3.1.1 Charge-Pumping Techniques

In this section a review of the charge-pumping techniques used to characterize the interface traps is given. Other techniques to analyze interface states, which are applicable to small MOSFETs, are pointed out. If not explicitly said, we assume -channel devices.

All charge-pumping techniques have one common property: they are direct methods to sense the interface states. In addition, they can be applied directly on small devices. Three different groups of the charge-pumping techniques have been proposed in the literature:

1. Large-signal techniques: Their general feature is that the gate-pulse amplitude is larger than the difference between the threshold and the flat-band voltage (Figures 3.2, 3.3). All four processes, capture and emission of both electrons and holes, take place in these methods. We distinguish between two variations:
   1. The trapezoidal gate-pulse waveform is the most frequently applied, Figure 3.2. This waveform is essentially that applied by J.Brugler and P.Jespers in [43] and investigated further by A.Elliot [117]. A thorough theory is given by G.Kaden and H.Reimer [233][232] using the theoretical investigation of the emission processes by J.Simmons and L.Wei [435][434]. Based on these results, an engineering and very useful model is proposed by G.Groeseneken et al. in [154]. A quite general but hardly practical model is developed by U.Cilingiroglu in [49]. The basic physical phenomenon involved in the determination of the charge-pumping current is the non-steady-state emission of electrons and holes during the falling () and rising () edge of the gate pulse, respectively (Figure 3.2). The emission times are controlled by the rise time and by the fall time . Roughly speaking, does not depend on the surface potential and the bulk doping. By this technique it is possible to measure:
      (1) an average interface-trap density in the active energy interval.
      (2) the interface-trap distribution in the energy across the whole band gap, except close to the midgap (approximately from at room temperature) and close to the band edges (within above the valence band and below the conduction band ), by varying the rise time and the fall time . To penetrate closer to the midgap, very long and are necessary, which leads to very low because of the low frequency. During increasing temperature the part closer to the midgap can be scanned, but the inverse leakage currents of the junctions become a limitation. The scanning close to the band edges is limited due to the very short required or which can produce the geometric component (effect 2; investigated in Section 3.4), and due to influence of the steady-state emission at such short times (Section 3.3).
      (3) by the triangular variant of the large-signal technique the geometric mean value of the capture cross-sections can be extracted [233][154]. It should be stressed, however, that a slow variation of with energy is a prerequisite for the proper extraction of in this way.
      (4) the spatial trap distribution (along the interface) can be extracted, close to the drain and source junctions (Section 3.5.2).
      A further refinement of the large-signal charge pumping represents the variant (Figure 3.3), proposed by G.Przyrembel et al. [377] and later in [206]. A mid level is introduced to precisely control the electron (hole) emission time which equals to . Typically holds. Here, one process strongly dominates over other three during the top level , mid level and the bottom level (electron capture, non-steady-state electron emission and hole capture, respectively). By varying one can determine . Using three-level pulses it is possible to penetrate a bit closer to the midgap than by using trapezoidal pulses.

       

   2. It employs the three-level gate pulse shown in Figure 3.3. This method is proposed by W.Tseng in [479] and applied in [83]. By holding the mid level () sufficiently long, a steady-state value of the quasi-Fermi level is established at the interface. This value determines the energy in the band gap , where is to be measured. By varying the mid level the energy can be changed in the upper part of the band gap. can be determined by this technique without any knowledge of the capture cross-sections . However, in this technique it is necessary to determine the level for the particular . This can be done by the standard Q-V or QS - methods (pages 92-95 in [331]), or by analytical calculation. Note that the surface-potential fluctuations could have an influence in this charge-pumping technique.

       

   3. Improved charge-pumping techniques have recently been proposed, which extend and combine the techniques (a) and (b). A spectroscopic variant of the large-signal trapezoidal method (a) is proposed in [97][96] by G.Van den bosch et al. The variation of temperature is applied to extract , by keeping and constant. A larger part of the band gap can be scanned than by method (a). In its spectroscopic variant (like DLTS), this technique enables one to extract separately both, the electron and hole capture cross-sections and as functions of energy. Unfortunately, these measurements have a low accuracy. In addition, an inconsistency has been systematically observed in the experimental results for the minority carriers in [97].
      The second class of new charge-pumping methods attempting to measure the capture cross-sections is proposed by M.Ancona and N.Saks. They combine the three-level techniques (a) and (b) to extract , and separately [397]. The measured are related to the trap emission processes. Note that experiments clearly shown a unique value for the cross-sections at the specific energy level [397], which is a basic assumption in the theory of the charge-pumping techniques of type (a). In further work, by considering the capture mode in the three-level techniques, they extracted the cross-sections and for the electron and hole capture processes, in addition to ones for the emission processes [395][11]. We will critically evaluate this method in Section 3.4. The approach proposed by M.Kejhar in [243] belongs to this class of advanced techniques too.

