Journal paper Speed Optimized Diode-Triggered SCR (DTSCR) for RF ESD Protection of UltraSensitive IC Nodes in Advanced Technologies Transactions on Device and Materials Reliability 2005 A novel diode‐triggered silicon‐controlled rectifier (DTSCR) (Mergens et al., 2003) electrostatic discharge (ESD) protection element is introduced for low‐voltage application (signal and supply voltages ≤1.8 V) with extremely narrow ESD design margins. Trigger‐voltage engineering in conjunction with fast and efficient SCR voltage clamping is applied for the protection of ultrasensitive circuit nodes, such as SiGe heterojunction bipolar transistor (HBT) base regions (e.g., fTmax = 45 GHz in BiCMOS 0.35µm LNA input) and thin gate oxides (e.g., tox = 1.7 nm in CMOS 0.09µm high‐speed input). Ultrathin gate protection requires a reinforced trigger diode chain to avoid SCR trigger‐speed issues resulting in critical trigger‐voltage overshoots for very fast ESD transients such as a charged device model (CDM). SCR integration can be realized based on parasitic n‐p‐n/p‐n‐p inherent to CMOS devices or can alternatively be implemented based on vertical high‐ speed SiGe HBT with adjacent p+ SCR anode. 532 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 5, NO. 3, SEPTEMBER 2005 Speed Optimized Diode-Triggered SCR (DTSCR) for RF ESD Protection of Ultra-Sensitive IC Nodes in Advanced Technologies Markus P. J. Mergens, Christian C. Russ, Koen G. Verhaege, John Armer, Phillip C. Jozwiak, Russell P. Mohn, Member, IEEE, Bart Keppens, and Cong Son Trinh Abstract—A novel diode-triggered silicon-controlled rectifier (DTSCR) (Mergens et al., 2003) electrostatic discharge (ESD) protection element is introduced for low-voltage application (signal and supply voltages ≤ 1.8 V) with extremely narrow ESD design margins. Trigger-voltage engineering in conjunction with fast and efficient SCR voltage clamping is applied for the protection of ultrasensitive circuit nodes, such as SiGe heterojunction bipolar transistor (HBT) base regions (e.g., fTmax = 45 GHz in BiCMOS 0.35-µm LNA input) and thin gate oxides (e.g., tox = 1.7 nm in CMOS 0.09-µm high-speed input). Ultrathin gate protection requires a reinforced trigger diode chain to avoid SCR trigger-speed issues resulting in critical trigger-voltage overshoots for very fast ESD transients such as a charged device model (CDM). SCR integration can be realized based on parasitic n-p-n/p-n-p inherent to CMOS devices or can alternatively be implemented based on vertical high-speed SiGe HBT with adjacent p+ SCR anode. Index Terms—BiCMOS, CMOS, electrostatic discharge (ESD), gate-oxide protection, radio frequency (RF), SiGe heterojunction bipolar transistor (HBT), silicon-controlled rectifier (SCR), trigger speed. S Fig. 1. Narrow ESD design-window application of ultrathin gate oxide (tox = 1.7 nm) protection in CMOS 0.09-µm: TLP oxide breakdown characteristic (I–V and leakage) and typical supply voltage of VDD = 1.2 V. Comparison to GGNMOS protection TLP I–V curve: Snapback trigger voltage Vt1 and holding voltage Vhold are too close to the transient Gox breakdown voltage. Manuscript received January 13, 2004; revised November 24, 2004. This paper is the expanded and modified version of a conference paper presented at the IEDM, Washington, DC, 2003. M. P. J. Mergens was with Sarnoff Europe, Gistel B-8470, Belgium. He is now with Infineon Technologies, Munich D-81669, Germany (e-mail: mmergens@onlinehome.de). C. C. Russ was with the Sarnoff Corporation, Princeton, NJ 08543 USA. He is now with Infineon Technologies, Munich D-81669, Germany. K. G. Verhaege and B. Keppens are with Sarnoff Europe, Gistel B-8470, Belgium. J. Armer, P. C. Jozwiak, R. P. Mohn, and C. S. Trinh are with the Sarnoff Corporation, Princeton, NJ 08543 USA. Digital Object Identifier 10.1109/TDMR.2005.853510 Decisively, the superior voltage-clamping capabilities make SCRs ideal protection alternatives in technologies, where the so-called ESD design-window (refer to below) becomes very narrow. Therefore, significant future relevance of these efficient voltage clamps is anticipated in particular for sub-0.13-µm CMOS applications. Fig. 1 provides an example for narrow ESD design-window application of protecting an ultrathin gate oxide (physical thickness tox = 1.7 nm) in a 0.09-µm CMOS technology. The plot compares the transmission line pulse (TLP) I–V characteristic (square pulse duration 100 ns, selectable rise time 200 ps, 2 ns, and 10 ns) of an unprotected NMOS gate oxide to a conventional GGNMOS protection I–V curve. As indicated by the leakage-current evolution of the stressed gate oxide, damage occurs at a transient breakdown voltage of approximately BVox ∼ 5.4 V. The GGNMOS snapback trigger voltage comes extremely close to this critical limit. Moreover, the relatively high holding voltage does not provide any ESD design margin. Therefore, applying such an NMOS device for thin-oxide protection in this 0.09-µm CMOS and in more advanced I. I NTRODUCTION ILICON-CONTROLLED rectifiers (SCRs or thyristors) have long been used as on-chip electrostatic discharge (ESD) protection elements over a broad range of technologies because of their superior ESD-protection capabilities [1]–[7]. With SCRs, an extremely high failure current, low dynamic on resistance, and an ideal ESD performance width scaling can be accomplished. This is a result of the regenerative conduction mechanism due to double minority carrier injection into both wells leading to the low-resistive latch-up operation. The excellent high current behavior of SCRs can typically provide an area gain factor of 4 to 5 over standard silicide-blocked groundedgate NMOS (GGNMOS) ESD-protection elements. 