Theory of “Current-Ratio” Method For Oxide Reliability: Proposal and Validation of a New Class Two-Dimensional Breakdown-Spot Characterization Techniques M. A. Alam#1, D. Monroe#2, B. E. Weir#2, and P.J. Silverman#2 ECE Department, Purdue University W. Lafayette, IN-47906 (765-494-5988/[email protected]) #2 Agere Systems, Allentown, PA18109 #1 Abstract A theory of the Current-Ratio technique, which is widely used to locate gate oxide breakdown spots in one dimension (i.e., distance from source or drain), is proposed and verified. The theory shows that the Current-Ratio method is a special case of generalized van der Pauw technique, and as such, can easily be generalized to locate oxide breakdown spots in two dimensions. We develop the theoretical framework of this new class of breakdown-spot characterization techniques and then validate the theory by experiments. We conclude by discussing the implications of locating breakdown spots in two dimensions for reliability projections of ultra-thin gate oxides. Introduction The 1D “current-ratio” technique (1D-CR) has been widely used to locate breakdown spots (BD-spots) in thin gate dielectrics and to interpret oxide reliability [14]. Yet surprisingly, no one has justified the technique theoretically. As shown in Fig. 1(a) the idea of 1D-CR is simple [1]: once an oxide develops a soft breakdown spot with only a localized increase in gate current (IG), the device can be biased in accumulation such that IG is balanced by source (IS) and drain (ID) currents. And the ratio of source to drain-current locates the BD-spot, i.e., x/L=ID/(IS+ID) . (1) Fig. 1(a): Side view of an NFET biased in accumulation. IG is injected at the BD-point x. (b): n1 is the hole density at the BD-spot and S/D are perfect sinks. The slopes of the carrier concentrations determine IS and ID. (c, d) Top view of the oxide (dashed regions are S/D). Two different BD-spot distributions are indistinguishable by 1D-CR (same x, but different y), yet would have very different soft-BD lifetimes ([4], see Fig. 11 for reliability implications). Yet, the prescription does not tell us (a) If the formula is exact, (b) If it is valid in inversion, (c) If it is limited to 1-D, and if not, if it can be generalized. This 2D generalization is crucial because 1D-CR will lead one to incorrectly conclude that pair distances between BDspots 1,2,3 in Figs. 1c and 1d are identical (1D-CR measures x only!), although BD-spots in 1c is significantly more localized (therefore has reduced TDDB lifetime [2-4]) than those in Fig. 1d. The theory proposed here provides the first interpretation of 1D-CR, and helps develop several 2D techniques to determine both x & y coordinates of SBD-spots. Theory of the existing technique Since J ( x, y ) = qnun ∇φ + qDn ∇n and ∇ • J ( x, y ) = 0 , nun ∇2φ + Dn ∇2 n = 0 . (2) In accumulation (Fig. 1a), the electric field in small (except near the source and drain) and minority (hole) diffusion dominates, so that (3) Dn ∇2 n = 0 . Fig. 1b shows that the solution of Eq. (3) in 1D is given by , IS=qADnn1/x and ID=qADnn1/(L-x) and their ratio satisfies Eq. 1. This derivation is simple, exact, and wrong! Fig. 2(a) explains why: the 1D derivation assumes line injection (the side-view of Fig. 1a is misleading), while calculation of current from a BDspot requires a 2D solution (Fig. 2b). Surprisingly, however, both the Monte-Carlo solution of Eq. 3 (a) (b) (d) (c) Fig. 2: Top view of oxide: (a) 1D analysis of CR method assumes a line charge, while (b) actual diffusion from the BD-point to S/D (dashed region) is 2D. However, 2Ddiffusion (c) can be mapped onto 1D diffusion with variable delay (d) – explaining why the 1D solution of CR (a) gives the exact result for the 2D problem. Before we verify Eq. (5) experimentally, note that instead of determining the VD (VS) by injecting IS (ID) at each x, the Reciprocity Theorem allows determination of all VD(x) in a single calculation with I injected at the drain (source) to determine V(x)=VD(x) [red line with filled square]. Later, we will use this theorem to calculate the potential induced in 2D geometry. Fig. 3: Both the Monte-Carlo method (diffusion simulated by following random walk of particles in 2D and 3D from BD- point in 2D) and the Image Charge method (multiple reflections on S/D from a point charge) confirm 1D-CR (Eq. 1). (computed by injecting particles at each point (x,y) and counting the number of