. 1998. [2] Alber H. Z. Wang, "On-chip ESD Proecion for Inegraed Circuis An IC Design Perspecive" 1s ediion, Springer ISBN -79-237647-1, 22 [3] "IEEE Guide on Elecrosaic Discharge (ESD): Characerizaion of he ESD Enviromen", IEEE Sd C62.47-1992, 1992 [4] MIL-STD-883G: Mehod 315.7 Elecrosaic Discharge Sensiiviy Classificaion, Deparmen of Defense Handbook Sandard, 1989 [5] ANSI/ESD SP5.1.1-26: Sandard Pracice for Human Body Model (HBM) and Machine Model (MM) Alernaive Tes Mehod: Supply Pin Ganging Componen Level", Elecrosaic Discharge Associaion, 26 [6] EIA/JESD22 Tes Mehod A115-A: Elecrosaic Discharge (ESD) Sensiiviy Tesing Machine Model (MM), Elecronic Indusries Associaion, 1997. [7] ANSI/ESD STM 5.2-1999, Elecrosaic Discharge Sensiiviy Tesing - Machine Model (MM) Componen Level, Elecrosaic Discharge Associaion, 1999. [8] JEDEC JESD22-C11C: Field-Induced Charged-Device Model Tes Mehod for Elecrosaic-Discharge-Wihsand Thresholds of Microelecronic Componens, JEDEC Solid Sae Technology Associaion, 24. [9] ESD STMS 3.1: Charged Device Model (CDM) Componen Level for Elecrosaic Discharge Sensiiviy Tesing, Elecrosaic Discharge Associaion, 1999. [1] T. J. Maloney and N. Khurana: Transmission Line Pulsing Techniques for Circui Modeling of ESD Phenomena, Proc. 7h EOS/ESD Symp., p. 49, Sep. 1985. [11] H. Hya, A. Alonzo and P. Bellew: TLP Measuremens for erificaion of ESD Proecion Device Response, IEEE Transacions on Elecronics Packaging Manufacuring, ol. 24, No. 2, pp. 9-98, April 21. [12] J. Barh, K. erhaege, L. G. Henry and J. Richner: TLP Calibraion, Correlaion, Sandards and New Techniques, IEEE Transacions on Elecronics Packaging Manufacuring, ol. 24, No. 2, pp. 99-18, April 21. [13] G. Noermans, P. de Jong and F. Kuper: Pifalls when correlaing TLP, HBM and MM esing, Elecrical Oversress/Elecrosaic Discharge Symposium Proceedings, pp. 17-176, 6-8 Oc. 1998. [14] W. Sadler, X. Guggenmos, P. Egger, H. Gieser and C. Musshoff: Does he ESD-failure curren obained by Transmission-Line Pulsing always correlae o Human Body Model ess?, Elecrical Oversress/Elecrosaic Discharge Symposium, Proceedings, pp. 366-372, 23-25 Sep 1997. [15] L. G. Henry, J. Barh, K. erhaege, and J. Richner, Transmission-Line Pulse ESD Tesing of ICs: A New Beginning, Compliance Engineering, hp://www.ce-mag.com/ce-mag.com/archive/1/3/13ce_46.hml, March/April 21. [16] Julio Zola, Gonzalo Pacheco, "TLP: ESD models correlaion and aproximaion", Proceedings of he Argenine School of Micro- Nanoelecronics, Technology and Applicaions 29, pp. 93-97, 29. [17] J. inson and J. Liou, Elecrosaic Discharge in Semiconducor Devices: Overview of Circui Proecion Techniques, Elecron Devices Meeing, 2. Proceedings. 2 IEEE Hong Kong, pp. 5-8, June 2. [18] Ajih Amerasekera, Charvaka Duvvury, "ESD in Silicon Inegraed Circuis Second Ediion", John Wiley & Sons, Ld. ISBN -47-49871-8, 22 [19] TLPSoluions, "Model TLP5 Transmission Line Pulse Teser - User's Manual v1.3", hp://www.lpsol.com/files/download/tlp5 User's Manual.pdf, 27.