1530-4388/$20.00 © 2005 IEEE MERGENS et al.: SPEED OPTIMIZED DTSCR FOR RF ESD PROTECTION OF ULTRA-SENSITIVE IC NODES Fig. 2. TLP NMOS gate-oxide breakdown and GGNMOS snapback parameters (trigger voltage Vt1 and holding voltage Vhold ) as a function of physical gate-oxide thickness in different CMOS technologies (0.18–0.065 µm). Technology advancement results in a dramatic shrink of the ESD designwindow. technologies becomes dangerous. Moreover, for radio frequency (RF) design, the requirement for efficient ESD voltage clamps is even more challenging. Here, a local GGNMOS protection approach cannot be used due to the high parasitic junction capacitance (order of magnitude of several picofarads) being introduced to the RF pin. Such a high ESD capacitance would largely compromise high-speed performance. Therefore, in conjunction with NMOS-type clamps, RF ESD implementations require ESD diodes to be locally connected to the capacitance-sensitive pins (e.g., in the “dual-diode” protection approach) to minimize the parasitic junction capacitance. However, for worst case ESD stress conditions on the integrated circuit (IC) level, this implies that there is always a diode voltage drop in addition to the protection clamp in the ESD discharge path. Fig. 1 demonstrated how this additional diode voltage drop would further minimize the ESD design margins by pushing the total I–V curve of the protection solution outside the ESD design-window. In conclusion, the application of the conventional GGNMOS protection approach becomes extremely critical if not impossible for ESD-protection design of RF inputs containing ultrathin gate oxides in sub-130-nm CMOS technologies. This general behavior is corroborated in Fig. 2, which depicts the general trend of shrinking gate oxides with CMOS technology advancement. The plot compares the transient NMOS oxide breakdown voltage to the scaling behavior of the corresponding GGNMOS ESD parameters (Vt1 , Vhold ) as a function of the oxide thickness for a number of advanced technologies (0.18–0.065 µm). Apparently, the ESD susceptibility of the gate oxide is increasing at a rapid pace as technologies advance to thinner gate oxides resulting in a dramatic shrinking of the design-window. On the other hand, the corresponding GGNMOS clamping capabilities do not improve to the same degree. As a result, the trigger and holding voltage eventually exceed the dynamic breakdown voltage of the ultrathin gate oxides, thus, disabling the protection type. 533 Fig. 3. Narrow ESD design-window application of SiGe HBT input protection in BiCMOS 0.35 µm: BE breakdown characteristic (I–V and leakage) and typical input voltage of Vin = 1.0 V. Comparison to GGNMOS protection TLP I–V curve: Snapback trigger Vt1 and holding voltages Vhold are outside the ESD design-window. The dramatic implications on high-speed input–output (I/O) driver design in the (low-voltage) thin-oxide domain becomes evident: the common concept of self-protecting MOS drivers in I/Os does not work anymore due to input oxide damage. Salman et al. reported the same behavior for a CMOS 0.1-µm technology [8]. Except for applying SCRs for ultrathin gate-oxide input protection in CMOS technologies, these devices can also be used in bipolar or BiCMOS technologies to protect the sensitive HBT base–emitter (BE) junctions, cf. Fig. 3. Particularly, if stressed in forward mode, the inherent BE diode fails at low clamp voltages (here approximately 6 V). In this case, TLP experiments were conducted for a relatively large HBT cell resulting in a fairly high HBT failure current. For other RF applications, the HBT size can be up to a factor of 4 smaller with a low failure current of approximately 0.3 A. In accordance to the thin gate-oxide situation described above, n-p-n bipolar-based protection (e.g., GGNMOS) cannot provide sufficient voltage clamping to keep the HBT safe. This is clearly demonstrated by the GGNMOS TLP I–V curve, which is entirely situated outside the HBT ESD design-window, i.e., in the HBT damage regime. In conclusion, for presently applied product technologies and for future aggressive technology scaling, conventional bipolarbased protection (e.g., parasitic n-p-n on GGNMOS) does not work anymore for many ultrasensitive IC nodes due to insufficient voltage clamping. SCRs are a viable solution alternative to conventional ESD elements due to their excellent voltage-clamping capabilities and ESD performance per area. Moreover, the parasitic junction capacitance per ESD is an order of magnitude lower compared to NMOS transistors making SCRs suitable candidates for local RF I/O protection. Note that the junction capacitance roughly 534 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 5, NO. 3, SEPTEMBER 2005 Fig. 4. (a) G1 -triggered DTSCR forward biasing SCR G1 –cathode junction. Rsub indicates intrinsic p-well connection to the substrate. (b/c) G2 -triggered DTSCR forward biasing to G2 –anode junction. (c) Holding diode in series with SCR for normal operation latch-up immunity replaces one trigger diode. equals a diode capacitance (e.g., in CMOS, a p+/n-well), which can be reliably estimated by using simulation program with integrated circuits emphasis (SPICE) parameters and junction geometry parameters. To customize latch-up elements for ESD design in leadingedge technologies, one has to overcome a number of design challenges. SCR-device engineering techniques and SCR ICprotection application are the topics of this paper, covered in the following sections. 