Behind the Photomask

The extension of optical projection lithography through immersion to patterning features with half pitch ≤65nm is placing greater demands on the mask. Strong Resolution Enhancement Techniques (RETs), such as embedded and alternating phase shift masks and complex model-based optical proximity correction, are required to compensate for diffraction and limited depth of focus.

To fabricate these masks, many new or upgraded tools are required to write patterns, measure feature sizes and placement, inspect for defects, review defect printability, and repair defects.

BEHIND THE MASK

Analysis by SEMATECH was coupled with an informal survey of several captive and merchant mask suppliers. Findings showed that the greatest technical capability gaps are in mask cleaning, mask pattern generator throughput with adequate Critical Dimension (CD) control, mask CD metrology precision at high throughput, patterned mask inspection and actinic defect review with sufficient resolution to account for polarisation effects and predictable defect printability.

Beyond the significant technical challenges, suppliers of mask fabrication equipment face the challenge of being profitable in the small market for mask equipment. They also encountered significant R&D expenses to bring new generations of mask fabrication equipment to market.

The total available market for patterned masks is estimated to be $2.5bn per year, about 1% of the predicted total available market for integrated circuits, fabrication equipment and materials (see Figure 1). The patterned mask market is about 20% of the market size for lithography equipment and materials. The largest R&D affordability issue arises for the makers of equipment for fabricating masks, where total available unit sales per year may be less than ten.

SEMATECH has used discounted cash flow models to predict affordable R&D while maintaining industry-accepted internal rates of return. The results have been compared to estimates of the total R&D cost to bring a new generation of mask equipment to market for various types of tools. The analysis revealed that affordability of the required R&D is probably the greatest problem for suppliers of pattern generators, repair tools, cleaning tools and actinic defect review tools. However, affordability of R&D concerns suppliers of all mask fabrication tools and the suppliers of the masks themselves.

POLARISATION AND ABSORPTION

The extension of optical lithography will place greater demands on mask blanks. Depth of focus scales as the inverse square of the numerical aperture, so mask flatness as mounted in the exposure tool must be improved. This improvement will require mask substrates with smaller flatness error. Furthermore, the mask blank substrate is birefringent, affecting polarisation of the transmitted radiation; however, present levels of stress birefringence in fused silica mask substrates are probably low enough to have little impact on imaging.

Immersion tools, at least for the first generation, will have four times the demagnification. For patterning 50nm wide features on the wafer, the mask features are 200nm wide, which is approximately equal to the 193nm exposure wavelength. When feature sizes are in the range of 0.5 to two times the wavelength, the mask patterns partially polarise the transmitted radiation in the transverse electric polarisation state. This enhancement will be manifested as dose variation in systems that do not have purely and uniformly polarised illumination of the mask at all locations.

Simulations by Flagello et al for 32nm–50nm equal lines and spaces on the wafer (mask features are approximately 193nm) indicated that the transmitted field will become polarised by about 12%. Software for designing resolution-enhanced masks, such as embedded or alternating phase shift masks, will require more complex rigorous electromagnetic models, perhaps increasing cost of RET preparation by up to 40%.

Polarisation by mask features might eventually lead the industry to strongly consider using greater than a four times demagnification factor. Understanding and accurately emulating polarisation effects will also be essential for patterned mask inspection and actinic defect review tools, such as aerial image-emulating microscopes to predict defect printability and mask error factor.

Selection of absorber materials may play a key role in managing or using polarisation by the mask to increase the process window. As described by Grenon, the chromium-based absorbers prevalent today are not optimal for reactive ion etching, mask cleaning or repair. Chromium was chosen originally for its hardness and durability, and it has been extended due to its high absorption over a wide bandwidth from visible to deep ultraviolet radiation. Along the way, grading of film composition has reduced absorber residual stress and reflectivity at the exposure and inspection wavelength values. The materials components and surface of the chromium-based absorber stack make it difficult to remove hydroxyl groups to prevent contamination by cleaning residues and atmospheric trace gases cracked during exposure using 193nm radiation.

Chromium’s hardness and durability makes it difficult to remove chemically without significant sputtering in reactive ion etch and focused ion beam etching processes sued for patterning and repair. Although materials with high atomic number increase the magnitude of electron beam proximity effects, an absorber stack-based tantalum may have better properties for patterning and repair. Furthermore, recent simulations by Smith of mask polarisation effects by small features predict that mask absorber material optimisation may be needed to reduce these polarisation effects.

To optimally extend 193nm wavelength optical lithography using immersion, significant changes to absorber films may be required to meet technical specifications, maintain defect-free masks during use in the lab, and achieve desired yield values for mask fabrication processes.

PATTERN GENERATORS

In reducing feature size on the mask, tools such as electron beam and optical pattern generators must be significantly enhanced. Electron beam pattern generators are presently used to write most critical layer masks due to the inherently high resolution of electron optics. Geometrical aberrations and stochastic space charge blur limit the simultaneous achievement of high resolution and fast writing time, since space charge blur is proportional to beam current, and writing time is inversely proportional to beam current.

