The photomask industry adopts the 50keV e-beam
09/01/2002
Millions of dollars are spent each year keeping up with the frenetic pace of leading-edge manufacturing requirements. As each new manufacturing tool upgrade becomes available, users must ask the obvious questions regarding whether or not their incumbent tools can limp along for a few more years or survive another technology node reduction.
In the photomask electron beam lithography world, many 10keV and 20keV platforms have been installed over the past 15 years or so. While resolution capability was superior compared to that of the optical steppers and other pattern generators at the beginning of that era, write times increased dramatically as design grids decreased. In addition, the critical dimension (CD) uniformity and linearity requirements imposed by semiconductor customers taxed the tools' capabilities with respect to stripe butting, subfield stitching, and stage errors.
 |
Ultimately, even resolution capability became sub-standard when attempting to replicate optical proximity correction (OPC) features that have decorated circuit patterns these last few years. Add that to the fact that circuit pattern density has become so high that beam fogging became problematic [1].
What was needed was a smaller-diameter electron beam that had the complementary stage and subsystem architecture capable of producing photomasks for the ever-shrinking patterns of the 130nm and 90nm technology nodes. Regardless of the original equipment manufacturers' (OEM) beam shape and pattern-generation strategy, all seemed to jump from wherever they were - 10keV or 20keV - into producing new platforms with 50keV accelerating voltage columns. In other words, regardless of imaging strategy (Gaussian or shaped-beam) and scan type (vector or raster), all camps decided that 50keV was the right solution.
Increasing beam voltage: a good thing
In order to understand why increasing beam voltage is a good thing, let us first look at the physics involved in electron beam lithography. There are three major factors that need to be understood and controlled in any e-beam lithography situation: forward electron scattering, backward electron scattering (also called backscattering), and substrate heating.
 Figure 2. Energy distribution a) in resist at 10keV and b) in resist at 50keV [3]. |
The lithographic exposure of resist is a result of inelastic collisions of the electrons with resist molecules. Electron energy losses resulting from collisions are transferred to ionization, excitation, or heat, leading to molecular chain scission or cross-linking - both forms of exposure. Other electrons fall victim to scattering. Forward scattered electrons (FSEs) have a net motion in the same direction as the incident beam, and backscattered electrons (BSEs) are those that have a net motion in the opposite direction. Most forward electron scattering can be attributed to elastic collisions with atomic nuclei in resist molecules and mask substrates. Backscattered electrons can be reflected both from the underlying substrate and from the e-beam tool's column ring itself.
Using Monte Carlo simulations
Problems occur when scattered electrons deflect in a direction that exposes resist in an unwanted area. A diameter as large as 10-20μm can be affected by column ring and substrate BSEs. The old adage - if you can't measure it, you can't improve it - holds true, so in order to reduce unwanted lithographic effects, the trajectories of scattered electrons must have quantifiable and accurate predictive simulation results.
 Figure 3. Electron distribution at a) 10keV and b) 50keV [3]. |
Electron scattering can be predicted using several simulation strategies. Bethe (1938) first proposed a semi-empirical model. Among others, Tomlin (1966) and Green (1963) both proposed different types of probabilistic models based on Boltzmann's differential transport equation, solved either numerically or with the use of simplifying assumptions. Of interest in this evaluation, however, are Monte Carlo simulations, first derived for x-rays by Green and Bishop.
In Monte Carlo simulation analysis, an electron possessing an initial energy E0 loses some of this energy as it randomly undergoes inelastic collisions with atoms while still traveling in a straight line. From time to time, the electron collides elastically with an atom's nucleus, thereby changing its direction. After another straight-line episode, it will collide with yet another atomic nucleus and change its direction again.
This process is repeated until the electron has lost all of its kinetic energy (Fig. 1). During the collisions, the electron has given its energy to other atoms in random locations, some of which may have been resist molecules not intended to be exposed. The higher the electron energy density near the mask surface, the more unwanted exposure will undoubtedly occur. The goal is to achieve resist exposure using a small-diameter beam that sends most of its scattered electrons deep into the substrate, away from the rest of the resist. Once achieved, one would expect better dose latitude and CD linearity in the fully processed photomask.
 Figure 4. Process performance at a) 10keV and b) 50keV [4]. |
To understand more about resist exposure effects, Monte Carlo simulations of many electrons at varying acceleration voltages can be examined. Figures 2a and 2b show the contour lines from actual Monte Carlo simulation runs where 100,000 electrons were tracked as they traveled through 400nm of resist. Figure 2a is from a 10keV accelerating voltage column; Fig. 2b is from a column running at 50keV.
One can easily see by a comparison of deposited energy distributions that the 50keV beam has a much smaller effective diameter than the 10keV beam. Looking specifically at 16.0 eV/ cm3/electron, the effective diameter at the resist-substrate interface is approximately 450nm at 10keV, while it is only 60nm at 50keV.
Monte Carlo simulations of actual traces of 100 electrons are shown graphically in Figs. 3a and 3b. In these plots, the trajectories of each electron were plotted as defined by the model in Fig. 1. The result is a cross-section of what would have been experienced by the resist, chromium, and fused silica - a standard binary photoblank.
Noting that the scales of Figs. 3a and 3b are vastly different, the effective beam diameter of the 10keV column is wider in the resist and shallower into the substrate than the 50keV column. As a matter of fact, all 100 electron trajectories from the 10keV case can be contained within the depth of the first y-scale tick mark in the 50keV figure! Likewise, the maximum BSE spread from the 10keV simulation can be contained within one x-scale tick mark. The higher energy beam irradiates a considerably larger substrate volume due to its deep penetration. As an example, a 5× increase in accelerating voltage creates a 20× increase in vertical and lateral electron spread.