   The large-signal techniques have common drawbacks: they suffer from the parasitic geometric component (effect 2 in Section 3.1) and they cannot be employed to extract the trap density and the capture cross-sections at the midgap (which exclusively contribute to the generation current in MOS gated-diodes).
2. Small-rectangular-pulse technique: Unlike the large-signal techniques, the small-rectangular-pulse (SRP) technique is capable to resolve traps at the midgap. The technique is developed by R.Wachnik and J.Lowney [494] and applied later in [495]. The amplitude of the gate pulses is chosen smaller than the difference between the charge-pumping threshold and flat-band voltages, Figure 3.4 (left). The rise and fall times are assumed to be much shorter than the top level and bottom level duration, in order for negligible capture and emission of carriers to occur at the pulse edges. The emission and capture processes determine a narrow energy interval close to the midgap, where the capture processes are dominant: electron capture at the top level and hole capture at the bottom level . To calculate the surface concentrations of electrons and holes, corresponding to and , respectively, need to be calculated. For higher interface state densities, this should be done selfconsistently, because of the charge-potential feedback effect (Section 3.1.3). In addition, the minority-carrier concentrations can be nonuniform along the interface, due to a finite time for the minority-carrier response to the gate-bias variations. Consequently, the capture can be nonuniform during , leading to deviations from the simple theory, as observed experimentally in [494]. The SRP technique is related to the surface potential (i.e. concentration) and, therefore, is sensitive to the surface-potential fluctuations (Section 6 in [331]), as investigated in [494]. By the SRP technique one is able to measure , to evaluate , to propose the sets of traps which fit the experimental characteristics (all this close to the midgap) and to evaluate the standard deviation of the surface-potential fluctuations. Different sets of traps mean that only a rough information on the dispersion of the capture cross-sections can be obtained. The current measured by the SRP method is proportional to the generation current of the depleted interface [133], but the former current is much larger [495] (both currents depend solely on the density of midgap traps with large cross-sections).
3. Pulsed-interface-probing technique: This technique has been developed by U.Cilingiroglu in [51]. In this method we monitor the surface generation current. In conventional measurements of the generation current of the depleted MOS surface, the source and drain junctions are reversely biased with respect to the bulk. As a consequence, the quasi-Fermi levels of electrons and holes are separated at the interface. By conveniently placing the gate bias so that the interface becomes fully depleted, the surface generation current is produced by the electron and hole emission (mostly from the midgap traps) [156]. Emitted holes travel to the bulk (negative ), while emitted electrons are collected by the source and drain junctions (positive ). The generation current is proportional to (all quantities are related to the narrow interval at the midgap). In the pulsed-interface-probing (PIP) method a narrow inverting pulse in duration () is added to the constant gate bias which produces the interface depletion (), Figure 3.4 (right). The level is chosen sufficiently high to fill all traps at the interface by the electron capture during the short pulse . During the depletion time , both, the electron and hole emission take place. It is assumed in [51] that all electrons emitted from the interface during will travel to the source and drain junctions; thus not contributing to the charge-pumping current (). We will critically analyze this assumption in Section 3.4, using a rigorous numerical approach. The difference between the current for the waveform shown in Figure 3.4 and the generation current for constant measured against the depletion time , yields the information on and , separately (for the energy interval about from ). Additionally, dispersion of can be extracted for the midgap traps [51]. Contrary to the SRP technique, the PIP technique does not rely on the surface potential; the PIP is insensitive to the surface-potential fluctuations. Note that the parasitic geometric component (effect 2) can be produced in the PIP technique, when becomes too short.
   The main idea in processing the difference between the currents, measured with the short pulse and without it, is to eliminate all leakages (due to junctions and bulk traps in the depletion region), where it is believed that these effects are not influenced by the presence of the very short pulse . This idea is used in a further work [52] by the same author, to improve the capability of the three-level large-signal technique (a) to penetrate closer to the midgap.

In common, the charge-pumping techniques cannot yield detailed information on the physical properties of traps (at the present stage of development). By measuring the DC component in currents a lot of information on the physical phenomena involved in the effect is lost, in comparison with the techniques which rely on observing some quantity in the time domain, like DLTS adduced below. The charge-pumping measurements are, however, simple and very fast.