Optimizing I-line Lithography for 0.30-UM poly-gate manufacturing

03/01/1997

Optimizing i-line lithography for 0.3-?m poly-gate manufacturing

Jo Finders, Plamen Tzviatkov, Kurt Ronse, Luc Van den hove Interuniversity Microelectronics Center (IMEC), Leuven, Belgium

With advanced steppers and resists, i-line lithography (365 nm) is capable of printing sub-0.35-?m geometries. Individual and overlapping process windows for grouped and isolated 0.3- and sub-0.3-?m lines were optimized, using advanced resists and bottom antireflected coating (BARC). The entire poly-gate patterning sequence, including process characterization by scanning electron microscope (SEM) inspection after resist development, BARC-dry etch, and polysilicon etch, has been investigated. Automated electrical linewidth measurements (ELMs) were proven to be a fast and reliable method to measure the after-etch process window.

Recent work has indicated the feasibility of using i-line lithography for printing 0.3-?m design-rule circuits [1, 2]. Besides advanced i-line resist materials, this resolution improvement required advanced i-line steppers with variable NA, partial coherence (s), and optical enhancement technology. In this paper, we will demonstrate an optimization methodology that enhances the process windows of 0.3- and sub-0.3-?m geometries. Optimized individual process latitudes and adequate common overlap of the process windows is necessary for a process to be qualified for manufacturing.

For the initial lithographic pattern definition, several parameters such as resist choice and exposure tool settings can be varied. To understand the impact of tool NA and s settings on the CD control, in-line SEM measurements were done after development of three different resists. In this way, the initial process windows and the optical proximity effect (OPE) can be characterized for fine-tuning the lithography process. We found that the OPE shows a strong nonlinear - and sometimes oscillating - behavior as a function of pitch. To vary the pitch for a given CD, the space between the lines was changed with the linewidth held constant. Linewidth/spacewidth (L:S) ratios in the range from 1:1 (i.e., dense lines) to 1:100 (fully isolated lines [IL]) were considered.

It is well known that photo process latitudes may change during the next steps after development, i.e., the processing of the BARC and poly. Because of different etch biases for lines of different pitches, the overlap of the process windows may be altered as well. Therefore SEM measurements were also done after BARC- and poly-etch to monitor these effects. We used the dry-etch settings that were optimized for IMEC`s standard 0.35-?m poly-gate patterning process. The ultimate process windows after poly-etch also have to be analyzed in detail. Owing to the large number of input variables (NA, s, resist choice) and the large number of figures-of-merit (e.g., latitudes, overlap, linearity), ELMs were used as a fast and reliable method to define the process windows after poly-etch.

Experimental conditions

Three commercially available positive-tone i-line resists from different vendors were used: OiR 643 from Olin Microelectronic Materials (OMM), FHi 620 BC from Fuji-Hunt, and PFi 60A from Sumitomo. The resists were coated on top of 80-nm BARC (BARLi-0.065-?m grade) using a TEL Mark-8 wafer track. The resist thickness was held at 0.76 ?m. All resists were soft baked at 90?C for 60 sec. Postexposure bake was performed at 110?C for 60 sec. Development was carried out using OMM OPD 262 developer.

The exposure tool was an ASM-L 5500/200 i-line stepper with variable NA (048-0.6) and s (0.34-0.85). The advanced AERIAL illuminator of this production tool allows keyboard-controlled implementation of annular illumination without loss of throughput [3]. A binary mask with dedicated structures for ELM had feature sizes ranging between 0.5-0.25 ?m (at 1?) with an increment of 0.025 ?m. The L:S ratio varied from 1:1 (denoted as L:S or 1:1) to 1:100 IL. In-line SEM measurements were done on a Hitachi 8820 SEM. Cross-sections of resist profiles were obtained using a Jeol JSMT 330 high voltage SEM. The ELMs were performed on a HP 4062 parameter analyzer. A LAM-TCP 9400 dry-etched the BARLi using O2 and N2, and a Tegal 1511 performed the poly-etch. Recently, we switched to a more advanced poly-etch process using a LAM-TCP 9400 SE system

Processing results: after resist development

We will discuss the trends at different stages during the poly-gate patterning sequence starting with after resist development.

