Cathode-ray tubes are dead. Walk into any electronics store and you'll see aisles of LCD screens for TVs and computer monitors. Even the competing, and initially superior, plasma display technology is under attack from the latest LCD TVs that are more efficient, lighter and deliver sharper images.

Unlike plasma displays, LCD screens employ a backlight unit (BLU) to project white-light emission through the liquid-crystal panels. Cold-cathode fluorescent lamps (CCFLs) are often used for this task, but LED-based backlighting offers an alternative that is also becoming a popular option in laptop computers, where it can extend battery lifetime, thanks to its higher efficiency.

The advantages of LEDs include a further reduction in the weight, thickness and energy efficiency of BLUs. The solid-state source is also compatible with dynamic contrast control, a technology that produces contrast ratios of up to 10,000:1 by switching off LEDs behind dark areas of displayed pictures.

LED options
BLUs produce white light through the careful mixing of the output of red, green and blue LEDs, or by using blue LEDs coated with a yellow-emitting phosphor. Both options involve GaN-based blue LEDs, which share a weakness found in all forms of this device: the trapping of a high proportion of the light generated by the active region.

This inefficiency means that many LEDs are needed to form a BLU, which adds to the cost. To improve light extraction, researchers have developed several technologies to extract light trapped in the chip by total internal reflection.

Commodity LED manufacturers favor a roughening of the wafer's top surface. Random texturing, which is simple and cost-effective, widens the escape angle and allows a significant fraction of trapped light to escape. However, this is inappropriate for BLUs, owing to the lack of directional control over the light distribution. Roughened LED chips produce an essentially omnidirectional light intensity profile within the emission cone, which is fine for many general illumination requirements but undesirable for LCD backlighting applications, where a more structured beam profile is needed to direct the light to the most appropriate places.

Working out how the light gets trapped, and how it can be coaxed from the LED, can pay dividends. Calculations and simulations reveal that there are horizontal trapped modes in the GaN overlayers and the sapphire substrate, and the incorporation of an ordered hole-based grating structure on the emitting facet of the LEDs is a practical and effective way to extract them. The most popular choices are periodic and quasiperiodic arrays of shallow blind holes that form a two-dimensional photonic crystal lattice.

Studies show that LEDs incorporating etched photonic crystal lattices can double the surface brightness of unstructured equivalents and influence the device's spatial emission profile. The Lambertian profile produced by planar LEDs is not ideal for BLUs, but optimization of a photonic crystal structure enables tailoring of the radiated intensity distributions to the characteristics of the particular light diffuser and brightness-enhancement films.

US firm Luminus Devices Inc is already enjoying success in this field with its photonic crystal PhatLight LEDs. These devices are used in some top-end TVs, such as Samsung's 56 inch rear-projection model, because this manufacturer believes that the higher costs associated with that class of LED can be accommodated into flagship products. However, if photonic crystal LEDs are to make a major impact on the overall market, and the key LCD TV segment, then their manufacturing costs will have to plummet to levels that are comparable to CCFLs.

Developing a lower-cost manufacturing technique for photonic crystal LEDs is the primary goal of the Photonic Quasicrystal LEDs for Display Illumination (PQLDI) project, which started in May 2007. This two-year program has £1.2 million ($2.4 million) of funding from the UK's Technology Strategy Board. It involves our team at Glasgow University's Department of Electronics and Electrical Engineering, the Institute of Photonics at the University of Strathclyde and Sharp Laboratories of Europe.

Quasicrystal benefits
Conventional photonic crystal LEDs, such as those made by Luminus, have attracted a strong patent portfolio, so we have directed our efforts towards quasiperiodic photonic crystal structures. When it comes to display backlighting, this type of photonic crystal has a major advantage over its conventional cousin: the quasiperiodic arrangement of holes delivers freedom for optimizing the output-beam profile, and this in turn simplifies design tweaks.

Creating the photonic crystal structure is a key process step for fabricating this type of device. Modern "deep submicron" optical lithography systems are an obvious choice but not necessarily a good one. Although they can print photonic crystal features with deep submicron-scale patterning, this expensive tool is not found in most LED fabs.

A cheaper alternative is electron-beam lithography. This is an affordable option that is widely used by researchers investigating photonic crystal devices. However, we have ruled it out. Its incredibly slow writing speed means that pattern-writing costs are prohibitively high. Covering an area of just one square inch takes several tens of hours, which translates into beam-writing costs of thousands of pounds – unacceptably high for volume manufacturing of commodity chips.

Instead we have selected nano-imprint lithography (NIL). This technique involves imprinting patterns on LED epiwafers with a stamp made from a suitable, hard material, such as quartz or silicon. High-resolution techniques, such as direct write electron-beam lithography, can produce the stamp but costs are acceptably low because the stamp can imprint many wafers before it is replaced. In manufacturing environments a master stamp would be made and used to create several working copies.

