Correlation Between Low-frequency Noise Overshoot In SOI MOSFETs And Frequency Dependence Of Floating Body Effect

Abstract

In this paper, a new mechanism is proposed to explain the well-known kink-related noise overshoot in SO1 MOSFETs. We found that there is a correlation between the frequency dependence of the kink’s onset voltage and the characteristic frequency (ao) dependence on drain bias of Lorentzian-like noise overshoot. It was concluded that the noise overshoot is due to the frequency dependence of floating body effect, which amplifies the noise arising from electronic tunneling transitions ( zT = 2n/w,) between front interfacial oxide traps and the channel. Introduction Recently SO1 MOSFEiTs have been proposed as a candidate for high speed communication applications. Low-frequency noise is an important consideration for these analog circuits. For example, low-frequency noise can be up-converted in RF mixers and oscillators, resulting in phase noise [ l ] . Due to the floating body, SO1 MOSFETs exhibit a low frequency kink-related noise overshoot, which has a Lorentzian spectrum: a flat low-frequency plateau, with constant amplitude, followed by a l’ roll-off, illustrated in Fig. 1. Different mechanisms were proposed to explain this excess noise, such as the trap-assisted generation-recombination noise model [2] and the G-R noise induced by back interface state [3]. However, these models cannot completely explain the characteristic frequency (wo) dependence on drain bias of overshoot noise, as shown in Fig. 4. In this paper we propose another mechanism to explain the above noise overshoct phenomenon. A similar noise behavior was also found in bodytied devices, but the amplitude is significantly reduced. Experimental Result A. Device description and low-frequency noise measurements Near fully depleted thin film SO1 nMOSFETs were used in the study [4]. An extra p+ implant step is applied on the source to form sourcebody-tied (SB-tied) SO1 nMOSFETs. SB-tied devices eliminate the kink at the output characteristics as shown in Fig. 2. Low-frequency noise measurements were made using an HP 3561A Dynamic Signal Analyzer with the gate electrode of the MOSFET ac shorted to ground by a capacitor. The output noise power is then transferred to inputreferred gate noise power. B. Drain bias dependence of noise overshoot in the kink region The devices were biased in the saturation region with low VGT, where the flicker noise is dominated by the number fluctuation model [5]. The Lorentzian spectrum noise overshoot was observed in the floating body SO1 nMOSFET (Fig. 4.a). As bias voltage increases, 0, increases and the noise level of the plateau decreases. Plotting the frequency times S,, versus frequency (Fig. 4.b), the characteristic frequency (w,) on drain bias could be determined and listed in Table 1. It is important to note that such noise overshoot behavior exists even for body-tied devices. The same drain bias dependence phenomena were observed (Fig. 5 ) with smaller overshoot amplitude and weaker drain bias dependence of a,,. Discussion C. Frequency dependence offloating body efect Onset voltage of the kink effect, which occurs as body voltage changes from low to high state, increases with frequency as illustrated in Fig. 3. This is due to the sourceibody junction capacitance acting as a low pass filter for the holes [6] . In order to suppress the floating body induced instability, SB-tied SO1 devices were fabricated without the I-V kink characteristics. However, these devices still show a smaller Lorentzian-like noise overshoot (ASvG) and weaker drain bias dependence of 0,. This implies that the floating body effect, which is due to body charging, still occurs as the body ties do not provide a zero impedance path for the holes. D. The correlation between frequency dependence of the kink effect and drain bias dependence of CO, Coupled wo dependence on drain bias (0, increases as VDs increases) with the frequency dependence of kink’s onset voltage (V&) increases as frequency increases), it suggests that there is a strong correlation between kink’s onset voltage and noise overshoot’s coo. Alternatively, in the frequency domain, at a constant bias there is a frequency (ao) corresponding to the kink’s onset voltage where body voltage quickly changes. The power spectral density of the fluctuation in the number of trapped electrons in the gate oxide(AV, AE) is given by [5], where zT = 2x/0,. Integrating the noise spectrum, the flicker noise (U’ given by the number fluctuation model can be obtained. Based on the correlation established above, the mechanism of noise overshoot can be explain as follows. At a constant drain bias larger than the DC I n k ’ s onset voltage, there is a frequency w0(VDs) at which body voltage induces the kink effect and amplifies the signal of frequency a,, in the channel. It enhances the noise event in (1) by a gain factor of A(w,) with characteristic time equal to zT (= 2nlw,). Finally, the total drain noise power is given by, 1 1 1 S,, = C , X + ( A 1 ) .C,X--. 2 ’ (2 ) f 1 + (w/wn) where C1, C2 are constants. The noise spectrum (2) explains the Lorentzian-like overshoot spectrum and the excess noise reduction as V,, increases. It can also predict the result by the spot noise analysis that as frequency increases, the noise overshoot’s peak shifts toward higher VDs and its amplitude reduces. Conclusion This paper presents the mechanism of the kink-related noise overshoot. It is due to the frequency dependence of the floating body effect which amplifies the tunneling noise with characteristic frequency respective to the frequency at which the kink occurs for a given bias voltage. Therefore, the key to avoiding the Lorentzian-like noise overshoot spectrum is to either suppress the floating body effect or optimize CO,, with the 99 4-93081 3-75-1 I97 1997 Symposium on VLSl Technology Digest of Technical Papers Floating body SO1 nMOS vDS fo ( q)/2x) 0.75V _ _