Photostabilization: Comparing DUV and i-line
02/01/2003
overviewThe direct transfer of i-line resist photostabilization processes to deep UV resists is a very common practice, but is a mistake and should not be done. Through a basic understanding of the resist chemistry, good research through design of experiments, and sound analysis, optimum processes can be realized for thin and thick films — for both deep UV and i-line processes.
Photostabilization, also known as deep-UV (DUV) hardening, has been in the mainstream of semiconductor manufacturing for more than 20 years, primarily for i-line resists, where it is still effective at improving process performances at ion implant levels. It is also effective with many etching processes [1].
 |
Since the mid-1990s, more and more manufacturers have been turning to DUV photoresists to enable resolution of smaller fine-line features. With the aid of resolution enhancements and process tricks, resolutions at the 248nm wavelength have been driven to sub-180nm. At the same time, attempts have been made to transfer photostabilization processes to these DUV resists.
In particular, the transfer of "copy exact" recipes from i-line resists to 248nm DUV resists have met with unexpected results. Many who attempted to copy exact subsequently discarded the process across the board as ineffective. Others took the time to properly characterize the new DUV resist process window for given layers, and achieved rewarding results. The main reason for this disparity is that the resist response to photostabilization is very different for the chemically amplified resists (CAR) used in DUV lithography, as compared to i-line resists. Therefore, understanding the response differences is critical to achieving success.
 Figure 1. 3.0µm i-line images cured with oxygen and a D-Mod bulb: a) before and b) after. |
null
Basic photostabilization
Upon intense DUV irradiation of resist in a photostabilization process, all of the resist's photochemical reactions are initiated. This is much like normal photolithography. By continuing to raise the temperature of the resist during the exposure, extensive cross-linking of the resist is achieved. This is due to the breakdown of the chemistry and the destruction of the photoactive compound (PAC) in i-line resists. In DUV resists, however, the heat causes the diffusion of the photo acid generator (PAG) and the subsequent cleaving and expulsion of the leaving groups.
A comparison of the chemical reactions between the two types of resists shows the difference in the by-products of each chemical reaction. The main by-product of i-line resist chemistry is nitrogen, while DUV resists eliminate hydrogen and the leaving group. This is a very important factor to keep in mind, since this expulsion of the leaving group and solvents is a major mechanism in the shrinkage seen with all DUV/CARs.
 Figure 2. Control features at a) 0.25µm and features reduced to b) 0.1875µm by a controlled shrink process. |
null
Differences between i-line and DUV resists
i-line resists are a three-component system consisting of a PAC, a resin, and various solvents. The molecular weight of the resin (novolac) is typically around 1000. The PAC (diazo napthoquinone — DNQ) is most commonly stitched to the polymer chain and is activated upon light exposure. The chemical reaction is immediate and the image is formed upon exposure. DUV resists are also a three-component system with variations in the polymer contents. Each has a PAG (initially based on PFOS — perfluoro-octane sulfonate), one or more polymers, and multiple solvents. The molecular weights of the polymers vary from 25,000 to 50,000+. The PAG resides within the polymer matrix and is activated by exposure to DUV photon, ion, electron, or x-ray.
When either the i-line or DUV resists are subjected to a postdevelop photostabilization process, there are varying results depending on the resist type. The main difference is the molecular weight of the resist. With DUV resists, the polymer chains are very long and have an enormous amount of free volume when coated onto the wafer surface. This is by design to allow the PAG to diffuse through the matrix and form the image. i-line resists, with small polymer chains, compact very tightly together during coating and form an image almost immediately upon exposure.
In i-line photostabilization, the resist is UV-irradiated while the temperature is simultaneously elevated. The intense DUV radiation begins to cross-link the resist, raising the glass transition temperature (Tg), thereby allowing the temperature to be raised as high as 250°C with no thermal flow. Shrinkage of most i-line resists has been measured to be on the order of approximately 5% (Table 1).
With DUV resists, however, elevating the temperature activates the acid and begins the diffusion process. At the same time solvents are released, leaving groups are cleaved and billions of tiny microvoids are created. This free volume loss and creation of the microvoids allows the polymer chains to contract upon themselves while at the same time beginning to cross-link. With increasing temperature, free volume loss is increased. The result is resist shrinkage as high as 25%. This shrinkage causes a change in the printed features and the critical dimensions (CDs). Grouped or isolated lines get smaller while contact holes and vias grow larger.
