Due to a very small depth of focus, standard photolithography techniques have insufficient fidelity for defining photonic crystal structures on LED epiwafers. But highquality, large-scale patterning is possible by turning to a novel self-imaging photolithography technique, say Harun Solak, Christian Dais and Francis Clube from Eulitha.

The introduction of light extraction techniques has spurred an increase in GaN LED efficiency. That’s because this type of technology can prevent a high proportion of light being trapped and eventually absorbed in the high refractive index semiconductor layers where emission is generated.

One of the most effective light extraction technologies involves etching regular arrays of holes into the emitting surface. Such photonic crystal structures cut the proportion of light propagating within the guided modes of the high index dielectric layer, and channel more emission out of the structure. Indeed, some of the highest performing devices have been created in this way by researchers at manufacturers such as Philips Lumileds and Osram. In addition, there are highly successful commercial products employing this concept, such as the series of Phlatlight LEDs from Luminus Devices.

There are many different ways to incorporate photonic crystal structures into LEDs. Arguably the most straightforward is the etching of a hole structure into the top layer, leaving air holes that can extend into the active region. Other approaches include incorporating a lower index dielectric - such as silicon dioxide - as pillars in GaN, or patterning the substrate with a photonic crystal structure.

Adding a photonic crystal pattern influences light emission in two ways: overall extraction increases by a factor of two-to-three; and the emission profile changes, becoming more concentrated around the surface normal. This enhanced directionality of the emission results in a brighter LED, which is especially important for applications where the light needs to be further guided, collimated or focused. Through careful design of the photonic crystal structure it is possible to tailor the LED’s emission pattern to the target application. Electromagnetic simulation codes are one tool for realizing this. They can determine the effect of parameters such as the period or lattice symmetry.

An alternative, popular practice within the industry for boosting efficiency is roughening or texturing of the LED surface. This introduces facets at different angles, making it easier for light to escape from the chip. One downside of this approach is that it offers no gain or control over the directionality of the emitted light. What’s more, by turning to an optimized and highly controlled photonic crystal pattern instead of a textured surface, it may be possible to increase the process and emission reproducibility across chips and wafers, leading to higher production yields.

Despite the research programs being undertaken by major manufacturers and their published results, the application of photonic crystals to LEDs has not yet been adopted extensively by this industry. That’s primarily because this approach is believed to add substantial complexity and cost to the lithographic process required for the fabrication.

Printing versus optics

Photonic crystals designed for light extraction from GaNbased LEDs typically have a lattice period of 300-600 nm. The lattice symmetry is usually hexagonal, although other geometries such as square grids are also considered. In order to realize such patterns, holes as small as 100 nm in diameter need to be printed on LED wafers. This rules out proximity (or contact) photolithography, which has a minimum resolution of about 500 nm. And features of this size are only possible in contact mode, where damage to the mask and process yield are both serious issues.

 



Photographs of a 600 nm period hexagonal pattern on a 2-inch wafer

Another optical method is holographic lithography, which involves interfering of two or more mutually coherent beams to obtain periodic structures. Resolution is not an issue for this UV method, but it is unsuitable for high volume production processes because the optical configuration has to be modified to realize different patterns. In addition, this approach requires a strict control of the environment to maintain stable fringe patterns.

Nanoimprint lithography (NIL) has been proposed as a suitable technique for high-volume manufacture of photonic LEDs because it promises to combine sufficient resolution with high throughput. Some manufacturers have indeed adopted this technique, and all main Nanoimprint tool providers now advertise equipment specially targeting this application. However, the NIL approach faces considerable challenges: process difficulty, cost and throughput. These difficulties are partly caused by the non-flatness of LED wafers and the particulate contamination commonly found on their surfaces. There are ways to circumvent this problem, but they add to process complexity, which ultimately increases cost. Meanwhile, deep UV lithography, as used by the IC industry, is not considered to be a viable option due to its prohibitively high cost. In addition, there are depth-offocus problems associated with printing high-resolution patterns onto non-flat LED wafers.

Staying focused

At Eulitha, a start-up founded in 2006 in the canton Aargau of Switzerland, we have developed a proprietary technology to address this important manufacturing roadblock. Our technology that is known as PHABLE – a shortening of “photonics enabler” - is a mask-based photolithographic technology that takes full advantage of standard photolithography infrastructure such as photoresists and associated processes. It enables fabrication of periodic structures required for photonic applications, such as arrays of holes arranged on hexagonal or square lattices, or linear gratings with sub-100 nm resolution. The unique property of PHABLE is that it forms an optical image with a very large depth of focus (DOF), which means that it is not a problem to print high-resolution patterns onto non-flat surfaces, such as LED wafers.

Self-imaging of gratings (or the Talbot effect) is a wellknown phenomenon where a mask with a periodic structure (grating) is illuminated with monochromatic collimated light to form images of the grating at periodic distances after it. These so-called self images have a DOF that scales with the square of the pattern period. A typical DOF value for a pattern period of 400 nm, illuminated with 365 nm light, is 50 nm. This value is so small that it prevents use of non-flat substrates or sufficiently thick photoresists, and requires very precise positioning and alignment of the wafer with respect to the mask. While there have been many demonstrations and research studies, up until now the limited DOF has prevented application of this method in industrial fabrication, especially for high-resolution structures. The PHABLE technology promises to lift this restriction.

