oxides, it was found that Qp and Qn are generated near the oxide/substrate interface andQpis initially much larger thanQn. After the continuous stressing,Qpsaturates and moves closer to the interface, butQnkeeps increasing and moves away from the interface, especially for those oxides after the post-poly anneal (PPA) treatment. Qp is very unstable and easily neutralized, either by a small stress of opposite polarity or the same polarity. For the latter,Qp is mainly dependent on the level of the final stressing field.
I. INTRODUCTION
H
IGH field stresses are commonly used to evaluate the gate oxide reliability. The stress-induced trapped charges in the oxide film result in threshold voltage shifts, excess leakage currents and even oxide breakdowns. Previously, the distributions of trapped charges were measured and discussed to analyze trapping mechanisms [1]–[8]. However, in the stress, both positive and negative trapped charges are simultaneously generated [1]–[6]. In previous studies, effects of and were not differentiated and their net effects were measured when stress-induced or curve shifts were analyzed. To fully understand the trappings and the related degradation mechanisms of gate oxide films, it is desirable to monitor and generation, respectively, during the stress.In this paper, a new method is proposed to monitor the distributions of and , respectively. Unlike , is very unstable and easy to be neutralized by a low reverse bias stress [6], [9]. Thus, the respective effects of and on and curve shifts were differentiated by neutralizing . Then their individual trapping quantities and centroids in the oxide film were extracted. Thus their dependence on stressing currents and injected charges under constant current stresses was monitored. Moreover, the relations between the oxide quality and the trapping characteristics were also studied by using the post-poly-annealing (PPA) method which was to degrade the oxide film [10]. In addition, the dynamic change of
Manuscript received November 15, 1996; revised January 17, 1997. The review of this paper was arranged by Editor C.-Y. Lu. This work was supported by the National Science Council of R.O.C. under Research Contract NSC86-2215-E009-025.
The authors are with the Department of Electronics Engineering and In-stitute of Electronics, National Chiao-Tung University, Hsinchu 300, Taiwan, R.O.C.
Publisher Item Identifier S 0018-9383(97)06127-3.
Fig. 1. The gate voltage shifts(1Vg) for samples under two consecutive
0100 mA/cm2 stresses. Between these two consecutive steps of negative
stresses, small opposite+0.1 mA/cm2stresses were applied for different times (s) to neutralize the pregeneratedQpin the oxide film.
trapping distributions in the oxide film under different injection conditions were also studied. It was found that has a reversible characteristics, with its steady-state trapping level determined by the final oxide field [9], [11], and additional stresses only generate extra but does not disturb the original .
II. EXPERIMENT TECHNIQUES
The samples used in this study were POCl -doped gate MOS capacitors, with an area of cm , on a p-type Si wafer. The 80 ˚A gate oxide was grown in diluted dry O (O /N ) at 900 C and annealed in N at the same temperature for 15 min. The poly-Si gate was metallized with Al followed by a 400 C anneal in N for 30 min. For some samples, the gate oxides were degraded by the PPA, before POCl doping, in N at 900, 950, and 1000 C, respectively, for 10 min. HP4145B and Keithley analyzer were used to measure sample electrical characteristics.
III. RESULTS ANDDISCUSSIONS
As mentioned previously, both positive and negative trapped charges are generated during high field stresses [1]–[6], and is very unstable and easy to be neutralized by a low reverse bias stress [6], [9]. An experiment was done by applying two consecutive 100 mA/cm stresses to a group of similar samples, but with some samples applied with small opposite 0.1 mA/cm stresses for different times between these two consecutive stresses. The first and the second stresses were both applied for 10 s. Fig. 1 shows
Fig. 2. The gate voltage shifts (1Vg) under +0.1 mA/cm2 stress for samples prestressed by 00.1, 01, 010, 0100, and 0500 mA/cm2 with the same total injected charges of 01 C/cm2. The neutralization of 0Jg
stress-generatedQpby the+Jg stresses led to the initial increase of their
1Vgcurves. The higher the0Jg, the larger the initial1Vg-increment, i.e., the moreQp neutralized.
