Solid-State Electronics Vol. 36, No. I 1, pp. 1593-1596, 1993 0038-1101/93 $6.00 + 0.00 Printed in Great Britain. All rights reserved Copyright © 1993 Pergamon Press Ltd BACK-GATE INDUCED RANDOM TELEGRAPH SIGNAL NOISE IN FULLY-DEPLETED SILICON-ON-INSULATOR n MOSFETs E. SIMOEN, U. MAGNUSSON~', J. VERMEIREN a n d C. CLAEYS IMEC, Kapeldreef 75, B-3001, Leuven, Belgium (Received 27 October 1992; in revisedform 5 April 1993) Abstract--Back-gate induced Random Telegraph Signal (RTS) noise in the front-channel drain current is demonstrated for the first time experimentally, in fully-depleted Silicon-on-lnsulator(SOl) n MOSFETs. The RTS is observed when the back-interface is in accumulation and is absent for zero, or positive back-gate bias. As will be demonstrated, hole-trapping at the back-interface occurs, generating a low-frequency, large-amplitude drain current step. This trapping is related to mobile charges in the back-oxide, which therefore causes a metastable switching of the threshold voltage. l. INTRODUCTION Low-frequency (LF) noise spectroscopy is becoming an increasingly popular tool for the study of devices fabricated in Silicon-on-Insulator (SO1) sub- strates[l-4]. The technique not only allows the as- sessment of the front- and the back-interface quality[4], but it enables one to evaluate the defective- ness of the film, for example by studying the magni- tude of the kink-related excess noise[5,6]. More interestingly, quite often Random Telegraph Signals (RTSs) are observed in SOI MOSFETs[5-9], even for fairly large device dimensions. In conventional sub- micron MOSFETs it has been convincingly demon- strated that RTSs are due to trapping by a defect centre in the vicinity of the Si-SiO2 interface[10,11]. In this paper, it will be shown that alternative mech- anisms may generate a discrete RTS-like step in the drain current Iv of an SOI transistor[8,9]. One mech- anism will be illustrated here for a fully-depleted nMOSFET and corresponds with back-gate induced RTS. At the same time, it will be shown that the LF noise of fully-depleted transistors increases when operating with the back-interface in accumulation. 2. EXPERIMENTAL The devices studied are processed in a double-poly 3 #m thin-film SOI CMOS technology. The starting material is commercially available 125 mm SIMOX wafers and the transistors have a designed area of W x L = 20 gm x 3 #m. The final Si film thickness is 100nm; the front-oxide thickness 50nm and the back-oxide thickness is 400 nm. The low-frequency noise measurements are performed using a HP 3562 dynamic signal analyser. A home-built, low-noise ?Presently at the Institute of Microelectronics, Kista, Sweden. biasing circuit enables the selection of a specific operation point in the linear region (constant drain voltage VDs), or in saturation (constant drain current It)). The RTSs are monitored simultaneously on an oscilloscope. 3. RESULTS Typical low-frequency noise spectra of the SOI nMOSFETs, recorded at a drain current ID = 50 #A are shown in Fig. 1, and correspond to zero back-gate bias V~ (curves a and e), or with the back-gate in accumulation (V~a =-20 V; b, c and d). A clear increase in the LF noise for the V~ = - 20 V case is obvious from the figure. This effect is even aggravated for higher drain voltages, enabling the generation of multiplication current at the drain, e.g. at Vas = 4 V, as shown by curves c and d. The corresponding spectra show a tendency to level off at low frequency f and roll-off according to Ill ~, with v between 0.5 and 1 (drain current noise). At the same time, huge RTSs emerge in the oscilloscope trace, as evidenced by Fig. 2: in Fig. 2(a), the oscilloscope picture for VaG = 0 V and Vas = 1 V is shown, while Fig. 2(b) corresponds to V~ = - 20 V and VDS= 4 V. In the latter case, a clear step in the drain current is obvious, with an amplitude of approximately 2/~A, i.e. 4% of the total drain current. This RTS shows metastable behaviour, i.e. it may switch to one (or more) RTSs with a different amplitude and different capture (down) and emission (up) times, as shown in Fig. 3(a) and (b), recorded for the same settings as Fig. 2(b), but at different moments in time. Further evidence for this is presented by the spectra c and d in Fig. 1 corresponding with a different RTS configuration. In some cases, the RTSs disappear completely from the spectrum after some time. Changing the negetive back-gate bias back to zero, a lower noise value is typically found than before the 1593