One of the key challenges for pattern generators is to maintain acceptable mask write times and achieve high resolution, especially to pattern prevalent subresolution assist and serif features added by model-based optical proximity correction. According to Kalk, present e-beam mask writing times range from 10 to 30 hours. The sensitivity of resist at 50kV electron beam energy significantly affects throughput. According to Ando, positive tone materials have high sensitivity of 5–10µC/cm2, and negative tone materials have lower sensitivity at 15–20µC/cm2.

Optical pattern generators have four to five times higher throughput than electron beam pattern generators, and throughput is not related to resolution to first order. Optical pattern generators have resolution limited by numerical aperture and exposure wavelength similar to projection lithography tools since lenses are required to focus the beams for patterning. Perhaps by incorporating immersion these optical pattern generator tools will achieve further resolution advances.

Another key challenge for pattern generators is achieving acceptable CD control and pattern placement. Advances in decreasing systematic CD and pattern placement errors are dependent on the precision of metrology tools. The utility of the metrology tools will be maximised if they can achieve precision values five to ten times smaller than required CD and placement specifications demanded by customers of mask makers.

Today’s leading-edge, scanning electron microscopy-based CD metrology tools have long-term repeatability roughly five times smaller than desired mask CD control targets, but tool-to-tool matching at the same level cannot be achieved. Placement metrology repeatability is roughly five times smaller than the placement error specs, but as with CD metrology tools, tool-to-tool matching is not achieved at the same levels. The number of mask CD measurements required per mask is increasing rapidly, so throughput at increasing resolution is a significant challenge. Quantifying complex pattern fidelity is also increasingly in demand.

Another issue for pattern generators is increasing data volume, which has increased by approximately ten times due to flattening of hierarchy and model-based optical proximity correction, adding many new geometrical elements to the pattern. A recent mask industry survey revealed that mask data file sizes are as large as 263GB, with a mean value of 1.5GB. To compensate for systematic errors in mask writing, mask makers are resizing features in these flattened files. According to Bloecker, data processing times are not only limited by CPU speed, but also by hard disk access times. In some cases, 30 minutes are required to save files to disk. The proposed development of optical and electron maskless lithography tools for direct writing wafer patterns may ultimately be used to achieve economical mask writing throughput with the desired CD control and resolution.

REPAIRS AND DEFECTS

The largest contributors to the cost of mask fabrication are yield, cost and throughput of pattern generators, and cost and throughput of inspection, roughly in that order. A recent survey of ten mask makers, representing approximately 85% of the worldwide mask market revenue, showed that yield loss aggregated over all types of masks, including critical and non-critical layers, is dominated by unrepairable opaque defects (see Figure 2).

To maximise yield and manage mask cycle time, patterned mask inspection tools have increasingly sophisticated image processing algorithms to predict defect printability and the effects of mask CD errors on wafer CD errors (mask error enhancement factor or MEEF). In addition, Zurbrick has recently proposed expanding the application of patterned mask inspection to more stages in the mask fabrication process, to find any disposition defects earlier before further costly steps in the fabrication process are executed.

Development of mask cleaning capability to remove sub-100nm particles without damaging the mask or leaving residues is critical. As noted by Grenon, Marmillion and others, residues left on masks from cleaning have been a source of defect generation during use of masks for both 193nm and 248nm lithography. Salts such as ammonium sulphate have been observed on masks with large areas without absorber patterns used for as few as 300 wafer exposures.

The cleaning processes with sulphuric acid and ammonium hydroxide may leave behind residues that act as precursors for salt formation from atmospheric contaminants cracked by absorption of 193nm photons. Novel cleaning techniques will probably be needed to augment traditional wet chemical and hydrodynamic approaches.

Once defects have been reduced to an average count that may be economically repaired, the capability of repair has a significant impact on mask cycle time. Significant innovation has occurred in the past five years in mask repair; namely, nanomachining and electron-beam gas-assisted repair tools have become available. Most mask makers acknowledge that several types of mask repair tools are needed to minimise unrepairable defects (see Figure 3).

Traditional focused ion beam tools will continue to be used for repair of opaque chromium absorber defects for the next generation or so. Yet the application of electron beam-based repair is likely to increase significantly for inducing deposition to repair clear defects and etch typical embedded phase shift mask materials. E-beam repair allows for potentially higher resolution and more selective repair, preventing damage to and staining of the underlying mask substrate.

Bald notes that one of the key challenges for mask repair is developing accurate defect dispositioning through increased automated handling of data on defects generated by inspection, review and repair tools. Repair of alternating phase shift masks is also a key challenge.

Unwanted fused silica material, typically called ‘quartz bumps’, may be removed using nanomachining or by gas-assisted etching with precise control of etch rate versus position. Atomic force microscopy metrology is being used to characterise these quartz bumps to develop the etch rate versus position recipe used by the e-bema or focused ion beam etch process. New technology must be developed to match the phase of clear defect repairs in embedded and alternating phase shift mask patterns.

Many technical challenges are posed by the continued rapid scaling of IC dimensions and the extension of optical lithography to meet the increasing resolution and dimensional control requirements. Suppliers of masks and mask equipment are likely to continue to be challenged with the required R&D costs. Consortia and government funding will continue to play an important role in enabling the required mask fabrication infrastructure development.