(a) (b)
Fig. 3. The distributions ofQpandQn, pregenerated by a0Vgstress, (a) before and (b) after application of the+0.1 mA/cm2 stress.Qp near the SiO2/Si interface had been neutralized and small quantity of newly generated
Qpappeared near the gate/SiO2interface due to the small reverse positive stress butQnstayed undisturbed.
the gate voltage shifts of these samples. The first stress generated which caused the initial decrease of . The amount of generated then saturated [5] and then , which was generated at the same time, kept increasing. This caused to increase henceoff. During the second consecutive stress, the sample which did not receive the reverse 0.1 mA/cm current stress had a constant increase in the curve. However, for the samples which did receive the reverse
0.1 mA/cm current stress, the decreased at the very initial stress, just like that of the first stress. This indicates that the generated by the first negative stress was neutralized by the reverse 0.1 mA/cm current stress and re-generated during the second stress. For these samples, it is also seen that the decrease in saturates after the 0.1 mA/cm stress for 20 s. This suggests that the precreated had been completely neutralized. And in the following studies, a 0.1 mA/cm stress for 30 s was applied to neutralize the negatively created and thus differentiate the respective effects of
and .
Fig. 2 shows gate voltage shifts, , of 0.1 mA/cm stresses for the samples to which negative stresses of different current densities had been preapplied, but with the same total injected charges of 1 C/cm . For the fresh sample, which did not receive any negative stresses, the curve decreased due to the generation of near the gate/SiO interface [3]–[5], [8]
Fig. 4. The shifts of quasi-staticC0V curves after the +Jg neutralizing stresses applied for samples prestressed with different0Jg(mA/cm2) stresses with the same total injected charges of01 C/cm2. The curves labeled with (R) were for samples neutralized.
upon stressing. However, for the samples which had been prestressed with negative currents, all curves increased rapidly initially then saturated. These initial voltage increments were resulted from recombination of the positively injected electrons with the preexisting ’s which were generated by the negative prestressing currents. The more ’s generated by the higher negative current stress [7]–[9], the more the initial rising of the curve. The final saturation of all curves indicated that for all cases, had been completely neutralized by the positive current stress. On the other hand, since the positive neutralizing stress was relatively small, only 0.1 mA/cm stress with 0.003 C/cm of total injected charges, it is reasonable to assume that the distribution of the preexisting
was not disturbed by it.
The above phenomenon can be explained with the aid of Fig. 3, in which Fig. 3(a) shows the generated and after the first negative stress, and Fig. 3(b) shows that near the SiO /Si interface had been neutralized and small quantity of newly generated appeared near the gate/SiO interface due to the small reverse positive stress, but stayed undisturbed.
The above phenomenon is also reflected by and curves of the samples, measured before and after they were stressed and neutralized. Figs. 4 and 5 show the measured quasi-static and negative curves, respectively, of those samples. In these figures, the curves marked with (R) correspond to the samples which had been neutralized by the 0.1 mA/cm stress. In Fig. 4, it can be seen that the curves of the negatively stressed samples shifted left with the 100 mA/cm shifting most. This indicates that was generated by negative stresses, where the amount of increased with increasing the stressing current. After the neutralizing 0.1 mA/cm stress was applied, all the neutralized curves shifted right, indicating that ’s had been neutralized and the remaining originally generated shifted the curves to the right. Comparing these two groups of curves, we find that they are the same in the shape, indicating that no new SiO /Si interface states were generated during the -neutralization process by the 0.1 mA/cm injection. In Fig. 5, all the negatively stressed curves also shifted
Fig. 5. The shifts of negative I0V curves after the +Jg neutralizing stresses applied for samples prestressed with different0Jg(mA/cm2) stresses with the same total injected charges of01 C/cm2. The curves labeled with (R) were for samples neutralized.
left with the 100 mA/cm -stressed curve shifting most. After the -neutralization, all curves shifted right. This exhibits the same results of the curves.
From the shifts of the above and curves, before and after -neutralization, we can derive the quantities and locations of centroids for and , respectively. This can be done in the following way: Comparing the curves of a sample, before and after its -neutralization, we can derive the flat band difference of these two curves. Similarly, we can derive the gate voltage shift of the negative curves of the sample, before and after it was neutralized. During measuring this , the curves were measured at a current level of 10 A/cm to prevent disturbing the unstable in oxide film. The obtained and , mainly due to the existence and nonexistence of
, can be used to be and , respectively, in the following equations to compute the quantity and the location of [12]
(1a) (1b) where , , and are the centroid of measured from the substrate, the oxide dielectric constant and the oxide thickness, respectively.
On the other hand, after the -neutralization, nearly only existed in the oxide film. We can also derive the quantity and the centroid of by computing and from the curves of Figs. 4 and 5. In these figures, the curves marked with are curves corresponding to the case where only is present in the oxide film. And comparing the curves with the curve of the fresh sample, we can derive and , which are used to be and , respectively. In this case, was extracted at a current level of 10 A/cm to minimize the amount of undetectable charges trapped in the tunneling distance of negative measurement. With the obtained and , the quantity and the location of the centroid of can be computed by using the similar formulas of (1).