particles collected by the source and drain) as well as the image charge solution of Eq. 3 (determined by using the equivalence of diffusion equation and Laplace equation and then obtaining the charge induced in source and drain by iterated images of a point charge placed at (x,y) ) in 2D and 3D still agree with Eq. (1) (see Fig. 3). This paradox is resolved by realizing that a particle injected from a BD-spot and collected, through diffusion, by two infinite parallel plates (S/D contacts) in any dimension can be transformed into an equivalent 1D problem (Fig. 2c): as far as progress toward the electrode is concerned, motion in the y-direction is indistinguishable from no motion. Therefore, y-motion corresponds to a delay for particle collection. Thus the 1D-CR technique is exact in any dimension, for infinite electrodes. Fig. 2a gives the correct result because each point of the injection line satisfies the 1D-CR individually. Fig. 5 shows that locations of BD spots, determined by the standard 1D-CR technique and 1D-VR technique (open circuit voltage ratio technique) just developed, are identical for all gate voltages beyond threshold. The VGindependence (above threshold) reflects the fact that it is the ratio of open-circuit voltages (in 1D-VR, Eq. 5) or the ratio of short-circuit currents (in 1D-CR, Eq. 1), rather than their absolute values, that determine the location of BD-spots. Below threshold (~0.5V), the agreement between the techniques is not as good, Fig. 4: With BD-spot located at x and the gate held at a certain bias, the source grounded, and the drain floating (open-circuited), the potential profile in the channel is given by the solution of Eq. 4 (black line with open circles). The floating drain voltage, VD, equals that of the BD-spot, because no current flows between drain and BDspot. The Reciprocity Theorem provides an efficient alternate method to calculate the VD(x) [red filled squares]. Generalization to inversion With the theory of 1D-CR established, we now generalize the technique to inversion, where Eq. 2 simplifies to (4) µn n( d 2φ / dx 2 ) = 0 because inversion is drift-dominated and the diffusion component is negligible. If one monitors Drain opencircuit voltage VD with IS flowing between BD-spot and Source, then VD=αISx/LC because no current flows between drain and BD-spot and the line joining them must have the same potential (black line with circles in Fig. 4). A similar measurement of current flow between drain/BD-spot determines the floating source voltage, VS=αID(LC-x)/LC. The ratio eliminates the constant α and locates the BD-spot, i.e., x/L=VDID/(VDID+VSIS). (5) Fig. 5: Locations determined by 1D-CR and 1D-VR techniques give identical results indicating equivalence of the techniques. Fig. 6: BD-spots located by 1D-CR and 1D-VR agree within the margin of error (as predicted by the theory). Results from 5 different devices are shown. because subthreshold transport is diffusion-dominated and 1D-VR can not be used. Fig. 6 summarizes the measurements for several different devices, indicating that in all cases the measured locations of BD-spots by these two techniques are essentially identical. Fig. 7: Four different test structures to locate (x, y) positions of a BD-spot (red square). (a) Tester 7a adds y-contacts (probes 3&4) to two standard x-contacts in 1D-CR, (b) Tester 7b simplifies Tester 7a by reducing probes 3 & 4 to point-probes). (c) The four point-contacts of Tester 7c are located at the corners. (d) Tester 7d indicates, like a general 4P-VDP technique, that the BDspot can be located by arbitrarily-placed probes. Generalization beyond 1D Fig. 7(a-d) shows four different test structures, any of which can be used to localize a BD-spot (indicated by red square) in 2D and they are all conformationally equivalent to 5-probe van der Pauw (5P-VDP) [5-7]). The theory of each of the testers is discussed below, because some may be easier to implement than others. Our measurements are based on test structure 2 (Fig. 7b). For the first test structure in Fig. 7(a), two “S/D” contacts in the y-direction allows numbering of probes (P) by respective current density, J3>J1>J2>J4. The BDcoordinates, x=0.5+[0.5J2/(J1+J2)]0.5 & y=[0.252(J1+J2)(0.25-x2)]0.5, are obtained by solving Eq. 4 (see Fig. 8 for validation by Monte-Carlo and [8,9] for special cases). One might wonder if measurement of only two currents, J1 and J2, is sufficient to ‘triangulate’ a point in 2D. The key to this puzzle is that J1 and J2 can be specified only if J3 and J4 are also measured, which provides the additional required information for ‘triangulation’ of the BD-spot. J1 ≅ (1 − y 2 ) (1 + x − x 2 ) (1 − x 2 ) J2 ≅ (1 − y 2 ) (1 − x + x 2 ) (1 − x 2 ) The test-structure in Fig. 7b simplifies 7a by reducing the top/bottom “S/D” contacts to point-probes. The xposition is obtained by 1D-CR or 1D-VR (with P3 & P4 floating), and R3,4=y/W=(V3-V4)/2(V3+V4). Fig. 9 provides numerical (Image Charge) and experimental verification. The test structure in Fig. 7c reduces all contacts to point-probes. Measurement of V1,3 (P2 & P4 sinking current from BD-spot) and V1,2 (P3 & P4 current sinks) locates the 2D-BD spot (Fig. 10). Finally in Fig. 7d, k=V2,3/ρ I. ρ is measured, before breakdown, by 4P-VPD. After breakdown, V2,3 is Fig. 8: J1 and J2 for test structure in Fig. 7a are plotted along y for fixed x (symbols: Monte Carlo, solid lines: analytical solution (Eq. in inset). Note that by symmetry, the solution needs to be plotted in only the triangular region of Fig. 7a. Therefore, in the plots, initial y-values increase with increasing x. two BD-spots are uncorrelated in 2D, their distance PDF p(l) is shown in Fig. 11 [10,11]. Any deviation between theoretical and measured p(l), by any of the techniques discussed in Fig. 7, would establish the degree of 2D-correlation of BD-spots, with corresponding modification of post-SBD oxide lifetime. p (l ) = 4l [1 − (0.5 + 0.25π )l + l 2 / 3] Fig. 9: 2D-position measured by test structure 7(b). Solution of left half of the test-structure is plotted. 1D-VR based on P1, P2, & BD-point determines BD-spot (dotted lines: V2=VX=0:0.05:0.5. The induced voltage between P3 & P4 determines the ylocation (solid lines: R3,4=-0.5:0.033:0.5. Measurement of locations of three BD-spots by two methods (open circles: VR, closed-box: CR) coincide, which validates the technique. 0< l <1 p ( l ) 2.5( 2 − l ) 3 for 1 < l < 2 Fig. 11: The approximate analytical PDF of pair distance l2=(xi-xj)2+(yi-yj)2 is supported by exact numerical calculation. (xi,yi) are locations of i-th BDspot (see Fig. 1c and 1d). The 2D-location determination techniques discussed in Figs. 7-10 can be used to explore deviation from this PDF to establish 2D-correlation. Summary Fig. 10: 2D-position determination by test structure 7(c). (solid-lines): P1 & P3 are V-probes and P2 & P4 are I-sinks from BD-point. (dashed-line): P1 & P3 are V-probes and P2 & P4 I-probe provides 2nd measurement. Contour values indicate the location of the BD-spot for measured V13 and V24. measured between P2 & P3 and I between the BD-spot and P1. Since k is a known function of the locations of P1, P2 & P3 [5-7], the BD-spot is uniquely determined. Reliability implications The lifetime after SBD depends on the spatial and temporal correlation of the BD-spots [3,4]. We have previously shown that BD-spots are temporally uncorrelated [3,4], but spatial decorrelation could be proved, by 1D-CR, only in 1D. This does not exclude the possibility of 2D correlation (Fig. 1c vs. 1d). The “5P-VPD” methods [7a-d] offer a new class of techniques to detect such correlation. Theoretically, if We have provided the first theory of position measurement by the current-ratio method, which resolves paradoxes and establishes its validity on a par with SILC, CP, etc. Our theory allows generalization of the 1D-CR to inversion and to 2D (four techniques, all conformationally equivalent to 5P-VPD, are proposed). This new class of techniques will help interpret BD-spot distributions in 2D to better predict gate oxide reliability. Acknowledgement: M. Alam gratefully acknowledges financial support from NCN (#671-1211-4533) and NSF/SRC (#2004-HJ-1238). References [1] R. Degraeve et al., TDMR, 1(3), 163, 2001. [2] J. Suehle et al., Micro. Reliability, 419, 2005. [3] M. Alam et al., Nature, 6914, 378, 2002. [4] M. Alam and K. Smith, IRPS Proc., 406, 2003. [5] D. S. Perloff, Solid State Elec., 20, 681, 1977. [6] L. J. van der Pauw, Philips Res. Rep, 13, 1, 1958. [7] M. Yamashita et al., Jap. J. of Appl. Phys, 23, 1499, 1984. [8] B. Berndt, ‘Ramanujan: Letters and Comments’, (3rd Letter to Hardy), 1995. [9] F. Bornemann, “The SIAM 100-digit Challenge”, Problem 10, SIAM, 2004. [10] W. Donath, SIAM J. Appl. Math, 16, 439, 1968. [11] P. Christie et al., Trans. VLSI Sys., 8, 639, 2000. x2 )
© Copyright 2026 Paperzz