1) Section II-A: SCR trigger-voltage engineering for narrow ESD design-window applications in advanced technologies: diode-chain-triggered SCR (DTSCR) including various trigger schemes, which allow SCR turn-on well below the critical voltage limit of IC damage. 2) Section II-B: SCR holding-voltage engineering for latchup immunity: avoid unintended SCR triggering during normal IC operation. 3) Sections II-C and II-D: DTSCR implementation techniques in advanced CMOS/BiCMOS (SiGe HBT) technologies. 4) Section III-A: DTSCR IC application example in 0.35-µm BiCMOS: sensitive SiGe HBT base protection, cf. design-window in Fig. 3. 5) Section III-B: DTSCR IC application example in 0.09-µm CMOS for ultrathin gate-oxides protection, cf. design-window in Fig. 1. DTSCR trigger-speed enhancement is required for fast ESD transients as produced by the charged device model (CDM). II. DTSCR P ROTECTION D ESIGN A. SCR ESD Trigger-Voltage Engineering The DTSCR employs an external trigger diode chain to latch the device during ESD stress conditions as soon as the diode string injects enough current into the SCR gates (or “trigger taps”) G1 or G2 , cf. Fig. 4. Triggering can either be accomplished by forward biasing the inherent SCR G1 –cathode junction, cf. Fig. 4(a), or alternatively the G2 –anode diode, cf. Fig. 5. DTSCR Vt1 engineering: TLP I–V characteristic of DTSCR (CMOS-SCR, see Section III-B) for three G1 trigger diode schemes: Vt1 ≈ (n + 1) · 0.8 V (n trigger diodes + intrinsic SCR G1 /Cdiode). Fig. 4(b), or both simultaneously. There is an important design tradeoff for the number of trigger diodes n required for leakage limitation during normal operation conditions (maximize n) and ESD trigger-voltage reduction (minimize n). For IC normal operation, the number of trigger diodes (n) must be sufficiently high such that the chain does neither leak nor trigger the SCR. Conversely, the SCR ESD trigger voltage increases with n. A reasonable design compromise can be achieved for lowvoltage applications where the supply or signal voltage does not exceed 1.8 V. For CMOS-based SCRs, see Section II-C, G2 -triggered devices bear the advantage that the SCR G2 (n-well)–anode diode can be exploited as a trigger diode within the diode chain due to the isolation of the parasitic n-well diode from a common p-substrate. In contrast, in G1 (p-well)-triggered elements, the diode chain sees an intrinsic p-well–substrate connection, which bypasses the G1 (p-well)–cathode diode as indicated MERGENS et al.: SPEED OPTIMIZED DTSCR FOR RF ESD PROTECTION OF ULTRA-SENSITIVE IC NODES 535 Fig. 6. DTSCR latch-up engineering: TLP I–V characteristic of a G2 -triggered DTSCR (CMOS-SCR) for 1.8-V application with six trigger diodes and no holding diode compared to five trigger diodes plus one holding diode. Result: Approximately 1-V difference in holding voltage, approximately same Vt1 , but higher It2 with series diode due to parasitic Darlington current into substrate. by Rsub in Fig. 3(a). Therefore, the G1 trigger configuration requires one trigger diode more than G2 triggering to satisfy normal operation leakage requirements. In other words, G2 triggering allows for the same DTSCR normal operation leakage, while using one trigger diode less (substituted by the inherent anode–G2 diode) as compared to G1 triggering. Consequently, for a predefined leakage specification, G2 -triggered SCRs can be activated by a one-diode-voltage-drop (∼ 0.8 V) lower trigger voltage. Fig. 5 illustrates the Vt1 engineering technique by showing TLP I–V characteristics of various DTSCRs with different trigger diode chains. In this example, a lateral SCR is incorporated based on CMOS layers as will be explained in Section II-C. B. Holding-Voltage Engineering for Latch-Up Immunity To ensure latch-up immunity during normal operation for VDD higher than the SCR holding voltage (e.g., if the DTSCR is used as a 1.8-V supply clamp), a series diode is needed to increase Vhold above VDD [1], cf. Fig. 4(c). This holding diode can also become part of the trigger chain in a G2 -triggered element, replacing one trigger diode. Fig. 6 shows the I–V characteristics of two G2 -triggered SCRs demonstrating a holding-voltage shift by approximately 1 V with a p+/n-well series diode. Thus, the design is latchup immune for VDD = 1.8 V. Furthermore, the substitution of a trigger diode by a holding diode does not alter the trigger voltage Vt1 . Surprisingly, a slightly reduced dynamic on resistance as well as a higher failure current can be observed for the SCR containing a holding diode. These effects can be attributed to the parasitic vertical p-n-p of the p+/n-well diode chain forming a Darlington p-n-p in conjunction with the anode/ n-well junction of the SCR as shown in the schematic in Fig. 6. Overall, this parasitic creates an additional ESD current path to GND in parallel to the SCR, thus enhancing the overall ESD failure current It2 . Fig. 7. Cross section and schematic layout of a CMOS-based SCR (not shown: External diode-trigger chain, which can be connected to G1 , or G2 , or G1 and G2 ). The local G1 /G2 trigger taps can be inserted between the interrupted cathode/anode diffusions, respectively [1]. In the DTSCR, a diode chain is applied to bias the wells through the gates G1 /G2 . C. SCR Integration in CMOS Conventional CMOS design integrates lateral SCR n-p-n-p structures using an adjacent p-well containing an n+ cathode