OPE and resolution for fully isolated and dense lines. Immediately after resist development, we parametrized the achieved DOF, the exposure latitude (EL), and the OPE (parametrized as the dose-to-size) between fully isolated and dense lines using a SEM. The DOF for different coherence settings at a NA of 0.6 for fully ILs and dense lines (L:S = 1:1) at the best exposure dose is shown in the table for each condition.

For conventional illumination and a low coherence parameter s, the DOF is limited for both dense and ILs to less than 1 ?m. Increasing s to 0.8 increases the DOF for the dense lines, without altering the DOF for ILs. This trend is even more pronounced for annular illumination (with the obscuration parameter si in the table); the DOF for the dense lines is increased to about 1.6 ?m, but the DOF for the fully ILs remains <1 ?m. A limited DOF of =1 ?m for ILs was found for all resists under investigation, for different NAs, and regardless of the illumination settings. For a poly-gate process, this will be one of the major limitations.

For the dense patterns (L:S = 1:1), annular settings are obviously favorable. The ELs for various illumination settings summarized in the table do not vary greatly. Annular illumination can thus be used to increase the DOF for the dense lines without degrading the EL. However, altering the illumination settings clearly changes the proximity effects: The exposure dose-to-size of 0.3-?m ILs changes dramatically as indicated in the table. Illumination settings with high s values result in severe CD differences between dense and fully ILs (isodense bias). A similar trend has been found for NA = 0.5. Because the observed trends are similar for both settings of NA, we focused for the rest of our study on NA = 0.6. This high NA value is necessary to achieve the required =0.3-?m resolution.

The observed isodense bias for 0.3-?m lines has been plotted in Fig.1 as a function of the inner and outer illumination limits, si and so. For large so, the proximity effect becomes severe (up to 60 nm). However, there exists a region of si, so settings, including both annular illumination (si>0) and conventional (si = 0) settings, where the OPE vanishes. Thus, the illumination settings can be fine-tuned in order to achieve zero isodense bias for 0.3-?m lines.

With respect to overlapping EL and overlapping DOF between fully isolated and dense (1:1) 0.3-?m lines, we found an optimized illumination setting: si = 0.35, so = 0.70. The ED windows for both isolated and dense 0.3-?m lines at this condition are plotted in Fig. 2 for FHi 620 BC resist. The DOF for the dense lines is enhanced beyond 1 ?m because of the annular settings. For the ILs, the stronger bowing of the ED window indicates a smaller DOF. However, because of the OPE, the dose-to-size is somewhat larger for the ILs than for the dense L:S pattern. The overlapping DOF (ODOF) is 1.1 ?m, which is larger than the DOF for ILs measured at best energy. Because of the large individual ELs for this resist (>25%), we achieved an overlapping EL of 21%.

At the optimum illumination settings, we made cross-sections of the resist profiles to check them at best resolution. An example of this is shown in Fig. 3 for FHi 620 BC; 0.3- and 0.275-?m lines are resolved with a good resist profile, while 0.25-?m lines are resolved for all pitches, but show a resist loss of 20% for the fully ILs. On BARLi, some of the advanced i-line resists show a small "foot." Because it is removed during the BARLi-etch, we did not include it in CD measurements made after resist development, even though it can be seen in top-down SEMs. In this way, its removal during BARLi-etch is omitted from the dry-etch CD bias.

OPE vs. pitch. The isodense bias is not sufficient to characterize OPE. Linewidth shifts for intermediate pitches must also be examined. However, to limit the number of SEM measurements, the intermediate pitches were studied only at best focus. ELMs were used after poly-etch to look for overlapping windows. Figure 4 shows the measured linewidth vs. pitch for 0.35-, 0.30-, and 0.25-?m lines for two illumination settings: annular (si = 0.35, so = 0.70) and conventional (si = 0, so = 0.70). It can be seen from Fig. 1 that for both conditions the isodense bias of 0.3-?m lines becomes very small.

