Correcting across-chip linewidth variation by mask substrate tuning
04/01/2007
Engineered arrays of micron-sized scattering pixels, written inside the fused silica mask substrate by a femtosecond pulsed laser, can selectively alter the effective illumination locally, correcting for CD variations at the wafer to a fraction of a nanometer.
As chip dimensions shrink, critical dimension control becomes even more of a challenge for lithographers. Better CD control contributes to improved chip binning (higher speed, lower current leakage, etc.), higher mask or wafer yield, and higher litho productivity. Adequate CD control for future nodes is among the International Technology Roadmap for Semiconductors (ITRS) “red bricks” with no proven solution. Sematech Chair, O.B. Bilous, predicted it will take a lot of “curiosity, imagination, creativity, and passion” to meet ITRS requirements (see example in Fig. 1) [1].
 Figure 1. CD control according to the International Technology Roadmap for Semiconductors, 2005 Edition, Lithography. |
The CD control and process window (PW) in the optical lithography process can be improved by correcting the photomask after its fabrication using a CD control (CDC) technology that employs femtosecond lasers. This fine-tuning method can make litho processes more robust, reduce the effects of error sources, extend the lifetimes of existing reticles, or compensate for reticle or exposure tool deterioration.
 Figure 2. Eliminating a mask CD error fingerprint. The maximum error was reduced from 10.4nm to 6.3nm with 3σ from 6.4nm to 4.1nm. |
In this article, we describe how modifying the transmission of the substrate of the reticle translates into a controlled change in CD on the wafer. Figure 2 shows one example of the improvement in across-chip linewidth variation (ACLV) demonstrated using Pixer’s Shade-In Element technology, which writes engineered arrays of micrometer-sized scattering pixels inside the fused silica mask substrate to selectively alter the mask transmission [2-4]. Because the Shade-In Element formed by these closely spaced scattering pixels is millimeters behind the patterned surface of the mask, it alters only the local illumination and cannot be imaged on the wafer. The CD uniformity was improved by 35%, falling from 6.4nm to 4.1nm (3σ).
Each scattering pixel is formed by tightly focusing a high peak power femtosecond laser pulse inside the mask substrate, resulting in localized breakdown and plasma formation in the quartz. The tiny structural change pixel that remains after the laser pulse has a different index of refraction than the fused silica material. This localized index change (Δn) scatters light out of the aperture of the projection lens, reducing the effective illumination. An array of such pixels with constant density constitutes one shading element. The required correction is predetermined according to CD variation maps which may be supplied by wafer optical CD (OCD), CD SEM, or reticle AIMS measurements. Thus, by changing the local reticle transmission levels, it is possible to correct for existing CD variations, which may result from stepper illumination nonuniformity, mask processing errors, or other phenomena.
In a transparent medium like fused silica (absorption <0.001% per cm), light interacts linearly and propagates mostly forward at normal intensities. However, at intensities of above 109 W/cm2 multiple photon absorption can promote electrons into the conduction band of the material, leading to avalanche ionization and near instantaneous total opacity (100% absorption) across the area that is above the threshold.
As a result of this tremendously fast process, a small mass of the medium is transformed into plasma without going through melting and evaporation. At the high temperatures that are associated with the plasma (105-106 °C), most of the plasma energy is radiated away within picoseconds, before thermal dissipation of phonons takes place. But even before that happens, the intense heat in the plasma zone creates enormous pressure (Mbar scale) that causes a shockwave-expanding at extremely high velocities (tens of km/sec).
The interaction of the shockwave with the cold atoms around it and the rapid cooling of the plasma result in a zone of structural change. This zone quickly freezes with a morphologically different structure from the original amorphous material, typically with a density distribution that is rarified in the center, compressed at the perimeter, and normal outside. Thus, the refractive index at the perimeter of the altered pixel is higher than the rest of the matrix and much higher than at the center. This Δn gives diffraction and scattering effects that can be used for optical elements in the visible and DUV spectrum. Typically, these spots are ~1µm in diameter and extend for ~15µm along the path of the laser. By controlling very accurately the pitch between the pixels, the CDC process can achieve high-precision transmission changes (attenuation), that will translate into a change in CD on the wafer during the lithography process [5].
In variable density filters that are based on such pixels, the attenuation is defined by the density of the pixels in a given area, which is proportional to the square of the pixel pitch. This means that a small change in pitch results in a large change in pixel density and therefore a large change in attenuation. The result is that a filter based on scattering pixels provides a very high dynamic range of attenuation compared to other types of variable density filters.
In today’s advanced DUV lithography, the normalized image log slope (NILS) is ≤2, implying that every 1% change in illumination will give rise to a >1% change in CD. The ratio of CD change in nm to attenuation change in %, is called the CD control ratio or CDC ratio. In practice, a typical CDC ratio is 1.5nm/%. With the laser-based CDC process, it is possible to achieve attenuation values from 0.1% to 5%, potentially correcting CDs by 7.5nm in steps of 0.15nm.
This range covers all typically required CD corrections in advanced masks. In cases where higher attenuation is needed, it is possible to apply multiple correction layers. The application that does that is called M-CDC (multiple-CDC) and it works very similarly to the standard CDC process, except that two or more layers of pixels are written, each 100µm above the previous one. Pixer has verified that such layers of scattering pixels attenuate independently, with total transmission obeying the Beer-Lambert law. M-CDC can achieve 10-20% attenuations, and perhaps even effective opacity, completely inhibiting resist development in the treated area. The M-CDC application also allows reverting a mistaken CDC process to the original condition (i.e. making the attenuation uniform) or fine-tuning a specific area to achieve higher yield at a specific die or to control areas that require different CDs (in SoCs, for example).