and an n-well containing a p+ anode, respectively [2]. A schematic example layout and a cross section of the SCR kernel including gate/trigger tap implementation (G1 : p+ in p-well; 536 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 5, NO. 3, SEPTEMBER 2005 Fig. 8. Cross section and schematic layout of SiGe-HBT-based SCR (not shown: external diode-trigger chain, which can be connected to G1 , or G2 , or G1 and G2 ). The symmetrical SCR device is formed by a vertical n-p-n HBT (emitter/SiGe-base/n-epi) in conjunction with a distributed lateral p-n-p bipolar (anode/ n-epi/SiGe-base). Note: alternatively, the same intermittent trigger taps as shown in Fig. 4 could be used, replacing the solid G1 /G2 stripes. G2 : n+ in n-well) are depicted in Fig. 7. Separating the external trigger element from the SCR allows the optimizing of the SCR independently regarding trigger speed and high current capabilities. For example, allowing for minimum anode–cathode spacing results in minimum turn-on time due to smallest bipolar junction transistor (BJT) base lengths composing the SCR. D. SCR Integration With SiGe HBTs The schematic layout and cross section of an SiGe-HBTbased SCR are shown in Fig. 8. Due to the vertical current flow from the SiGe HBT emitter to collector, the SCR will benefit from the high currents that can flow in a vertical bipolar structure in contrast to a predominantly lateral bipolar conduction in a lateral SCR (e.g., in CMOS). As demonstrated in Fig. 9, the ESD performance in terms of maximum current It2 and human body model (HBM) level drastically increases inserting more contact rows in the anode and cathode, respectively, due to the reinforced vertical junction area. Furthermore, the SCR holding voltage is relatively small (Vhold ∼ 1.0 V) as compared to regular CMOS implementations reflecting the high transistor current gain of the vertical SiGe HBT part of the SCR. In addition, the thin SiGe base, as well as graded doping concentration, makes this SCR a relatively fast device. Another interesting benefit of the described SiGe HBT-SCR is the fact that the SCR is fully isolated from the substrate. As a result, the gate G1 can be treated the same way as gate G2 (see discussion above: G1 trigger versus G2 trigger). This allows simultaneous G1 /G2 triggering with identical diode chains on both gates. As shown in [10], dual-gate injection enhances the SCR speed and thus helps to avoid transient trigger-voltage overshoots during the device turn-on time. See discussion in Section III-B. Moreover, due to the complete device isolation from the substrate, these DTSCRs can also be applied as local protection elements between I/O and VDD. A further benefit is the high latch-up immunity of the device induced by substrate current injection. III. DTSCR RF ESD-P ROTECTION A PPLICATION A. Input SiGe HBT Base in BiCMOS 0.35 µm SiGe HBT ( fTmax = 45 GHz) emitter–base protection applying the conventional dual-diode protection approach in conjunction with a GGNMOS or for example, an NMOS-triggered SCR power clamp is ineffective due to the low BE failure voltage (∼ 6 V), as illustrated in the ESD design-window in Fig. 3. Moreover, a direct local-diode chain protection between MERGENS et al.: SPEED OPTIMIZED DTSCR FOR RF ESD PROTECTION OF ULTRA-SENSITIVE IC NODES 537 Fig. 9. HBT-based DTSCR (W = 2 × 50 µm; two-sided) TLP I–V for three different variations of number of contact rows in anode and cathode, respectively. The performance boost with larger number of contact rows reflects the predominant vertical current flow in HBT-based SCRs. Fig. 10. Bipolar LNA circuit including local DTSCR protection of sensitive HBT BE diode in conjunction with application of capacitance-reduction scheme for the anode by reverse biasing the p+/n-well junction. ESD-protection scheme meets the RF capacitance spec at each pin (voltage-dependent capacitance C(V ) values calculated by SPICE parameters). All pin combinations exceed ESD spec of 2 kV-HBM and 200 V-MM spec (latter corresponds to approx 4 kV-HBM). Input and VEE (for both pins, a capacitance spec is defined) is not possible either, since a too-high capacitance of the forward-biased diode would compromise the RF performance. Introducing a diode chain between IN and GND in an input configuration, as shown in Fig. 10, would violate the ESD design-window: for worst case stress between IN and VEE, four width-limited RF diodes (three between IN and GND plus one between GND and VEE) with relatively high series resistance would compete with the parasitic HBT BE diode in forward conduction. This would result in HBT failure below the ESD spec. Local DTSCR protection, as shown in Fig. 10, succeeds for three reasons. 1) By applying three small trigger diodes in series, the SCR Vt1 can be reduced sufficiently below the critical HBT failure voltage at 6 V with an acceptably lowinput leakage during normal RF operation. 2) The SCR clamps the IN–VEE voltage below the HBT damage level, thus enabling a challenging 200 V-MM spec (roughly corresponds to 4 kV-HBM in this technology). 