High-end mask manufacturing using spatial light modulators
04/01/2005
Optical and electron-beam maskwriters using raster-scanning beams have been workhorses in the semiconductor industry for a long time. Today, e-beam writers using vector-scanning beams with small virtual address grids combine high throughput with high resolution for advanced reticles, but as patterning density increases on leading-edge photomasks, throughput drops and costs increase. The introduction of spatial light modulators (SLM) for optical writers opens the possibility of lower-cost production of the most advanced mask layers. This article describes how SLM works and how these systems can leverage new technology developments for wafer-exposure steppers in next-generation mask production.
Optical maskwriters based on spatial light modulators with millions of movable micromirrors have a lot in common with microlithography steppers. The SLM’s optical system is a reflective device, which uses partially coherent light for image formation and higher resolution in a fashion similar to a wafer stepper. The coherence in the illumination causes interference effects (“edge ringing”) that make the edges steeper and improves critical dimension (CD) linearity (Fig. 1).
 Figure 1. This conceptual rendering shows aerial image cross-sections with noncoherent (Gaussian beam) and partially coherent imaging for different linewidths. |
Looking closer at the technology (Fig. 2), the similarities to a stepper are apparent, although there are three fundamental differences: The SLM (corresponding to the mask in the stepper) is reflective, the pattern on the “mask” is controlled by a data channel, and the demagnification is in the range of two orders-of-magnitude higher than in a stepper. A pulsed excimer-laser beam illuminates the surface of a micromirror array, which is imaged onto a mask blank. Micromirrors on the SLM device can be tilted, and the tilt angles are controlled individually by means of electrostatic deflection. Alternatively, pistoning mirrors can be used for the SLM surface phase modulation. The laser light reflected from the micromirror surface is partially coherent, and the image is formed in a diffractive mode.
 Figure 2. Principle layout of an SLM-based optical maskwriter. |
A phase modulation, resulting from deflected mirrors, is converted to amplitude modulation in the Fourier stop (i.e., the aperture in the projection system). This conversion is not general and automatic; it is a result of a deliberate mirror design, mirror size, and layout of the tilt pattern in combination with the design of the optical system. In a more geometric interpretation, this can be understood as light being scattered to higher diffraction orders for tilted mirrors. Since the projected mirror size is small, the phase information has a high spatial frequency. Light scattered in the higher diffraction orders is blocked in the Fourier stop. In the projected image plane, the initial phase modulation has now become an amplitude modulation where the intensity of a mirror is darker with increasing tilt.
During the exposure sequence, the stage moves continuously and the stage interferometer issues flash commands to the laser. The laser flash time is short enough to freeze the stage movement during the flash. Each flash images adjacent portions of the pattern (known as SLM stamps) with a small tapered overlap until a full stripe is finished. Then the stage retraces, steps to the next stripe, and starts the exposure and so forth. Multipass writing with shifted SLM stamp positions reduces the effects of irregularities due to malfunctioning pixels, pulse-to-pulse dose variations, overlap effects, etc. This provides a balance between write time and pattern accuracy. In the data path, the vector format pattern is rasterized into grayscale images, with gray levels corresponding to dose levels for the individual pixels. The pixel grayscale values are converted to drive voltages for each pixel cell. The image processing is done in real time using programmable logic.
Calibration techniques
Maskwriting quality using SLM technology is largely dependent on the quality of the MEMS processing. Varying process properties, resulting in micromirror variations, call for various corrections in the maskwriter. The first and most obvious challenge is variation in deflection characteristics between mirrors, resulting in a need for calibration.
An integrated system-calibration tool, where the SLM is imaged onto a CCD camera, is used for this purpose. The transfer function of each individual micromirror is extracted and stored. During writing, the individual voltage-controlled deflections are corrected according to the stored transfer functions. In addition to corrections of mirror-to-mirror variations, the calibration system also takes care of other causes of dose variation over the exposure field.
Mirror layout
The design of the SLM micromirrors can, in principle, be made of several different types of structures. The majority can be divided into the two main groups of tilting and pistoning mirrors.
In the pistoning mirror arrangement, each mirror in a 2D matrix can be set to an individual height. A height change corresponds in this case directly to a phase change. A pistoning mirror pixel has to be defined in a scheme where at least two mirrors operate in a pair with movement in opposite direction to emulate a transmission mask. The reason for this is that the complex amplitude reflectance has to stay on the real axis to achieve grayscaling, which is not the case for a single pistoning mirror.
Another way of describing the criteria for emulating a transmission mask is that the mean phase over a resolved image area has to stay constant with deflection. The strong phase effect inherent in piston mirrors makes them susceptible to through-focus artifacts, unless the individual mirrors are very small. An advantage that follows from the average phase modulation of a moving piston mirror is that a strong phase-shift mask can easily be emulated.
For a tilting mirror design, the balance of phase with deflection is automatic. Therefore, tilting mirrors can be rasterized based on local data for the individual pixel. This makes the data path architecture simple. And since fewer micromirrors are required for a certain pattern, the tilting mirror design allows for less demagnification and higher throughput than an equal quantity and size of pistoning mirrors on the SLM.
A third conceivable option for parallel writing using SLM technology would be based on a grating light valve device from Silicon Light Machines, in conjunction with an array of focusing Fresnel zone plate lenses, to create a one-dimensional array image of individually dose-controllable dots. This architecture has the potential for high throughput, but does not benefit from the advantages of partially coherent imaging.