(a)
(b)
Fig. 6. The calculated quantities (Qp(n)) and centroids (dp(n)) of both (a) positive and (b) negative trapped charges generated by different negative current stresses with01, 05, and 010 C/cm2of total charges injected. Larger
QpandQn are generated for higher current stresses, and both charges are trapped near the anode.
It has to be mentioned that the above method can only be applied to monitor distributions of trapped charges gen-erated by negative gate voltage stresses. This is because the trapped charges are mostly generated near the SiO /Si interface [3]–[5], [8] under the stress. They can be detected by the and the negative curve shifts. However, for the positive stress, the generated charges are mostly near the gate/SiO interface. The curves and negative curves are not sensitive to their existence. And thus the positively created charges can not be fully and cor-rectly detected by our method. This is also the reason that, in Fig. 3(b), although a comparatively small quantity of were unavoidably generated near the gate/SiO interface during the 0.1 mA/cm neutralizing stress, they were neglected in the previous computation for determining and . Moreover, for the positive stress, an opposite 0.1 mA/cm stress is needed to neutralize the precreated . This 0.1 mA/cm stress will also inevitably generate new near the SiO /Si interface. And this can not be neglected as in the negative stress case, since errors will be introduced in the extracted data. The above method was used to extract the quantities and locations of the centroids of generated and in the oxide stressed by different negative current densities. Fig. 6(a) and (b) shows the extracted data for and ,
(a)
(b)
Fig. 7. The calculated (a) quantities(Qp(n)) and (b) centroids (dp(n)) of both positive and negative trapped charges generated by010 mA/cm2stresses with01, 05, and 010 C/cm2of total injected charges. In which, oxides of some samples were degraded by the PPA at 900, 950, or 1000C for 10 min.
respectively, for the total stressing charges of 1, 5, and 10 C/cm . Both generated trapped charges are found to be near the anode [3]–[5], [8], while the higher the stressing current density, the more away the centroid is from the substrate, in addition to the larger generated quantities [7]–[9]. This is easy to be understood since the higher current density needs a higher applied field which gives more energy to injected electrons to generate traps in the oxide. On the other hand, increases rapidly at the initial stress and is larger than for this range of stressing current. While when increasing the injected charges, saturates after the injected charges reaches 5 C [5], but keeps increasing. This suggests that, initially in the oxide, hole traps are much more than electron traps. Upon stressing, these hole traps are easily filled up, which depends on the stressing field, and afterward electron traps are continuously generated by the stressing current.
The above method was also used to investigate the reliability of oxides for the samples subjected to post-poly-annealing (PPA) at 900, 950, and 1000 C, respectively, by monitor-ing the stress generated charges. Fig. 7(a) and (b) shows the quantities of and , and their respective centroids,
Fig. 8. The changes of quasi-staticC0V curves after +10 or 010 mA/cm2 stresses, both with the total injected charges of 1, 5, 10, and 20 C/cm2.
and , of the samples stressed by 10 mA/cm with 1, 5, and 10 C/cm of total injected charges. These figures show that more and are generated for the samples applied by PPA at the higher temperature [10], and is more susceptible to the PPA. This indicates that PPA does degrade the quality of the oxide and the factor to cause oxide breakdown is mainly associated with the generation. Also, for the higher PPA temperature or the larger injected charges, the centroid of is more far away from the substrate, i.e. more near the injection interface. This also indicates that a shorter accelerating distance is sufficient for the injected electrons to generate trapping states in the degraded oxide which probably has more weak spots in it after PPA. But for , the trend is opposite. In which, its centroid, , becomes closer to the substrate for the higher PPA temperature or the larger injected charges.