3) To meet the specific input RF capacitance requirement CIN < 120 fF, the SCR-n-well can be pulled high to VCC through G2 , thus, reverse biasing the SCR anode–n-well junction and minimizing the parasitic capacitance of the anode at IN. This capacitance reduction scheme is the reason for using the G1 trigger scheme in the local DTSCR clamp rather than the otherwise preferred G2 trigger (cf. Section II-A). The TLP characteristic in Fig. 11 corroborates the functionality of the DTSCR structure, revealing the following I–V 538 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 5, NO. 3, SEPTEMBER 2005 Fig. 11. LNA TLP I–V characteristic: Worst case stress (positive on Input versus VEE on ground) protected by local DTSCR, cf. Fig. 1. The DTSCR triggers and clamps the voltage below HBT BE damage. The ESD spec of 2 kV-HBM and 200 V-MM is achieved within the RF capacitance constraints. Fig. 12. CMOS 0.09-µm: TLP I–V characteristic of DTSCR including a parallel gate oxide (tox = 1.7 nm) monitor to emulate IC input protection. The thin Gox can be successfully protected. regimes: HBT BE diode conduction, DTSCR triggering, joined HBT/DTSCR conduction (see inset including ESD current path for most critical stress case), HBT failure occurs after the input voltage exceeds the critical HBT failure voltage limit of 6 V at an ESD current of 2.2 A. This HBT RF input design passed ESD product qualification tests of 2 kV-HBM and 200 V-MM. B. Input Ultrathin Gox in CMOS 0.09 µm Fig. 12 depicts the TLP I–V characteristics of two DTSCRs in a G1 - and G2 -triggered configuration, respectively. Both structures have the thinnest NMOS gate oxide (tox = 1.7 nm) as a protection monitor in parallel to emulate an RF IC input configuration, as seen in Fig. 12. Note that the G2 trigger scheme essentially results in the same leakage current as compared to the G1 -triggered SCR. However, the G2 -triggered SCR reveals a lower trigger voltage Vt1 , since one trigger diode less can be used as explained in Section II-A. The leakage increase at It2 indicates that the gate oxide fails within the higher resistive roll-off regime of the SCR I–V curve. The physical mechanism of this I–V roll-off in SCRs is in detail explained in [2]. This means that almost the intrinsic failure current of the stand-alone SCR is reached, which proves that the DTSCR can successfully protect an ultrathin input gate oxide. Moreover, the margin of Vt1 and the clamping voltage with regard to transient BVox provides ESD design flexibility for protection integration on the IC level. As outlined in the introduction, additional ESD voltage drops across MERGENS et al.: SPEED OPTIMIZED DTSCR FOR RF ESD PROTECTION OF ULTRA-SENSITIVE IC NODES 539 Fig. 13. CMOS 0.09-µm: Fast TLP rise time (TR) experiments for DTSCR including holding diode and parallel gate-oxide monitor: for fast TR = 200 ps, the transient voltage overshoot damages the gate oxide in parallel to the DTSCR due to a too-slow protection response time as qualitatively shown in Fig. 14. Fig. 15. TLP I–V characteristic of DTSCR containing reinforced G2 trigger diodes and G2 trigger taps. The trigger voltage is reduced by approximately 0.7 V as compared to conventional DTSCR in Fig. 12. The device normal operation leakage is not affected by the larger trigger diodes. Fig. 14. Qualitative SCR TLP voltage pulses comparing slow TLP rise times (e.g., TR = 10 ns) versus fast TLP rise time (e.g., TR = 200 ps). For fast transient TR trigger voltage, overshoots may result in ultrathin oxide damage as experimentally confirmed in Fig. 13. lenge of SCRs protecting thin gate oxides constitutes the relatively slow trigger speed [9], as compared to (parasitic) bipolar elements. The physical reason for the slower speed is that SCRs are composed of interlinked n-p-n and p-n-p bipolar, which both need to be turned on to start the regenerative ESD latch-up operation resulting in high current conduction and effective voltage clamping due to conductivity modulation. Therefore, for ultrafast ESD transients such as CDM, the SCR response time can be too slow to clamp the voltage sufficiently fast below the damage voltage of the node to be protected. The resulting transient trigger-voltage overshoots can be particularly dangerous if ESD voltage stress is directly exposed to an ultrathin gate oxide as, for instance, that present for RF inputs or so-called decoupling capacitors between supply and ground in IC RF domains. These elements act as alternating current (AC) “shorts” to filter out supply noise and are frequently formed by thin gate-oxide capacitors for area reasons. During ESD stress, the voltage drop across the power clamp will also be fully seen across such a sensitive gate oxide. To worsen the situation for the case of a 1.2-V supply, the DTSCR power clamp requires a series diode for latch-up immunity. The above-described trigger-speed issue is demonstrated in Fig. 13 for a conventional DTSCR power clamp. For slow TLP rise times (TR = 10 ns), the DTSCR successfully protects the ultrathin gate oxide (e.g., decoupling capacitor) in parallel in accordance to data in Fig. 12. However, for fast TLP rise times of TR = 200 ps, the increase of leakage current at a largely diminished TLP amplitude indicates premature gateoxide damage. Fig. 14 explains qualitatively the observed behavior comparing two schematic TLP voltage pulses. For slow transients, the trigger-voltage overshoot due to the finite SCR response time is still situated below the transient oxide damage level according to the good performance level shown in Fig. 13. However, significantly increasing the TLP slew rate dV /dt for TR = 200 ps, the SCR cannot react in time to clamp the diodes and bus resistance need to be tolerated for efficient RF ESD design. The relevant device parameters extracted from the TLP I–V characteristics (trigger voltage Vt1 , holding voltage Vhold , dynamic on resistance Ron , failure current It2 , permissible maximum device current Imax ) are indicated in the plot. Imax represents the