However, for intermediate pitches, the behavior for 0.3-?m lines is quite different. The largest difference is observed in the range between L:S = 1:1 and 1:2. In that range, we observe a strong oscillating behavior of CD vs. pitch in the case of annular illumination. These oscillations were observed for different resists and occurred also for other annular settings (e.g., si = 0.5, so = 0.80). For 0.35- and 0.25-?m lines, these oscillations were not observed. In the case of features larger than 0.3 ?m, the swings in the CD vs. pitch curve are <10% of the CD. However, if one chooses illumination settings that show a larger isodense bias, the nonlinear behavior - especially in the range between 1:1 and 1:2 - increases. For sub-0.3-?m features, a (large) reduction in the printed linewidth for 1:1 =L:S=1:2 occurred for all illumination settings.

Figure 1. Experimentally observed isodense bias of 0.3-?m lines after resist development for different settings of si, so. NA = 0.6. Resist: OiR 643, FHi 620 BC.

Figure 2. ED windows for fully isolated and dense (1:1) 0.3-?m lines after resist development for optimized coherence settings: so = 0.7, si = 0.35.

NA = 0.6. Resist: FHi 620 BC.

Figure 3. Resist profiles for dense and isolated lines with nominal CDs of a) 0.3 ?m, b) 0.275 ?m, and c) 0.25 ?m. Coherence settings: so = 0.7, si = 0.35. NA = 0.6. Resist: FHi 620 BC.

Figure 4. CD vs. L:S-width ratio for 0.35-, 0.30-, and 0.25-?m lines. Illumination settings: so = 0.7, si = 0 (solid lines), so = 0.7, si = 0.35 (dashed lines). Resist: Sumitomo PFi 60A.

Results: after BARLi- and poly-etch

After BARLi- and poly-etch, in-line SEM measurements were performed to detect shifts in the process windows (ELs, DOF, and their overlap), which would indicate an etch bias. Figure 5 shows the ED windows for dense (L:S = 1:1) and isolated 300-nm lines after resist development, after BARLi-etch, and after poly-etch with the Tegal 1511. We see no significant change of the individual process latitudes during the etch sequence. However, the poly-etch step causes a strong etch bias for the ILs. As a result, the overlap of the process windows of dense (1:1) and fully isolated 0.3-?m lines, observed after resist development, has disappeared after the poly-etch. Thus, the optimized illumination settings that were found to minimize OPE after resist development must be revised to account for etch bias. A possible method to achieve better overlap after poly-etch is to change the illumination settings in order to have an OPE of about -40 nm after resist development, which is then compensated by the etch biases.

Etch bias vs. pitch. The etch biases depend in a systematic way on the pitch of the structures. In Fig. 6, the CD vs. pitch for 300-nm lines after resist development, after BARLi-etch, and after poly-etch with the Tegal 1511 are plotted for FHi 620 BC resist. After resist development, we again observe the oscillations in the CD vs. pitch curve, especially in the range of pitches between 1:1 and 1:2. Note the similar behavior for 0.3-?m lines for the Sumitomo PFi 60A (Fig. 4) and the FHi 620 BC resist (Fig. 6). The observed OPE is mainly influenced by optical parameters and not the performance of advanced i-line resists. A Fourier optics analysis of the imaging of L:S patterns suggests that the contrast of the aerial image will vary most dramatically for L:S ratios between 1:1 and 1:2. Such contrast oscillations will affect not only the CD value at the bottom of the resist, but also the wall slope and resist loss.

The etch biases for the BARLi- and poly-etch can be extracted from the CD differences measured after the two dry-etch processes. The BARLi-etch bias is <10 nm; the poly-etch bias, is much larger. It ranges from -20 nm for the dense lines (linewidth loss) to +30 nm for the ILs (linewidth increase). Switching to a more advanced poly-etch process using a LAM-TCP 9400 SE reduced the range of etch biases by half. Nevertheless, the oscillating behavior in the range of pitches between 1:1 and 1:2 appears again. Because the etch behavior during poly-etch depends on the slope of the resist, the reappearance of this optically induced oscillation implies that a resist slope variation in addition to a CD variation appears for the different pitches.

To confirm this mechanism for etch bias, we took cross-section SEMs of the resist profiles of 0.3-?m lines for different pitches. The extracted sidewall angle is plotted in Fig. 7. The dense (1:1) structures show a very straight profile, whereas the more isolated features show a bit more slope. Again, strong slope variations appear for 1:1 =L:S=1:2.