As mentioned previously, and of stressed oxides can not be derived by our proposed method. It is in-teresting also to investigate these trapped charge distributions in the oxide. Fig. 8 shows the changes of normalized quasi-static curves ( for MOS capacitors after they were stressed by 10 or 10 mA/cm , both with 1, 5, 10, and 20 C/cm of total charges injected. The larger distortions of negatively stressed curves reveal that the interface states generated by stress were more than those by stress, especially for large injected charges. This could be attributed to that, upon stressing, electrons were injected from the gate and thus more damages were created at the oxide/substrate interface to generate interface states [13]. On the other hand, at the initial stage of stresses, i.e., the 1 C and 5 C stresses, the curves shifted to the negative due to a large amount of generated near the oxide/substrate interface. When increasing the injected charges, remained almost constant and was continuously generated. But the effect of was not felt in the curves’ shifts, since , the centroid of , moved closer while moved away from the oxide/substrate interface. However, for the stresses, the more the stress, the more the curves shifted to the positive. This probably indicates that and were generated near the gate/oxide interface with their centroids,
Fig. 9. The gate voltage shifts(1Vg) under +0.1(00.1) mA/cm2stresses, which were used to neutralizeQpin oxide films, for samples prestressed by
010(+10) mA/cm2with01(+1) C/cm2of total injected charges. In which,
oxides of some samples were degraded by the post-poly-annealing at 900, 950, or 1000C for 10 min.
and , shifting to the gate/oxide and the oxide/substrate interfaces, respectively. Moreover, the curves during the neutralization process by applying an opposite small stress can also give some hint on the quantity of positively generated . Fig. 9 shows the gate voltage shifts of 0.1 mA/cm neutralizing stresses (in open dot curves) for the PPA treated samples which were prestressed by 10 mA/cm with 1 C/cm of total injected charges. The initial rapid increase of the curves indicate that stress created was neutralized by the latter stress. And the higher temperature PPA, the more ’s were generated upon stressing. For comparison, similar samples were applied with 10 mA/cm stress but neu-tralized by 0.1 mA/cm , and their curves corresponding to this 0.1 mA/cm neutralizing stress are also plotted in Fig. 9 (in black dot curves). Comparing the initial increments of these curves, we find that the stressed curves have smaller initial increments than those of stressed curves, i.e., generated by stress is less than that of stress. This may also be one of the reasons for that stresses result in smaller breakdown charges than stresses [14].
In the end of this work, the reversible characteristics of trapped charges, with their trapping levels being irrespective to the previous stresses but only determined by the final stressing field [9], [11], is investigated. In this study, samples were first stressed by a of different densities of 500, 100, 10, 1, and 0.1 mA/cm , all with 0.5 C/cm of total injected charges, to generate trapped charges in the oxides. Then they were stressed again by a following second of 0.1 mA/cm stress. During this stress, the gate voltage shifts, , were plotted in Fig. 10. For fresh sample, which was not stressed, the initial decrease of its curve indicates that was generated upon stressing. However, the other samples with prestresses have initial increments on their curves, indicating ’s generated in the first prestressing step were neutralized by the following stress. Also for the samples with higher stresses, the larger initial increments are observed. It is because the magnitude of the
Fig. 10. The gate voltage shifts(1Vg) under J2= 00:1 mA/cm2stresses for samples prestressed byJ1= 00:1, 01, 010, 0100, and 0500 mA/cm2
and all with00.5 C/cm2of total injected charges. The larger theJ1stress, the larger the initial1Vg-increment, i.e., the more excess pregeneratedQp
neutralized.
Fig. 11. The gate voltage shifts(1Vg) under the neutralizing +0:1 mA/cm2
stresses for samples prestressed first byJ1= 010, 0100, and 0500 mA/cm2
and then byJ2= 00:1 mA/cm2(open dot curves), or only prestressed by
J1(black dot curves), all with00.5 C/cm2 of total injected charges. The reductions of the initial1Vg-increments for samples with an additionalJ2 stress imply that someJ1stress-createdQpwere preneutralized.
following stress was comparatively smaller than that of ( 1 mA to 500 mA, except mA) and hence did not generate , instead, neutralized the pregenerated near the oxide/substrate interface. This can be further verified by observing the curves of the samples applied with a 0.1 mA/cm stress to neutralize the remaining ’s after the stress. The curves are shown in Fig. 11 (in open dot curves). For comparison, the curves of 0.1 mA/cm neutralizing stresses for another similar set of samples, which were not applied with the stress to neutralize the stress created ’s, are also shown in Fig. 11 (in black dot curves). The black dot curves have much larger initial increments, especially for the higher stress, indicating more ’s were neutralized by the 0.1 mA/cm stress. On the contrary, the three open dot curves have much smaller initial increments and almost coincide together. This again verifies that a large portion of had already been neutralized by the stress.
Fig. 12. The calculated quantities of both Qp(open dot curves) and Qn
(black dot curves) for samples after they were justJ1stressed (circular and square dot curves for the total injected charges of 0.5 C and 1 C, respectively), and thenJ2= 00:1 mA/cm2neutralized (diamond dot curves for the total injected charges forJ1of 0.5 C followed byJ2of 0.5 C).