Typical i-line processes for ion implantation take the resist temperature up to 230°C with a minimal amount of shrinkage. If one were to "copy exact" and attempt to use this same temperature for DUV resists, the shrinkage would be nearly 25%, resulting in serious changes in CDs and subsequently, device performance. The same applies to etch processes where such a high temperature is not necessary, but when copied exact, the resulting loss of CD control for DUV resists becomes unacceptable.
 Figure 3. a) Before and b) after pictures of improved line-edge roughness through a chemically assisted photostabilization process (Shipley UV6, 248nm ESCAP). |
null
The etch application space
The propensity for significant shrinkage of DUV resist during photostabilization was the source of some confusion during early investigations into the ability of the process to enhance etch selectivity. Observations initially showed no significant enhancement of selectivity because the photostabilized resist thickness after etching appeared to be the same as that of unprocessed resist after etching. Closer examination revealed that etch rates had actually improved (decreased) but were offset by the resist shrinkage.
Kishimura et al. conducted a rather thorough investigation of the effects of photostabilization under a variety of conditions on both DUV shrinkage and etch rate [2, 3]. A 43% improvement in etch rate with a concomitant 17% shrink could be achieved by photostabilizing an ESCAP-type resist under nitrogen. Although this corresponds to a net improvement in etch selectivity, it is not clear whether manufacturers will be able to tolerate feature shrinkage on this scale [4].
 |
As reported by Krisa, improvements on the order of nearly 40% were realized at metal etch through proper characterization of the stabilization process [5]. In this case, the best final temperature for optimum selectivity was realized at 120°C. However, for i-line resists, the temperature for this same type of process will vary between 150 and 180°C. As such, there is a large difference in what one would see in feature shrinkage and etch selectivity if the process were transferred as a "copy exact." The effectiveness lies in properly characterizing the operating space through design of experiments, and then analyzing the outcome to get the optimum desired process conditions.
It is also important to note that with elevated temperatures, there will be a decrease in the thickness and a change in the as-printed CD. While improvements may be seen in the etch selectivity, it may well come at the cost of CD changes. In cases where improvements in line-edge roughness (LER) are needed, allowances for changes in CDs may be made as a trade-off to improving the overall quality of the pattern transfer during the etch.
The implant application space
For ion implantation into DUV resists, the benefits are still seen, as in the past, with i-line resist [6]. The improvements in outgassing, particles, and improved strippability (Table 2) still exist [7].
The operating parameters in the photostabilization recipes, however, are markedly different than for a typical i-line recipe. Commonly, final temperatures for i-line resists vary between 190 and 240°C, while for DUV resists a more typical final temperature has been observed to be between 140 and 150°C.
The shrinkage and the consideration that must be given to maintaining features within design rule limits are important. Too much shrinkage can seriously impact the location of implant species, and subsequently, the performance of the devices. The same can be said for the maintenance of sidewalls, since changes in the angle will impact the ion penetration into the resist at all feature edges.
Photostabilizing with oxygen
To date, the gas used to purge the photostabilization chamber has been nitrogen, though in other cases clean dry air has been used. In some instances, the purge has been eliminated for a variety of reasons, one of which is process improvement. Without the nitrogen purge, it was noted that the process was slightly enhanced, either through sidewall improvement or by reduced process times. Increased frequency of quartz plate cleaning was required since the volatile outgassing materials from the resist would rise and adhere to the quartz plate. (Failure to clean the quartz plate would cause a serious drop-off in the UV irradiation.) The processes could have been optimized even with the nitrogen purge; it was deemed unnecessary to do so, however.
Until recently, there has been little reason to dispute or embrace the use of anything but nitrogen as the purge gas. As resist films began to get thicker for various processes, the stabilization process became more difficult and problematic. Increased reticulation, changes in sidewall profiles, and loss of film integrity became more prevalent. Investigation and root cause analysis led to the use of oxygen in place of nitrogen as the purge gas.