maximum current level, which must not be exceeded during ESD stress conditions to allow for efficient voltage clamping. Applying DTSCR protection in an IC product, the decisive SCR parameters (Imax ∝ W , Ron ∝ 1/W ) can be scaled linearly according to the ESD specification to stay in the quasi-linear regime of the I–V characteristics and avoid the higher resistive roll-off region. DTSCR Trigger-Speed Engineering for Thin Gate Oxides Under Very Fast ESD Transients: A well-known design chal- 540 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 5, NO. 3, SEPTEMBER 2005 Fig. 16. TLP I–V characteristic for TR = 200 ps: reinforced DTSCR with and without holding diode and parallel thin gate-oxide monitor (decoupling cap). The gate oxide can be successfully protected. voltage transient below the oxide damage level. The transient trigger peak temporarily exceeds the Gox damage voltage, cf. Fig. 13, whereas the quasi-static TLP clamping behavior does not reveal the issue. This slow response time, as indicated by the TLP analysis, can also limit the CDM protection level on product chips. The objective of the DTSCR trigger enhancement is to reduce the transient trigger overshoot during fast ESD transients. The SCR nucleus itself integrated within the DTSCR is already optimized regarding speed by minimizing the anode–cathode spacing. The answer to the problem is to integrate a fast and sufficiently strong parallel ESD back-up path, which is activated during the relatively long SCR turn-on time. An important option with a diode-trigger chain, apart from low-voltage SCR triggering is that the diode chain can serve as the ESD trigger backup path for the slower SCR by clipping possible transient trigger-voltage overshoots. Careful design of the trigger diodes (width, spacings) and trigger-current injection points (SCR gates G1 or G2 ) is required to form a sufficiently strong initial ESD current path through the diodes reducing the initial voltage peak. To understand that an optimized diode chain can protect during the first nanoseconds, it is important to consider two additional points. First, the transient oxide breakdown within the first nanoseconds of an ESD pulse is significantly higher than the 100-ns TLP value of BVox = 5.2 V shown in Fig. 1. A rough estimation for the CDM relevant time domain gives an order of magnitude of 1.5 higher breakdown values [11], i.e., BVox (ns) ∼ 8 V. Secondly, the increased diode width also results in a smaller voltage drop per diode at the triggering point. At the low trigger-current level, the diode n-well is not conductivity modulated yet, such that the diode resistance is still significantly high and width scaling has an impact on the voltage drop per diode and thus on Vt1 . As shown in Fig. 15, the DTSCR starts to latch at 2.7 V, meaning that the reinforced diodes start to conduct at this voltage level. On a CDM-related time scale, the diode chain needs to clip the voltage below roughly BVox (ns) ∼ 8 V before the SCR can start to clamp the voltage to safe levels for the majority of the pulse duration. The TLP data for TR = 200 ps in Fig. 16 indicates that this objective is accomplished for the reinforced local DTSCR clamp (no Vhold diode) as well as DTSCR power clamp (one Vhold diode). Both these test devices contain an ultrathin gate oxide in parallel and successfully protect this element. Note that the dV /dt effect for TR = 200 ps appears to further reduce the triggering voltage as compared to Fig. 15, which will further support fast turn-on. As can be seen for the DTSCR power-clamp with series Vhold diode, the TLP performance is reduced by roughly 1 A due to the additional voltage drop. The good TLP performance of roughly 2.8 A for this test element was confirmed by HBM tests with a 3.5-kV pass and 3.7-kV fail level. The DTSCR was also applied as a power clamp in the RF domain of product chips containing sensitive ultrathin Gox decoupling capacitors for supply. CDM product specifications of 500 V were met for all products. IV. C ONCLUSION This paper demonstrates that bipolar-based ESD elements, e.g., parasitic lateral n-p-n in GGNMOS, are not suitable anymore to protect sensitive nodes such as ultrathin gate oxides in advanced sub-130-nm CMOS technologies. This is due to insufficient high-current clamping capabilities. As demonstrated, SCRs represent excellent solutions if the trigger voltage is appropriately engineered to fit into an extremely narrow ESD design-window. A novel DTSCR for low-voltage application (signal and supply voltages ≤ 1.8 V) is introduced including different diode-trigger schemes, which allow for SCR triggervoltage reduction sufficiently below the critical voltage levels. Moreover, engineering of normal operation latch-up immunity for DTSCR power-clamp application is discussed within the context of the specific diode-trigger scheme. As demonstrated, due to a relatively slow SCR trigger speed, dangerous voltage MERGENS et al.: SPEED OPTIMIZED DTSCR FOR RF ESD PROTECTION OF ULTRA-SENSITIVE IC NODES overshoots may jeopardize ultrathin gates during very fast ESD transients such as CDM. A novel concept of applying an optimized SCR trigger backup in parallel to the SCR allows coping with this speed issue. Here, the provided backup current path can be integrated into the DTSCR by exploiting a reinforced, thus low-resistive, trigger diode chain, which would clip initial fast voltage transients and results in a reduced DTSCR trigger voltage. DTSCR integration can be based on lateral parasitic BJTs in CMOS technologies