Figure 5. ED windows for 300-nm fully ILs and dense L:S patterns a) after resist development, b) after BARLi-etch, and c) after poly-etch with the Tegal 1511, as derived from SEM measurements. Illumination settings: so = 0.7, si = 0.35. NA = 0.6. Resist: FHi 620 BC.

After poly-etch: The final process windows using ELMs

The results of the in-line SEM measurements done at three stages in the process (after resist development, after BARLi, and after poly-etch) made it clear that an investigation of the final process windows is of importance. A fast and reliable method is needed to characterize the common process windows using such figures-of-merit as the variation of CD vs. pitch, the ED windows, and linearity. Automated ELMs offer a convenient solution.

A binary reticle for ELM was fabricated in-house using MoSi dry-etchable absorber and e-beam proximity corrections. The test structures were positioned to minimize the effects of long-range etch nonuniformities. It is well known that raw ELM linewidths are always offset from those measured by a SEM (see Fig. 6). The nature of this offset is not yet completely understood. In order for ELM to be a valid alternative to SEM metrology in characterizing the process window, one must cross-calibrate the results of the different measurement techniques. In the case of sloped poly-profiles, a SEM measures the linewidth at the foot of the profile, which differs from the average linewidth across the profile determined by ELM. For all exposure and focus settings, the resulting offset between the two measurement modes turned out to be constant experimentally. An average offset of 71 nm with a 3s = 12 nm was found across the focus-exposure matrix. Taking the 3s value of 12-15 nm into account as a typical number for SEM repeatability, the offset may be considered to be constant. Thus, only a limited number of SEM measurements is required to determine the offset accurately. Once it is determined, the ELM data can be converted into SEM data.

Using ELM, all ED windows for eight different CDs ranging from 0.5-0.225 ?m at eight different pitches were measured. Until the final poly-etch step, the OPE vs. pitch was mainly monitored at best focus. With the help of ELM, we were able to have a detailed look at the OPE vs. pitch behavior through focus. The ED windows of 0.3-?m lines for different pitches were measured after poly-etch (for the illumination settings that gave best results after resist development - so = 0.7, si = 0.35). For fully isolated and dense lines, the ED windows were quite symmetrical around the best focal position (F = +0.3 ?m), whereas for intermediate pitches, the windows were deformed. This stresses the importance of measuring the complete window by ELM to judge the OPE for different pitches, not only at best focus, but also under defocused conditions.

Figure 8 shows the ED windows of 250-nm fully isolated and dense (1:1) lines after poly-etch with the Tegal 1511. The isodense bias for 250-nm features after resist development is almost perfectly compensated by the etch bias. Thus, it is possible to obtain ED windows for sub-0.3-?m features, which show a reasonable overlap. This final overlap exists only for some pitches and cannot be maintained throughout the complete range, as illustrated in Fig. 9. In this figure, the dose required to obtain a linewidth variation of ?10% after poly-etch using the LAM-TCP 9400 SE is plotted vs. the pitch for three different mask features: CD = 0.3, 0.275, and 0.25 ?m. If, for a given CD, a dose value can be found that for all pitches would result in a linewidth between CD-10% and CD+10%, there exists an overall common process window. For 0.3-?m features, the overlapping exposure latitude (OEL) is about 8% for all pitches. For smaller structures (CD = 0.275 ?m, CD = 0.25 ?m), this energy overlap becomes worse, falling to 4% OELfor 0.275 ?m and vanishing for 0.25 ?m. Optical proximity correction (OPC) is a possible method for improving the overlap. However, the OPC corrections must take etch bias into account. For both 0.3-?m and sub-0.3-?m features, the corrections need only to be applied for L:S ratios between 1:1 and 1:3.

Figure 6. a) CD vs. L:S ratio, and b) etch bias vs. L:S ratio for 300-nm lines after resist development, BARLi-etch, and poly-etch with the Tegal 1511. Illumination settings: so = 0.7, si = 0.35. NA = 0.6. Resist: FHi 620 BC.

Figure 7. Resist sidewall angle vs. L:S-width ratio. Illumination settings: so = 0.7, si = 0.35. NA = 0.6. Resist: FHi 620 BC.