How does this large increase in exposed area not adversely affect the end lithographic result? To answer this question, the energy differences between forward and backscattered electrons during photomask exposure need to be explored. Going back to Figs. 2a and 2b, regardless of how voluminous the 50keV electron density in the substrate, the affected area in the resist is very narrow. Hence, one would expect very high resolution when moving to a higher energy source. In Figs. 3a and 3b, the effective diameter where an occasional BSE might expose an unwanted resist molecule is much larger in the 50keV case. The BSEs have much lower energy than the FSEs, however.
Benefits of BSEs
Since backscattered electrons from a 50keV column have considerably less energy than from a 10keV column, errant unwanted exposure is unlikely when using a 50keV tool. The result is better dose latitude and CD linearity on the photomask. Figures 4a and 4b show how the process engineers would have more dose latitude if they used a 50keV e-beam.
 Figure 5. 50keV CD linearity improvement [4]. |
At 10keV, the change in CD over a given change in perecent dose is approximately 3.2nm with about a 225nm over-development. This is in contrast to that of the 50keV case where the CD change is only 2.2nm and there is only about 125nm over-development. Figure 5 shows similar linearity plots for both line/space and hole patterns at 50keV, while at 10keV, hole pattern linearity falls off dramatically. There are also high future expectations for CD linearity improvement at 50keV as beam proximity correction algorithms become better understood.
Challenges in moving to a higher beam energy
Even though CD resolution and process control improvement have been enough to sell many 50keV tools, there are still several issues surrounding the move to higher beam energy that could result in making performance sacrifices and tradeoffs. These include throughput, beam blur, and resist/substrate heating.
A phenomenon that occurs with increased beam energy is an atomic energy loss reduction/unit length in the resist. In other words, as the higher energy electrons travel through the resist during exposure, they do not transfer as much energy to the resist molecules. The result is an apparent reduction in resist sensitivity, and hence, a noticeable slowdown in throughput.
Unfortunately, physics is physics, so there is little that can be done with the e-beam tool itself aside from raising beam current. If current were raised substantially, many other negative factors would be exacerbated, including beam blur and substrate heating. What has been done by photomask makers is to migrate to chemically amplified resist (CAR) strategies. By using a more sensitive resist, less dose is required to expose the same volume of resist polymer, which allows the stage to move more quickly under the beam.
 Figure 6. Process latitude vs. beam blur increase [5]. |
Another problematic area comes from increased electron Coulombic interaction as shaped beam energy is increased. The result of this interaction is a widening of the intended beam area. This phenomenon is called beam blur.
The graph in Fig. 6 depicts how, with increasing beam blur, CD control degrades for all resist thicknesses. There is still a great deal of work in progress at both the tool and process levels to minimize the effects of beam blur.
The final major area of concern with higher acceleration voltage e-beams is resist and substrate heating. At high voltages and current densities, considerable energy is transferred into the resist, chrome, and fused silica. This energy is converted into heat, which raises the resist temperature locally and could result in unwanted polymer chain scission in positive resist or cross-linking in negative. Since most of the electrons wind up in the fused silica substrate (an insulator), considerable cooling time is required to keep the unexposed resist temperature from reaching the point where an unwanted thermal "exposure" reaction could occur. If pattern densities are sparse, the mask has the chance to cool down, but with leading-edge memory and logic devices, pattern density does not allow for adequate substrate cooling.
Several simulations have been performed to study the effects of resist heating on surrounding features [6-8]. The net result shows that the resist image in one location is affected by heat transferred from a previous exposure made in fairly close proximity.
A simple checkerboard write was chosen to simulate dense circuit patterns. In this pattern, substrate temperature rises in areas located well into the pattern, while the pattern edges are substantially cooler. The study points to the fact that beam proximity effect correction algorithms need to incorporate heating effects as a function of dose and pattern density in order to maintain CD integrity across a nonuniform pattern load [9].
Conclusion
Resolution, process latitude, and CD linearity benefits currently outweigh the negatives regarding 50keV e-beam adoption. An impact has already been felt in the industry as 130nm and even 90nm node photomasks are being produced as a result of this enhanced capability. Future challenges for OEMs and process engineers will certainly include increasing throughput, and reducing beam blur and substrate heating. Things will only get tougher as the industry inevitably drives 50keV tool capability to meet the 65nm node and to provide development samples for the 45nm node. It will take the efforts of all engineering functions to pursue the mask industry's endeavors as aggressively as requested by its most advanced customers.
Acknowledgments
The author wishes to thank Dr. Pierre Sixt of Photronics-Europe for his advice and his editing of the technical content of this article.
References
- N. Takahashi, et al., Proc. SPIE Vol. 4186, pp. 22-33, 2000.
- D.F. Kyser, K. Murata, Proc. 6th Int. Conf. on Electron and Ion Beam Science and Technology, San Francisco, pp. 205-223, 1974.
- C.A. Mack, Proc. SPIE Vol. 3236, pp. 216-227, 1997.
- B.-C. Cha et al., Proc. SPIE Vol. 4186, pp. 508-512, 2000.
- S.-H. Yang et al., Proc. SPIE Vol. 4186, pp. 468-473, 2000.
- S. Babin, Proc. SPIE Vol. 3546, pp. 389-397, 1998.
- H. Sakurai, et al., Proc. SPIE Vol. 3748, pp. 126-136, 1999.
- F. Abboud, et al., Proc. SPIE Vol. 3748, pp. 385-399, 1999.
- A. Wei, et al., Proc SPIE Vol. 4186, pp. 482-493, 2000.
Benjamin G. Eynon received his BS in microelectronics engineering from the Rochester Institute of Technology. He is director of product development and integration at Photronics Inc., 201 Michael Angelo Way, Austin, TX 78728; ph 512/248-6169; [email protected].