Image enhancement
Since a maskwriter with an SLM can be viewed as a scaled-down stepper with an adjustable mask, performance enhancement techniques used in wafer-exposure technology - such as optical proximity correction and off-axis illumination - can be applied. In addition, the dynamic nature of the SLM compared to a static mask allows for some extra resolution enhancement and image control features. Pattern adjustment operations can be performed on the bitmap level of data processing. Examples of such operations are x/y-biasing, feature corner enhancement, optical distortion correction, and dose correction from one writing pass to the next. Edge fidelity improvements also can be performed by taking advantage of the fact that a negative reflection phase is accessible for mirror tilt angles beyond the normal black level.
In a further development of the technology, the tilting mirrors can be modified with a λ/4 phase-step at the center of the mirror to give strong phase-shifting. Since half of the mirror area is shifted 180° (after reflection), the image from the mirrors is dark when the mirrors are not actuated. Tilting the mirrors to one side brightens the image up to ~50% reflection. Tilting them to the other side also makes the image brighter, but now with reversed phase. Thus, with the described phase-step mirrors on the SLM, the maskwriting system can be used arbitrarily in binary, attenuated, high-transmission attenuating, three-tone, alternating aperture, phase edge, and CPL modes. The only disadvantage is a loss in maximum brightness of ~50%, which has to be corrected by increased laser power.
An additional advantage of phase-step mirrors is that mirror tolerances become less demanding. For instance, the contrast remains high for moderate mirror nonplanarity. This is a property that simplifies stitching of consecutive SLM images and improves yield in the manufacturing process.
The SLM chip
The basic building block of an SLM optical writer is a high-voltage CMOS circuit with an integrated micromirror MEMS device on top. The SLMs with electrostatic actuation and tilting micromirrors are tailor-made for maskwriting.
Every pixel cell has an electrode under a part of the mirror, connected to a storage capacitor and a transistor. This design allows the electrode to be charged to an analog voltage and then be isolated (Fig. 3). All pixels are blockwise addressed in sequence during the loading of a new frame by scanning columns and rows, loading an individual analog voltage into each cell. The electrostatic force pulls the mirrors and causes them to tilt. The balance between the analog voltage and the stiffness in the mirror flexure hinge determines the exact angle. High address resolution with a virtual address grid is enabled through grayscaling by analog-addressing the micromirror tilt angle. The actual resolution is limited by the digital-to-analog converters providing the drive voltages.
 Figure 3. SEM photo of mirror, hinge, and post areas for a tilting micromirror MEMS device. |
Micromirrors made from a sputter-deposited aluminum alloy offer good optical properties. These structures are also fairly straightforward to manufacture, resulting in a high yield and good mirror flatness. One serious concern is material creep, due to mechanical hysteresis caused by grain boundary sliding in micromirror hinges under stress. The tilt history of individual mirrors causes a modification in performance and deflection. At first glance, this problem seems impossible to correct, but it turns out that the amplitude of the hysteresis effect can be almost completely suppressed for the tilting mirror layout by using a specific addressing scheme. The scheme benefits from reducing the micromirrors’ deflection angles during the addressing cycle. A global voltage is applied on the micromirror surfaces that is halfway between the addressing voltage for “white” and that for “black,” thus having electrostatical forces both at the address and the counter electrode, which compensate for each other (Fig. 4). This global mirror electrode voltage is switched off just before the laser flash to allow all mirrors to assume the requested address deflection states. The duty cycle for large deflections, critical for the hysteresis magnitude, is thereby greatly reduced, in turn decreasing the impact of material creep on address resolution to a negligible level.
Digital light processing (DLP) mirror technology, developed by Texas Instruments Inc. and used for digital projectors, is not suited for optical photomask writers for several reasons. The DLP mirrors, with only two steady deflection states, achieve grayscaling by time modulation, not by amplitude modulation. While DLP is a sophisticated and highly successful technology for visual displays, it is not compatible with a pulsed excimer laser as a light source. The pulse duration is far too short to allow for any time modulation during the laser pulse. If, on the other hand, a continuous laser were used, grayscaling in the time domain would be possible but at the price of an architecture where the stage could not move during stamp exposure. The noncontinuous movement of the stage would lead to a more complex system with much lower throughput. The information rate also is inherently lower for a digital technology like the DLP than it is for an analog SLM technology.
Future outlook
Because of the similarities, SLM maskwriters can benefit by migrating down the same technology roadmap as microlithography steppers. These advances include transitioning to shorter wavelengths - even going to EUV in the distant future - and immersion technology. Stepper manufacturers will have to solve many of the challenges associated with these new lithography technologies years before photomask writers will need them. Therefore, SLM maskwriters have the advantage of adopting new technology once it has reached a mature state. With future SLM maskwriters, featuring a larger number of even smaller micromirrors and improved data-channel architecture for faster transfer rate, optical maskwriting will be in a position to follow the development of the semiconductor industry for many years to come.
Acknowledgment
CPL (chromeless phase lithography) is a trademark of ASML MaskTools.
Ulric Ljungblad received his PhD in physics from the U. of Gothenburg and is senior specialist in SLM technology at Micronic Laser Systems AB, PO Box 3141, 18303 Täby, Sweden; ph 46/31-703-0683, e-mail [email protected].