In order to explain the principle behind this new technology, we show the intensity distribution and selfimage planes formed behind a linear grating in Figure 1a. According to the conventional method, the photoresistcoated substrate is precisely positioned at one of these self-image planes to record the pattern which has a DOF smaller than p2/2λ, where p is the pattern period and λ is the wavelength. In the PHABLE method the wafer is not kept stationary at a self-image plane. Instead, it is moved toward the wafer by a full Talbot period (p2/2λ) to record an integral or average image (Figure 1b).

 

 



Figure 1 (a) Calculated image produced by a linear diffraction grating illuminated with monochromatic collimated light. The dashed lines show two of the self-image planes. (b) Diagram illustrating the PHABLE concept. (c) Resultant image obtained with the innovative method showing the invariance along the longitudinal direction and hence elimination of the DOF limitation faced in conventional photolithography

The resultant image is shown in Figure 1c. This image is also periodic along the lateral direction but, interestingly, is not sensitive to the starting distance of the wafer from the mask. Therefore the image has effectively no DOF limitation. A further advantage is that the printed pattern has half the period of the grating in the mask, therefore a resolution gain is achieved with respect to the mask.

To ensure a reliable and reproducible lithographic process, the contrast of the aerial image has to be high enough so that the non-linear response of a photoresist converts the image into a binary pattern. An inspection of the calculated image in Figure 1c reveals a peak-to-valley intensity ratio of about three – this is ample contrast for photoresist exposure. In general, simulations show that images obtained with this method have high contrast, which is supported by the experimental results presented below.

Any pattern you like

This principle illustrated in Figure 1 is applicable to both one-dimensional patterns, such as lines and spaces, and two-dimensional patterns, such as hexagonal or square lattices. Examples of patterns printed using this method are shown in Figure 2. A hexagonal pattern of holes with 500 nm period printed in photoresist is seen in Figure 2a, and top-down and cross-section images of a hexagonal array of holes with 600 nm period etched in a silicon wafer are shown in Figures 2b and 2c, respectively. Exposures were performed with a PHABLE tool using collimated UV light and a standard photoresist. In each case the wafer was displaced over one Talbot period during exposure to print large-area patterns over 2-inch wafers.

 



Figure 2 SEM images of photonic patterns with hexagonal symmetry. (Bottom left) Pattern in photoresist with 500 nm pitch. (Right) and (Top left) Top-down and cross-section images of 600 nm-period hexagonal hole array etched to a depth of about 800 nm in silicon

 

Evaluation of the printed structures showed that good uniformity and reproducibility were obtained despite an uneven gap and large resist thickness, proving that the pattern is indeed insensitive to the distance between the mask and the wafer. The large gap between the mask and the wafer ensures a practically unlimited lifetime for the masks.

Since PHABLE is a mask-based photolithography method, printing a different pattern simply requires a change of mask. What’s more, many different patterns can be simultaneously printed on a single chip or a wafer in much the same way as different circuits are printed on silicon wafers. The limiting resolution of the printed features depends on the wavelength of the light used, with the smallest period being close to half the wavelength.

PHABLE is ideally suited for patterning LED wafers because of its non-contact nature and ability to print over large topographical features and on non-flat surfaces. Photonic nanostructures can be created on LED surfaces after epitaxial deposition steps or on sapphire substrates before the device layers are grown. Relatively thick standard photoresists can be used, such as those with a thickness of 0.5-1.0 μm. This enables etching into semiconductor layers without the added complexity or cost of hard masks, such as SiO2. Photonic crystal patterns with various different periods, orientations or symmetries can be incorporated on individual chips to effectively tailor and control the distribution of light emission. The high reproducibility and uniformity of the lithographically produced patterns can improve yield and reduce the costs associated with binning products with large performance variations.

Other emerging technologies also stand to benefit from this innovative photonic patterning technology. For example, lithographic patterning for nanowire-based LEDs and photovoltaic devices can be accomplished with PHABLE. Heteroepitaxy on patterned silicon substrates and epitaxial lateral overgrowth for Blu-ray laser production are other potential applications. Wire-grid polarizers needed in both LCD displays and projectors are other areas where this technology can make a strong impact.

The PHABLE technology enables low-cost fabrication of photonic patterns. The time tested approach of a maskbased UV exposure and its associated infrastructure will ensure a smooth adoption of this approach. In particular, there is no requirement to invent or develop new materials. Standard photoresists with optimized resolution and etch properties are available from multiple vendors. The infrastructure for mask fabrication is also already in place. This means that the HB-LED and other industries can rely on the usual, well-established sources for the required consumables and a low-cost process for realizing their photonic nanostructures.

We are now offering samples and wafer batch processing services to companies and researchers developing nanostructure-based products, who are interested in taking advantage of this breakthrough technology. We are also currently offering laboratory lithography tools for 2- inch to 4-inch wafers that are suitable for product development. High-volume production tools with throughput in excess of 100 wafer per-hour will be made available to manufacturers in the near future. Many future photonic devices will shine even brighter with the introduction of our proprietary technology.

Further reading

Jonathan J. Wierer et al, Nature Photonics, (2009)

K Bergenek et al, IEEE J. of Quantum Electronics, (2009)

F. Rahman, Optics and Photonics News, (2009)