New Reliability Issues of CMOS Transistors with 1.3nm Thick Gate Oxide

In this paper, we will discuss several new reliability issues facing CMOS transistors with ultra thin gate oxides and their impacts on projection of operation voltage V10Y for 10-year lifetime. The most important findings are: 1). Oxide lifetime is more meaningfully determined by an event taking place much earlier than oxide breakdown: a strongly transistor-size dependent increment in gate leakage current. Contrary to oxide time-to-breakdown TBD , the newly defined oxide lifetime is shorter when the transistor size is smaller. 2). By proper processing of DCIV signal, DCIV technique is demonstrated for direct monitoring and accurate determination of interface trap generations in CMOS devices during stresses even with tox=1.3 nm. 3).Contrary to previous reports on thicker gate oxide, we have shown that Vg=Vd is the worst-case hot carrier degradation condition for both n-MOS and p-MOS, with pMOS showing smaller V10Y than n-MOS. 4). A “dynamic” NBTI during AC stressing is reported for the first time which clearly demonstrates that the conventional (static) NBTI 4 underestimates p-MOS device lifetime. A physical model is developed for this phenomenon. The DNBTI determines the overall CMOS device lifetime and will have significant impact on projection of maximum operating voltage in practical dynamic operation of digital circuits. Device fabrication: CMOS devices were fabricated using standard dual-gate CMOS technology. Gate oxide of 1.3 nm thickness was grown by RTO followed by an exposure to high-density nitrogen plasma. Interface traps measurement in CMOS devices with 1.3 nm gate oxide thickness: For n (p)-MOS, in the range of Vg=0 to -1(+1) V where the DCIV Ib peak appears, the intrinsic tunneling bulk current Ib corresponds to bulk to gate hole (electron) tunneling and is much lower than the gate current, because the holes (or electrons) face Si energy gap in the poly gate. The gate-to-interface trap tunneling current 5 is eliminated by deducting ∆Ig from ∆Ib. Further, by increasing the forward bias, the disturbance of drain-to-bulk thermal-trap-tunneling effect 6 is reduced and clear DCIV peak ∆IDCIV that is proportional to the number of interface traps ∆Nit can be successfully obtained. V 10Y projections with different stress conditions: Oxide degradation: For 1.3 nm gate oxide, TBD is too long to be considered as a limiting factor of device degradation. Instead, a steady increase in gate leakage Ig taking place much earlier than breakdown is observed. The increase of leakage current density Jg shows a strong area dependence, which implies that the oxide degradation is very much localized, unlike the uniformly distributed SILC in thick oxide 7,8 . Contrary to TBD as described in [1], the oxide lifetime defined by the time it takes to reach 100% increment in Ig is shorter when the transistor area is smaller. Hot-carrier degradations: gm and IDCIV variations under hot carrier stress conditions with Vg@Ib,max and Vg=Vd were compared. For n-MOS, the worst case is Vg@Ib,max for long channel devices and Vg=Vd for short channel devices, while for p-MOS, all devices show the worst degradation at Vg=Vd. The gm dependence on ∆Nit (∆IDCIV) was also monitored. The discrepancy between short and long channel devices can be attributed to the different ratio of channel to S/D resistance RCH/RSD. Negative Bias Temperature Instability (NBTI) for pMOS: NBTI in p-MOS has been suggested as the dominant factor for CMOS reliability 4 . We have observed a very interesting phenomenon of NBTI during a “dynamic” stressing. When an electric field of opposite polarity is applied during NBTI stressing, a reduction of ∆Nit is observed. This is interpreted by the interaction between Nit and hydrogen species. During negative biasing, the hydrogen is released from Si-H bond at the SiO2/Si interface, moving towards the gate electrode and resulting in ∆Nit. When the bias polarity is reversed, the hydrogen is driven back to the SiO2/Si interface under the applied electric field and passivates the Si dangling bond, resulting in ∆Nit deduction. We have shown that the “electrical annealing” of interface states can be very significant for p-MOS operating in a CMOS inverter, and hence the device lifetime under such “dynamic” NBTI stress can be much longer than that projected under the conventional “static” NBTI stress. Acknowledgement Devices used in this study were fabricated by CSM. References: [1]J.H.Stathis, IRPS 2001, p.132. [2] A.Neugroschel et al, IEEE TED, v.42, p.1657, 1995. [3] E.Li et al, IEEE TED, v.48, p.671, 2001. [4] N.Kimizuka et al, VLSI Tech. 1999. [5] A.Getti et al., VLSI Tech. 2000 ,p.218 . [6] G.Chen et al, IEEE EDL, v.22, p.233, 2001 [7] Y.Wu et al, IEEE EDL , v.20,p/262,1999. [8] D.J. Dumin et al, IEEE ICMTS 1991, p. 61.