 Figure 3. Insertion of pixels into the quartz mask substrate during the CDC process. |
Figure 3 schematically shows the apparatus that inserts pixels into the fused quartz mask substrate. The laser beam is incident from the back of the reticle, which is scanned back and forth through the mask bulk as pixels are written with the necessary density to correct whatever CD error is present at each mask region. By accurately controlling the pitch between the pixels inserted into the glass, the CDC process accurately controls the level of attenuation achieved. The process can be applied after the pellicle is attached to the reticle. No post-processing is required as all effects take place within the bulk of the substrate. The surfaces of the corrected reticle remain pristine and can be cleaned using standard technology.
Using CDC can improve wafer CD-uniformity by a factor of 2, and create a tight distribution (in the range of 1nm 3σ or the actual precision of the measurement tool used to characterize the CD uniformity) in spite of a potentially large mask error enhancement factor (MEEF). All that is required is that the region having the same CD error be large enough. CDC permits changes in CD in local regions as small as 100×100µm. Other techniques allow only global corrections. The process, moreover, is additive; if the first correction isn’t adequate or if further fine-tuning seems desirable, it can be repeated over the entire field and/or at a particular location. Since the CDC process does not require removing the mask pellicle, it can be implemented both by maskmakers and at wafer fabs.
 Figure 4. Eliminating scanner slit related CD error. The CD range fell from 6.7nm to 4.3nm while 3σ was reduced from 3.5nm to 2.2nm. |
An example of practical production correction is demonstrated in Fig. 4 where the wafer CDU 3σ and range were improved by a factor of 1.6. Each field in Fig. 4 contains multiple chips, but there are evident systematic CD variations across the field, with the spatial signature of nonuniform illumination along the scanner slit. The goal is to have all regions in the exposure field yielding high performance chips, and that requires correction of that nonuniformity, which was accomplished by varying the opacity of the mask substrate using CDC. If further correction were desired at particular chips (i.e., at the upper right), M-CDC could be employed. If the scanner were repaired and the initial correction needed to be “reversed,” M-CDC could do that, too.
The process
The actual CDC process is relatively simple and takes ~5 hours/mask. The key is accurate and fine-grained CD sampling. Typical wafer CD metrology recipes might include 5-10 targets per exposure field. Such distribution can only map the global variations, but doesn’t have the capability to identify the smaller-scale ACLV shifts that the technology can correct. Even the 20-50 target CD uniformity verification in mask manufacturing likely will not be sufficient.
 Figure 5. CDC process flow. The micrograph at right shows actual pixels written into a fused silica substrate. |
Figure 5 outlines a generic CDC process that takes place on Pixer’s tool (model CDC101), which can vary depending on the specific application and targets. Every correction starts with a dense array of actual CD measurements that describe the CD distribution to be corrected with the required resolution. This is inputted into the system as a correction file. Based on the control targets and the expected CDC-ratio (the relationship between the attenuation change inside the reticle and the CD change on the wafer must be measured previously), a map is generated of the density of pixels that will be written inside the reticle substrate. The density map is then converted into a “cluster” map where each cluster has the same density and is sized with highest lateral resolution of transmission change. The pixels themselves are then automatically “written” into the reticle glass using a computer controlled femtosecond pulsed laser, a galvo scanner for laser writing, and an open frame linear stage for translating the mask from cluster to cluster and for load/unload operations.
Once the reticle is loaded to the tool, the entire correction is done automatically. The reticle needs no further processing after the laser exposure and can be packaged for delivery. Alternatively, an AIMS tool can be used to confirm the improvement in aerial image CDs.
Conclusion
The CD uniformity improvements obtained with this mask-substrate tuning process clearly translate into major improvements in terms of productivity, yield, better chip binning, etc. The actual control level of DUV attenuation can be a fraction of a percent; leading to precise local exposure control. This high resolution control of attenuation, applied to the mask, allows control of CD errors of <0.1nm even after mask fabrication. There is no other known way to achieve such advanced performance in lithography. The CD uniformity specifications demanded by the ITRS in future nodes likely can be met using this innovative technology.
Acknowledgement
Shade-In Element is a trademark of Pixer Technology Inc. AIMS is trademarked by Zeiss GmbH.
References
- Chip Industry Driven to Innovate, “Sematech Chairman Tells ISMI Symposium,” Semiconductor International, November 1, 2006.
- Y. Morikawa, T. Sutou, Y. Inazuki, T. Adachi, Y. Yoshida, K. Kojima, et al., “In-field CD Uniformity Control by Altering Distribution of the Photomask Using Ultra-fast Pulsed Laser Technology,” SPIE 6283-29, PMJ 2006.
- E. Zait, G. Ben-Zvi, V. Dmitriev, S. Oshemkov, R. Pforr, M. Hennig, “Irradiation Resistance of Intravolume Shading Elements Embedded in Photomasks used for CD Uniformity Control,” SPIE 6283-11, PMJ 2006.
- “CD Variations Control by Local Transmission Control of Photomasks Done with a Novel Laser-based Process,” Semiconductor International, April 2006.
- Guy Ben-Zvi, Eitan Zeit, “Introduction to Mask and Wafer Intrafield Critical Dimension Control,” SEMICON Japan 2007, STS/Lithography.
Contact Gidi Gottlieb, director of marketing, at 44 Maale Camon, Karmiel, Israel; ph 972/4-908-8603, e-mail [email protected].