as well as on vertical high-speed SiGe HBTs. Moreover, two IC product application examples are described: protection of an SiGe HBT emitter–base junction (advanced BiCMOS) and ultrathin gate oxide (sub-0.13-µm CMOS). R EFERENCES [1] M. Mergens, C. Russ, K. Verhaege, J. Armer, P. Jozwiak, R. Mohn, B. Keppens, and S. Trinh, “Diode-triggered SCR (DTSCR) for RF-ESD protection of BiCMOS SiGe HBTs and CMOS ultrathin gate oxides,” in IEEE Int. Electron Devices Meeting (IEDM) Tech. Dig., Washington, DC, 2003, pp. 21.3.1–21.3.4. [2] C. Russ, M. Mergens, K. Verhaege, J. Armer, P. Jozwiak, G. Kolluri, and L. Avery, “GGSCRs: GGNMOS triggered silicon controlled rectifiers for ESD protection in deep sub-micron CMOS processes,” in Proc. Electrical Overstress/Electrostatic Discharge (EOS/ESD) Symp., Portland, OR, 2001, p. 22. [3] L. R. Avery, “Using SCRs as transient protection structures in integrated circuits,” in Proc. Electrical Overstress/Electrostatic Discharge (EOS/ESD) Symp., Las Vegas, NV, 1983, p. 177. [4] A. Chatterjee and T. Polgreen, “A low-voltage triggering SCR for on-chip ESD protection at output and input pads,” IEEE Electron Device Lett., vol. 12, no. 1, pp. 21–22, Jan. 1991. [5] A. Amerasekera and C. Duvvury, ESD in Silicon Integrated Circuits. New York: Wiley, 1995. [6] C. Diaz and G. Motley, “Bi-modal triggering for LVSCR ESD protection devices,” in Proc. Electrical Overstress/Electrostatic Discharge (EOS/ESD) Symp., Las Vegas, NV, 1994, p. 106. [7] M. Mergens, C. Russ, K. Verhaege, J. Armer, P. Jozwiak, and R. Mohn, “High holding current SCRs (HHI-SCR) for ESD protection and latchup immune IC operation,” in Proc. Electrical Overstress/Electrostatic Discharge (EOS/ESD) Symp., Charlotte, NC, 2002, p. 73. [8] A. Salman, R. Gauthier, C. Putnam, P. Riess, M. Muhammad, M. Woo, and D. Ioannou, “ESD-induced oxide breakdown on self-protecting GG-nMOSFET in 0.1-µm CMOS technology,” IEEE Trans. Device Mater. Reliab., vol. 3, no. 3, pp. 79–84, Sep. 2003. [9] J. Wu, P. Juliano, and E. Rosenbaum, “Breakdown and latent damage of ultrathin gate oxides under ESD stress conditions,” in Proc. Electrical Overstress/Electrostatic Discharge (EOS/ESD) Symp., Anaheim, CA, 2000, p. 287. [10] K.-C. Hsu and M.-D. Ker, “SCR device with double-triggered for on-chip ESD protection in sub-quarter-micron silicided CMOS processes,” IEEE Trans. Device Mater. Reliab., vol. 3, no. 3, pp. 58–68, Sep. 2003. [11] H. Gieser and M. Haunschild, “Very-fast transmission line pulsing of integrated structures and the charged device model,” in Proc. Electrical Overstress/Electrostatic Discharge (EOS/ESD) Symp., Lake Buena Vista, FL, 1996, pp. 85–94. Markus P. J. Mergens received the degree in physics in 1996 from the Technical University of Aachen (RWTH), Germany. In 2001, he received the Ph.D. degree in electrical engineering from the Swiss Federal Institute of Technology, ETH, Zurich, for his work conducted at the Robert Bosch GmbH, Germany. In 2000, he joined Sarnoff Corporation in Princeton, NJ, where he was the Project Manager responsible for electrostatic discharge (ESD)/input–output (I–O) product design programs with Japanese and US customers. He transferred to Sarnoff Europe, Belgium, in 2003. Since February 2005, he has been with Infineon Technologies in Munich, Germany, as a Manager of ESD Engineering in the automotive and industrial product area. 541 He has around 10 patents and authored/coauthored more than 20 articles in the field of ESD protection design and simulation. Dr. Mergens received the Best Presentation Award for a presentation at the EOS/ESD Symposium in 2001. Since 2001, he has been an active contributor to the EOS/ESD Symposium’s TPC while supporting ISCAS and IRPS ESD/LU subcommittees as well. Christian C. Russ received the M.S. and Ph.D. degrees in electrical engineering from the Technical University of Munich (TUM), Germany, in 1991 and 1999, respectively. His Ph.D. thesis was entitled “ESD-Protection Devices for CMOS Technologies: Processing Impact, Modeling, and Testing Issues.” He worked at the TUM on modeling of ESD phenomena and pulse-measurement techniques from 1991 to 1994, and was with the R&D Laboratory IMEC, Leuven, Belgium, on a fellowship from the European Community for research on ESD and reliability issues in CMOS technologies from 1994 to 1998. From 1998 to 2003, he worked for Sarnoff Corporation, Princeton, NJ, on the creation of innovative on-chip ESD protection and was responsible for ESD technology transfer and licensing programs. In 2003, he moved back to Munich, joining Infineon Technologies as a Senior Staff Engineer involved in the development of ESD-protection concepts for deep submicrometer CMOS (down to the 32-nm node) and bipolar processes. He has published over 30 conference and journal publications and has been awarded over 10 patents in his field with others pending. Dr. Russ was a recipient or corecipient of several Best Paper/Best Poster/Best Presentation Awards at the ESREF Conference (1993, 1995) and at the EOS/ESD Symposia (1993, 1996, 1998, 2000, and 2001). He is a member of the ESD Association, where he is active in the EOS/ESD Symposia. He was also the Chairman of the Technical Program Committee in 2003, as well as a session moderator and workshop panelist. Koen G. Verhaege received the M.Sc. degree in electrical and mechanical engineering from the University of Leuven, Belgium, in 1991. He is currently leading Sarnoff Europe BVBA, Belgium, a company fully focused on value-added silicon IC design solutions intellectual property: innovation, transfer and licensing, and a subsidiary of Sarnoff Corporation (Princeton, NJ). He has authored about 20 peer-reviewed published articles in the field of on-chip ESD protection and testing and holds several ESDprotection design patents. John Armer is a Member of the Technical Staff at Sarnoff Corporation, Princeton, NJ. He joined Sarnoff Corporation (formerly, RCA David Sarnoff Research Center) in 1979. Since then, he has been involved in a variety of analog and mixed signal IC design projects utilizing both custom and languagebased design methodologies. While continuing to be actively involved in those activities, he became a member of the ESD device design team in 1999 where he focuses on practical implementations of ESD-protection structures in mixed signal ICs. He is a coauthor of four papers on ESD, three papers in the field of video processing, and currently has four U.S. patents with several pending. Phillip C. Jozwiak received the degree in physics from Maria CurieSclodowska University, Lublin, Poland, and the M.S. degree in physics from Rutgers University, New Brunswick, NJ. From 1978 to 1980, he worked at BMI Textron, FL, on automation of sputtering systems and development of thin film coatings for photomasks. In 1980, he joined the RCA Solid State Division, Florida, where he was engaged in all phases of processing for very large scale integration (VLSI) fabrication. In 1981, he transferred to RCA Research Laboratories (currently Sarnoff Corporation) in Princeton, NJ, where he has worked on the development of new processes for hybrid Microwave Circuits and GaAs metal-semiconductor fieldeffect transistors (MESFETs) monolithic microwave integrated circuit (MMIC) technologies. He is currently an Associate Member of the Technical Staff at Sarnoff Corporation. His work also involves testing, measurement analysis and layout of the ESD-protection structures and I/O pads. He holds five U.S. patents and is a coauthor of several technical papers. Presently, his interests are the development of on-chip ESD-protection devices for new and emerging CMOS technologies. 542 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 5, NO. 3, SEPTEMBER 2005 Russell P. Mohn (M’01) received the B.E. degree in electrical engineering from the Cooper Union for the Advancement of Science and Art, Cooper Square, NY, in 2001. At the Cooper Union for the Advancement of Science and Art, he focused on signal processing, multiresolution analysis, and artificial neural networks. At Sarnoff Corporation, Princeton, NJ, he has contributed in the areas of mixed signal circuit design and the design of on-chip devices that protect against ESD. Bart Keppens received the I.E. degree in electronics from the Polytechnical School Groep T, Leuven, Belgium, in 1996. In 1996, he joined IMEC and was responsible for device electrical characterization, support for the ESD group and support for the Non Volatile Memories group. Since May 2002, he has been with Sarnoff Europe, Belgium, solving ESD-related problems for customers worldwide and is now Director for Technical Operations at Sarnoff Europe. In 2001 and 2002, he acted as a Workshop Panelist on TLP during the EOS/ESD Symposium. His master thesis, together with colleague S. Servaes, “Transmission Line Pulsing (TLP) technique for analyzing ESD reliability,” performed at IMEC, Leuven, Belgium, received the BARCO award for best Industrial Engineer thesis in 1996. Cong Son Trinh received the M.S. degree in software programming from the High Institute of Applied Microelectronics (ISMEA), Marseille, France. From 2000 to 2002, he was responsible for ESD-protection development at ALCATEL Microelectronics, Oudenaarde, Belgium. Then he joined Sarnoff Europe, Gistel, Belgium, as a Member of the Technical Staff and shortly thereafter, Sarnoff Corporation, Princeton, NJ, where he is responsible for ESD/I–O design programs for customers worldwide. About Sofics Sofics (www.sofics.com) is the world leader in on‐chip ESD protection. Its patented technology is proven in more than a thousand IC designs across all major foundries and process nodes. IC companies of all sizes rely on Sofics for off‐ the‐shelf or custom‐crafted solutions to protect overvoltage I/Os, other non‐ standard I/Os, and high‐voltage ICs, including those that require system‐level protection on the chip. Sofics technology produces smaller I/Os than any generic ESD configuration. It also permits twice the IC performance in high‐frequency and high‐speed applications. Sofics ESD solutions and service begin where the foundry design manual ends. ESD SOLUTIONS AT YOUR FINGERTIPS Our service and support Our business models include • Single‐use, multi‐use or royalty bearing license for ESD clamps • Services to customize ESD protection o Enable unique requirements for Latch‐up, ESD, EOS o Layout, metallization and aspect ratio customization o Area, capacitance, leakage optimization • Transfer of individual clamps to another target technology • Develop custom ESD clamps for foundry or proprietary process • Debugging and correcting an existing IC or IO • ESD testing and analysis Notes As is the case with many published ESD design solutions, the techniques and protection solutions described in this data sheet are protected by patents and patents pending and cannot be copied freely. PowerQubic, TakeCharge, and Sofics are trademarks of Sofics BVBA. Version May 2011 Sofics Proprietary – ©2011 Sofics BVBA Groendreef 31 B‐9880 Aalter, Belgium (tel) +32‐9‐21‐68‐333 (fax) +32‐9‐37‐46‐846 bd@sofics.com RPR 0472.687.037