Photostabilizing i-line resists with oxygen
The use of nitrogen does not promote cross-linking of novolac-based resist systems. The PAC to resin cross-linking will occur with or without oxygen. However, the resin-to-resin cross-linking requires oxygen as explained by Moreau et.al., who state the need for oxygen to enhance the hardening (cross-linking) of these types of resist [8, 9].
During the stabilization process, UV energy begins to break down the PAC in the resist as soon as it is enabled in the process. This breakdown, coupled with the thermal energy, begins the process of cross-linking the PAC with the resin. One of the by-products of the breakdown of the PAC in the resist is nitrogen, which is released as soon as the conversion of the ketene chains to caboxcylic acid begins. In a nitrogen-rich environment, this cross-linking is retarded (catalyzation cannot take place), since this nitrogen "blanket" does not allow sufficient amounts of oxygen to penetrate the resist.
The amount of this inhibition is a function of many factors: resist thickness; nitrogen flow; the amount of ambient "air" present; and the bake temperatures employed at post-apply bake and post-exposure bake.
Resist thickness is the primary factor, however. The penetration of oxygen into the entire resist film is key to cross-linking.
For thin films, the PAC-to-resin cross-linking may be adequate to enable a sufficiently stabilized film. One could surmise that there appears to be little or no inhibition, but quite the contrary is true. The reaction is the same, but the effects are less obvious because the cross-linking of the entire thin film takes place at a much faster rate. Slight improvements have been seen with thinner films without nitrogen. Those IC manufacturers who do not use the nitrogen purge can now easily explain why the purge is not used, yet it does not preclude the constant window cleaning required by not using any purge.
For thicker films (i.e., >1.5µm) such as seen in Fig. 1, oxygen (most commonly coupled with the Axcelis Technologies D-Mod bulb) provides a process enhancement resulting in improved film integrity. The reticulation is eliminated, the sidewall profiles are easier to maintain, and the resist holds up better to high-dose and high-energy implants, as well as etch processes such as silicon trench etch.
Table 3 shows the characterization to find the optimum final temperature for the features seen in Fig. 1. In this case, the optimum temperature for achieving the least amount of sidewall change is 190°C. The main reason is the oxygen has enhanced the cross-linking and thus, the film is better able to undergo incremental changes during the stabilization process without damage to itself or to a feature. Similar performance can be achieved in nitrogen ambient; however, the time to obtain a good cure of the film is prohibitively longer and requires considerably more characterization.
The level of success using oxygen as the purge gas on thick resist films is evident and highly recommended. There is also no reason not to use oxygen for thinner films, since its primary function is to enhance cross-linking. In addition, a third party has tested the effluent while stabilizing with a pure oxygen purge. The results showed no ozone or other damaging by-products [10].
Photostabilizing DUV/CARs with oxygen
While the use of oxygen is highly recommended for a novolac-based resist system, DUV resists, which commonly contain poly hydroxy styrenes, are a different matter. While oxygen does in fact improve the cross-linking as with novolac resists, shrinkage becomes worse. Since the cross-linking is enhanced, the rate of shrinkage is also increased.
Therefore, the resists will shrink much faster at lower temperatures. For example, if the resist shrinks by 10% at 140°C with nitrogen, it commonly shrinks by nearly 15% with oxygen. This is not to say that oxygen cannot be used with DUV resists, but one needs to keep in mind that increasing the shrinkage also has an effect on the final thickness and the CDs of the features being treated.
DUV resists: Controllable shrink
If the amount of photostabilization-induced shrinkage is characterized, predictable, and understood, the end user could take full advantage of it by designing it into the process at the outset. Reticles could be made at a given feature size and the process set to give a smaller feature size post-photostabilization [11]. A process was characterized to shrink features from one chip generation's minimum feature size of 250nm to the next at 187nm (Fig. 2).
