Introduction to Pattern Generators
The concept of maskless lithography typically refers to an application that makes the use of a photomask economically unfeasible because the volume and value of the objects written will not support the cost of a mask. It does however imply a volume significantly larger than one and a rate of production of minutes rather than hours. This has many implications that will not be covered below such as, for example, how much time we can spend on the data preparation. This tech talk is about the optical side of things.
Even when we focus on optics only, there will be differences between maskless and a mask writer. Both do practically the same thing, they write a pattern on a surface covered with a photo-sensitive material. The difference is that the mask-writer typically targets significantly higher image quality and the volume is one pattern per object written, which we typically refer to as a photomask. We have time to adapt the data to the projection system and adapt the projection system to the data, although this is rarely, if ever, used. Available time is usually one to several hours while for the maskless application, it is one or several minutes. There is no time to adapt anything.
Imaging Theory Principles
Every pattern generator requires an active optical element and the choice is not only a matter of technology and practicality. However, no matter the choice, we can take for granted that the projected image using this active element is much smaller than the size of the photo-sensitive surface we intend to expose with a pattern. Therefore, a scanning or stepping motion is always implied.
The choice of pattern generator also affects ultimate resolution. Scanning the workpiece with a spot of light, which we move with an acousto-optical deflector, results in incoherent image formation because we can only add intensity over time. If the modulator exposes a two-dimensional object, the imaging becomes coherent or partially coherent, we can now add or subtract amplitude, and the achievable resolution can be pushed toward a quarter of the wavelength. However, even if we stay with simpler modulators that don’t modulate the phase of the light, the achievable resolutions can still be pushed below half of the wavelength even without adapting the illumination to the pattern. Some details of this are covered in an earlier tech-talk.
The DLP – An Unexpected Champion
From a pure imaging theory point-of-view, the DLP is the strange kid on the block. Pattern generators need, generally speaking, to be able to create gray-scale images and the DLP is a binary, wax-on, wax-off, type of device. The light source is of prime importance for any imaging system. This is true -with interest- for a pattern generator. The DLP was once designed for the high-intensity arc lamp while a spatial light modulator aspiring to be a small programmable mask is expected to be illuminated a laser source, such as an Excimer laser.
Origins and Limitations of the DLP
For someone like myself, who started out in the KrF – tilt-mirror camp, the DLP appeared to be a poor choice. It doesn’t have the same limit resolution since the implied gray-scaling method somewhat degrades the effects of partial coherence. However, all of that which seemed so important when optical mask writers still held the promise for 90, 65 and 45 nm nodes, has gone by the way of the dodo bird two decades later.
The Changing Landscape of Pattern Generators
The race for pattern generators today is no longer resolution. This fight was lost to EBeam many years ago. The battle is now in the field of write capacity and price where the DLP has established a foothold that may be impossible to break. In this segment, the resolutions are moderate. Usually, not even smaller than the wavelength. The value of the exposed wafers will no longer cover the cost of an Excimer laser. Well, strictly speaking they might but the value of the masks will not cover the cost of developing such a machine.
Light Sources and Pattern Generators
The properties of the light source cannot be overstated. The excimer laser which generates hundreds of thousands of independent modes in a single 10 nanosecond pulse solves pretty much all the problems we encounter when designing a pattern generator that exposes a continuously moving workpiece. It fits perfectly the high-end mask-writer application with a two-dimensional modulator. It’s a match made in heaven because, I presume, in heaven we don’t care about money.
What are the alternatives then? As alluded to above, we need independent modes (or photons) to create (partial) spatial coherence and we need those modes to be squeezed into a sufficiently small area (high irradiance), or we need them to be emitted into a sufficiently small cone, and preferably both at the same time. And to add insult to injury, we need this light source to be pulsed so that it can be projected on a moving surface.