Our nano-imprinting process employs an Obducat tool and involves pattern transfer from a relief on the stamp to a special NIL resist through the application of pressure and temperature (figure 1). The cross-linking of polymers occurs within seconds to form a hardened resist before the stamp is removed. Dry etching replicates this pattern in the underlying wafer (figure 2).

Our efforts began with silicon NIL stamps patterned with electron-beam lithography, which were dry etched to form relief patterns. These stamps work well and our limited number of research and development samples have not suffered from erosion. However, we discovered that silicon stamps cannot handle high process pressures, and in this case it is better to employ stamps made from quartz or SiC.

The transparency of quartz also allows it to be used for "flash NIL". This involves curing the NIL resist through exposure of radiation transmitted through the stamp material, and this eliminates stamp heating. These advantages have encouraged us to adopt an advanced etch process for making high-definition NIL stamps for imprinting on GaN LED wafers with a 360 nm light source.

NIL-based patterning requires a dual etch because imprinting leaves a thin layer of resist, even over "clear" regions. This is caused by material displacement of the resist, which is in a pliable state during the imprinting contact phase. Fortunately, resist-development processes are not required for NIL. Instead the residual layer is removed before the etch proceeds inside the epilayers. This is usually carried out with an oxygen plasma process before the GaN is dry etched. Several gas mixtures can be used. Combinations of methane and hydrogen are one option, but we and others have found that chlorine-containing gases produce better results.

We carry out our dry etching with an inductively coupled plasma etch tool that produces an almost vertical etch profile. This approach also increases etch rates to several hundred nanometers per minute, which means that it takes about five minutes to define a photonic crystal structure across an entire wafer. This short process time makes the technique compatible with high-volume manufacturing.

Developing a process for mass-manufacturing has required us to address a range of challenges. These include fabricating a stamp with sharply defined patterns; ensuring a good contact between all of the stamp's patterned area and the resist-coated substrate; and preventing the stamp from adhering to the NIL resist so that it can be removed easily after imprinting and resist curing. Soft-resist reflow dynamics mean that a thin layer of resist is left at imprinted locations, and photonic LED fabrication processes must also include a step to remove this layer prior to the transfer of the pattern into the substrate material.

We have developed a proprietary process flow that addresses all of these issues and delivers high-quality imprinted features that are subsequently etched into the substrate. The precision of our process is demonstrated by scanning electron microscopy images of our NIL stamp on silicon (figure 3) and a section of one of our imprinted resist-covered wafers (figure 4). These images reveal the sharp definition of the photonic crystal holes. The fidelity of nanoscale pattern replication is an outstanding feature of NIL, which makes it a strong contender for the next-generation lithography of CMOS chips.

To speed up the nanoscale patterning process for high-volume manufacturing environments, NIL stamps must be used in step-and-repeat tools that imprint tens of thousands of dies in a few hours. This is not as difficult as it may seem, however, because alignment requirements are far more relaxed than those demanded for silicon-chip production.

We, alongside others, are also looking at the possibility of roller imprinting. This has the potential to pattern very large areas at high speeds, and it should be simpler than step-and-repeat imprinting because it only requires a rotary-tool motion. The process is based on imprinting a hard roller with an embossed pattern onto a resist-coated substrate. Pattern transfer can occur by heat-and-pressure or flash-based techniques because GaN epiwafers are deposited on UV-transparent sapphire substrates that allow UV illumination through the wafer's backside.

Work is ongoing in this area and initial results indicate the benefits of a "horses for courses" approach. Smooth wafer surfaces seem to respond better to flash imprinting, but rougher topographies are easier to pattern with the thermal process. Although these techniques still require more development, we believe that they will be taken up by commercial LED manufacturers, probably at the point when GaN LED wafers exceed 6 inch diameters.

After patterning, standard fabrication steps are employed to produce finished LEDs. In our case we make arrays of 350 µm × 350 µm chips that feature a proprietary p-type current spreading layer that uniformly spreads the current over the entire top face of the device. This is a crucial step because the current spreading layer has to be structurally compatible with the etched photonic crystal pattern.

Our devices (figure 5) can operate at more than 200 mA and feature very uniform emission over the device area. Their angular intensity profile is different from the Lambertian profile of conventional LEDs and we are looking to optimize it under the guidance of our backlight-development team.

The final year of our project will be devoted to optimizing processes, improving device characteristics and exploring new soft-mask approaches for the NIL process. Beyond that we are likely to seek further government and industry support to work on large-chip LEDs for mainstream lighting. Such devices pose their own set of challenges, such as effective thermal management, large-area current spreading and high-efficiency color conversion from blue to a panchromatic spectrum.

•  This article first appeared on our sister website, compoundsemiconductor.net