A thorough knowledge of the resist's chemistry is mandatory to obtain a repeatable process. As the features shrink three-dimensionally, the impact on overlay registration and fit to the specified design rules in both the horizontal and vertical (x and y) can be affected. Additionally, the thickness (z) will shrink by the same proportion and must be considered for calculating etch selectivity.
 |
While the resist shrinks, it also becomes increasingly more cross-linked, enhancing the etch resistance. In determining the actual etch selectivity, one needs to take into account both effects. Regardless, the potential to allow for an extension of a lithography tool set, given the fact that features can be shrunk to a smaller dimension, has been shown. Again, this would have to be considered in the initial mask
eticle design to be able to shrink to a final dimension using this controlled shrink process with photostabilization.
Recently, researchers at IMEC tested the feasibility and applicability of this controlled shrinkage process of DUV resist [12]. Features were shrunk from 100nm to 50nm lines with good repeatability. Further, it was reported that linewidths of 75nm were shrunk to 35nm with some limitations. Nonetheless, 84% of all data points across the wafers fell within the requisite specification.
The data demonstrate a process worthy of investigation. The use of this process could lead to the extended lifetime of current tool sets, which would otherwise require replacement due to resolution limitations. The research also investigated the extendibility of this process into 193nm resists. Initial results proved likewise promising and more tests are planned for the near future.
Line-edge roughness
Preliminary data suggest that the photostabilization process is effective in helping to reduce LER. Resist sidewalls have become much smoother after stabilization through the use of a chemically assisted photostabilization process (Fig. 3). Research will continue with 193nm and other resists as they become available.
Conclusion
There is a considerable difference in how novolac/DNQ resists and CARs should be processed through stabilization. While the resist chemistries are dissimilar and respond differently to the photostabilization process, experimentation and application in manufacturing flows has demonstrated that the benefits of photostabilization can be realized across all resist types.
Device engineers should determine the limitations of the shrinkage — a major issue with DUV resists — to know exactly how much is acceptable. By working in concert with device engineers, controllably shrinking printed features to a size of nearly a third of what was printed by using the controlled shrink process can be advantageous. This process, which can be manufacturable if fully understood and characterized, can have other benefits.
Photostabilization has consistently shown its value as part of the process flow. Understanding the interaction of the UV light, thermal energy, and the process timing is very important for creating a robust, manufacturable process — one that allows for variations in both light and thermal energies that would not normally be tolerated. Knowing the key response or responses required from the process is also essential.
Acknowledgments
Many thanks to the Process Technology and R&D Groups at Axcelis, in particular, Donna Whiteside, Tony Sinnott, John Hallock, Dwight Roh, Ivan Berry, Kevin Stewart, Gary Dahrooge, Alan Becknell, Eric Tien (now with SMIC), Ted Rafka, and Todd Mazzie. The author also extends a very special thank you to Ivan Pollentier and his group at IMEC for their excellent work on the controlled shrink process studies ("CD TRIM Techniques").
References
- 1.R. Mohondro et al., Future Fab International, Vol. 1, Issue 3, pp. 235–247.
- 2.S. Kishimura et al., Jpn. J. Appl. Phys., Vol. 38, pp. 250–255, 1999.
- 3.S. Kishimura et al., SPIE, Vol. 3049, pp. 944–954.
- 4.Internal communication with John Hallock, based on his work with shrinkage and etch selectivity improvements.
- 5.W.L. Krisa et al., MNE '96 Conf., Glasgow, Scotland, Sept. 23–25, 1996.
- 6.C. Norton et al., IIT Conference, 2000.
- 7.M. Jones et al., International Ion Implant Conf., Austin, TX, June 12, 1996.
- 8.G. Jordhamo, W. Moreau, SPIE, Vol. 2724, pp. 588–600, 1996.
- 9.B. Ranby, J. Rabek, Photodegradation, Photo Oxidation, and Photostabilization of Polymers, J.Wiley, pp. 165–183, 1975.
- 10.Third-party report on file at Axcelis, Rockville, MD; available on request.
- 11.The process is patented by Fusion Technologies (now Axcelis Technologies Inc.) under US patent # US06117622, J. Eisele, R. Mohondro, Sept. 2000.
- 12.I. Pollentier et al., FutureFab International, Issue 12. Contact [email protected] for more details of the work done at IMEC.
Robert Mohondro is technical marketing manager at Axcelis Technologies Inc. CCS, 7600 Standish Place, Rockville, MD 20855; ph 301/284-5743, fax 301/284-5003, e-mail [email protected].