Let me sum this up. This light source does not exist. If we exclude the Excimer, we will have to compromise
A Misfit with Traditional Light Sources
The DLP projector and the traditional lithography shop had one thing in common, the high-pressure arc lamp. But the similarity ends right here. The Mercury lamp was used to illuminate a large mask projected with relatively “large” numerical aperture. Even after the Mask Maker’s Holiday, the numerical apertures combined with the size of the illuminated field, the “size” of the light source was still a good fit for a mask while it is a terrible fit for a SLM. With size of the light, I am of course referring to étendue, but that essentially means size or extent so we’ll go with it.
The laser is king in the SLM – Optical Lithography world, at least it used to be. Before we can dive deeper with DLP, I must digress a little. We need to talk about gray scaling.
Revisiting Gray Scaling and Resolution Limits
With analog mirror array SLMs, presumed to be 2D, gray scaling is done in the pupil, using it as a spatial low-pass filter. The degree of filtering can vary depending on how we want to the imaging system to behave close to the resolution limit (mostly). The filtering factor is usually parametrized as a ratio of the numerical aperture and the angular distance between the diffraction modes of the mirror array. Choosing this factor as one-third or a quarter of the fundamental diffraction angle (wavelength divided by mirror period) is usually a good choice, although both smaller and large ratios can be motivated, but values larger than one cannot ever be motivated because that destroys gray scaling through interference.
Analog Mirror Arrays vs. DLP
However, with the DLP, the plot thickens. For analogue devices, size of the light source that can be accepted by the projection system scales the aforementioned ratio squared times number of pixels. Why only the number pixels matter is because the numerical aperture scales inversely with pixel size while the size of the field is proportional to the pixel size and so the pixel size factors out. (please forgive the simplification implied by equating étendue with the pixel count. There is a dimensional error here since étendue is an area times a solid angle of the light cone and the inverse length which provides the numerical compensation is hidden inside the light cone together with the wavelength. However, we are comparing different technologies at a fixed wavelength. Hence the shortcut)
Here, the DLP offers an interesting twist. If the tilt angle, pixel size and wavelength match, the DLP may behave like a blazed grating with good diffraction efficiency. White areas in the pattern will look uniform white, and black will be uniform black even when the aperture includes higher array diffraction orders, and the shape of the mirrors will only be revealed in the transition regions. The system numerical aperture can now be opened up to a ratio above one and the device will now reveal the shape of the pixels in the transition regions. This is a small complication that can easily be solved inside the procedure used to generate gray scale.
The Future of Maskless Lithography: A New Era for DLP
So what’s going on here? Why is this relevant? The reason this the excitement is because the projection system now accepts a light source with a much larger size (étendue). Can we use the I-lamp again? I guess, but it’s even better. it becomes viable to drop the laser altogether and use a UV-LED.
DLP – A Write Capacity Game Changer
There are two aspects that make this exciting. The DLP all but requires a partially coherent light source and creating one using laser diodes is a lot of work, not to mention cost. It requires many LDs, fibers and patience. The LED is a surface-emitting multi-mode source. The only problem we had with it was that we couldn’t squeeze enough light through the projection system. It was too “large”. It’s still too large but the loss of light is beginning to be acceptable even for the pixel count available for today’s UV-DLPs, and if (or when) Texas Instruments releases the 4096 x 2160 modulator for UV, it’s pretty much game over for any other maskless writer solution. This will be king, although the DLP9000XUV device is pretty much already there. The strange kid on the block will, if not already, figuratively speaking, run circles around anyone else when it comes to write capacity using a dirt-cheap source of light.
Final Words
The DLP together with a UV-LED is undoubtedly a very interesting choice for the maskless application, one that will be increasing attractive with the projected development of DLP technology. That said, there are still both design challenges and production challenges such as how to adapt the aspect ratio of the LED to the modulator, preserve etendue, set design targets for the projection optics that match the system requirements, and more. For all those topics, you can find support here at Senslogic. You are welcome to schedule a free initial consultation.
Leave a Reply