The quantities of and of the above samples after they were just stressed and then neutralized, respectively, were also extracted by using our proposed method. Fig. 12 shows the extracted quantities of and for the above samples. It can be seen that just after stresses (the circular dot curves of total injected charge of 0.5 C and the square dot curves of total injected charge of 1 C), the higher stressing current or the larger injected charges resulted in the larger and . However, when the stress was applied (the diamond dot curves of the total injected charges for of 0.5 C followed by of 0.5 C), the pregenerated is found to decrease, especially for higher current densities. And the resulted ’s for all densities are very close in their quantities. This indicates that ’s generated by stresses were neutralized by and their final quantities depend somehow on the level of the final stressing current. That is, is irrespective to the previous stresses but only determined by the final stressing field. On the other hand, keeps increasing for the additional stress. This also indicates that precreated , not like , is very stable and not easy to be disturbed by the following stress.
IV. CONCLUSION
In this paper, a measurement method was proposed to monitor the respective quantity and centroid of charges, and , which were generated by the negatively applied high field stress. The method was based on the -neutralization by a low positive current stress to differentiate the effects of and on and curves. By using this method to extract the quantities and centroids of and , we found that and are generated near the oxide/substrate interface by the negative stress. And higher fields or more injected charges generate larger quantities of and . During stressing, is initially much larger than , but then saturates in its quantity and moves closer to the oxide/substrate interface, while keeps increasing and moves away from the
2
of metal-oxide-semiconductor structures,” Appl. Phys. Lett., vol. 44, p. 202, 1984.
[5] M.-S. Liang, C. Chang, Y. T. Yeow, and C. Hu, “MOSFET degradation due to stressing of thin oxide,” IEEE Trans. Electron Devices, vol. ED-31, p. 1238, 1984.
[6] C. S. Jeng, T. R. Ranganath, C. H. Huang, H. S. Jones, and T. T. L. Chang, “High-field generation of electron traps and charge trapping in ultra-thin SiO2,” in IEDM Tech. Dig., 1981, p. 388.
[7] Y. Nissan-Cohen, J. Shappir, and D. Frohman-Bentchkowsky, “Deter-mination of SiO2 trapped charge distribution by capacitance-voltage analysis of undoped polycrystalline silicon-oxide-silicon capacitors,”
Appl. Phys. Lett., vol. 44, p. 417, 1984.
[8] , “High-field and current-induced positive charge in thermal SiO2 layers,” J. Appl. Phys., vol. 57, p. 2830, 1985.
[9] Y. Nissan-Cohen, J. Shappir, and D. Frohman-Bentchkowsky, “High field current induced-positive charge transients in SiO2,” J. Appl. Phys., vol. 54, p. 5793, 1983.
[10] R. Mehta, A. B. Bhattacharyya, and D. N. Singh, “Post-growth process-induced degradation in thin gate oxides,” J. Appl. Phys., vol. 69, no. 12, p. 8247, 1991.
[11] Y. Nissan-Cohen, J. Shappir, and D. Frohman-Bentchkowsky, “Dynamic model of trapping-detrapping in SiO2,” J. Appl. Phys., vol. 58, p. 2252, 1985.
[12] D. J. DiMaria, “Determination of insulator bulk trapped charge den-sities and centroids from photocurrent-voltage characteristics of MOS structures,” J. Appl. Phys., vol. 47, p. 4073, 1976.
[13] P. P. Apte and K. C. Saraswat, “SiO2degradation with charge injection polarity,” IEEE Electron Device Lett., vol. 14, p. 512, 1993.
[14] Y. Hokari, “Stress voltage polarity dependence of thermally grown thin gate oxide wearout,” IEEE Trans. Electron Devices, vol. 35, p. 1299, 1988.
Yung Hao Lin was born in Taichung, Taiwan,
R.O.C., in 1969. He received the B.S. and Ph.D. degrees from the Department of Electronics Engi-neering, National Chiao-Tung University, Hsinchu, Taiwan, in 1991 and 1996, respectively. He is currently continuing postdoctoral research at Na-tional Chiao-Tung University. His research interests include the fabrication technology and the reliability analysis of thin gate dielectrics.
Chung Len Lee (S’70–M’75–SM’92), for a photograph and biography, see
p. 159 of the January 1997 issue of this TRANSACTIONS.
Tan Fu Lei, for a photograph and biography, see p. 159 of the January 1997