Figure 8. ED windows for 250-nm fully isolated (dashed lines) and dense L:S patterns (solid lines) after poly-etch with the Tegal 1511. s0=0.7, si=0.35. NA=0.6. Resist: FHi 620 BC.

Figure 9. Energy required to obtain a linewidth of CD+10%, CD, and CD-10% after poly-etch with the LAM-TCP 9400 SE for various L:S ratios for a) CD = 0.3 ?m, b) CD = 0.275 ?m, and c) CD = 0.25 ?m. s0=0.7, si=0.35. NA=0.6. Resist: PFi 60A.

Conclusion

This optimization study for 0.3-?m structures found that illumination settings involving high partial coherence (s>0.7) result in severe proximity effects (up to 60 nm) between fully isolated and dense lines (L:S = 1:1) after resist development. Furthermore, this effect shows a strong nonlinear behavior for intermediate pitches, especially in the range of pitches between 1:1 and 1:2. Tuning the partial coherence (outer sigma, so) and the obscuration (i.e., si) revealed a region of (so, si)-settings, where the OPE between fully isolated and dense lines becomes very small and the dynamic behavior for the intermediate pitches is reduced. This region includes both annular (si>0) and conventional settings (si = 0). However, using an annular condition (so = 0.7, si = 0.35), we benefited from an enlarged DOF for the dense lines, while the ELs were not substantially affected. In this way, an overlapping DOF of 1.1 ?m and an overlapping EL between dense and fully isolated structures of 21% was achieved. The OPE after resist development showed similar results for three advanced i-line resists, which clearly demonstrates that this effect is predominantly defined by the optics of the stepper. Automated ELM was found to be a convenient and powerful method for exploring after-etch CD variations throughout the process window, and facilitating the full optimization of a poly-gate process.

The ELs and the DOFs are not significantly changed during the BARC- and the poly-etch. However, especially in the poly-etch, we observed a systematic pitch-dependence of the etch bias of the 0.3-?m lines. For dense lines, the linewidth of the structures decreased during etch; for fully ILs, it increased. Therefore, the individual process windows shift with respect to the exposure energy. The optimum overlap shown after resist development decreases. There are two possible methods of overcoming this etch-bias problem: correcting the mask using OPC with rules that incorporate etch bias, or tuning the optical settings of the stepper to minimize pitch-dependent CD shifts after etch - that is, purposely allowing an OPE in lithography that is compensated by the pitch-dependent etch bias. Our results indicate that the latter method alone can result in a manufacturable 0.3-?m process, but that OPC will be needed for sub-0.3-?m features. Either method will only result in a specific solution for an individual process, because the etch biases depend on the specific process conditions. Since illumination can be varied from the stepper keyboard, and mask redesign is a months-long process, it is more convenient to account for engineering changes in etch process conditions by tuning the stepper.

The results shown in this paper indicate that it is feasible to develop a manufacturable 0.3-?m poly-gate process using i-line lithography. Beyond that, we obtained process windows after poly-etch for 0.25- and 0.275-?m structures with remarkable individual ELs (more than 15%) and DOF (0.7 ?m for ILs). The lines were resolved for all pitches and there was a clear overlap of the process windows for some pitches. Therefore, we believe that, especially when OPC can be used, the resolution limit for advanced i-line lithography is actually lower than 0.3 ?m, and that in the near future there will be some applications at 0.275- and 0.25-?m dimensions.

Acknowledgment

AERIAL is a trade mark of ASM Lithography.

References

1. P.Tzviatkov et al., "0.30-?m and Sub-0.30-?m i-Line Lithography for Random Logic Gates," Proceedings Interface, pp. 1-5, 1995.

2. K-Y. Kim, et al., "Implementation of i-Line Lithography with 0.30-?m Design Rules," Microlithography World, pp. 6-11, Autumn 1995.

3. R. Rogoff et al., "Photolithography Using the AERIAL Illuminator in a Variable-NA Wafer Stepper," Proceedings SPIE, Vol. 2726, pp. 54-71, 1996.

Contact Jo Finders at IMEC, Kapeldreef 75, B-3001 Leuven, Belgium; ph 32/16-281-344, fax 32/16-281-214.

Geneva photo on Europe cover courtesy of Swiss National Tourist Board