Extreme ultraviolet imaging with hybrid AlGaN arrays

At first glance, the fields of solar astronomy and silicon chip manufacture are poles apart. But they do have one thing in common – the need for efficient and reliable imaging of extreme ultraviolet (EUV) radiation, which spans the range 10-120 nm. In solar physics, there is tremendous interest in phenomena occurring on the Sun’s surface (photosphere) and in its atmosphere (chromosphere and corona), such as coronal mass ejections and flares that cause staggering amounts of radiation to be emitted towards the Earth. Such processes occur at extremely high energies, so very short wavelength detectors are needed to observe what is taking place. Meanwhile, the silicon industry is continuing its neverending goal of shrinking transistor sizes by starting to develop lithographic processes involving EUV patterning. This requires detectors sensitive at these wavelengths, which can aid efforts to find and refine techniques for controlling the properties of the UV beam. Solar scientists and silicon engineers can use existing silicon devices for the EUV detection needs. But greater performance is possible by turning to detectors based on GaN, which have an inherently simpler system design and are more robust to UV radiation. Silicon’s weaknesses One of the biggest advantages of using silicon to build any device is that this material and its related process technology are mature, and consequently well understood. However, its bandgap of 1.12 eV means that it absorbs not only ultraviolet radiation, but the entire visible spectrum too, plus infrared radiation up to around 1100nm (see Figure 1). This makes silicon the perfect material for the most common digital cameras and advanced imagers operating in the visible part of the spectrum. But if silicon is employed as the active material for detecting ultraviolet radiation, its sensitivity to the longer wavelengths must be taken into account. Fig. 1. Cut-off wavelength (maximum detectable wavelength) of a material vs. its energy bandgap. Gallium nitride (365 nm) is sensitive only in the ultraviolet range (“visible blind”). Increasing the aluminum concentration in the compound above 40 percent (Al0.4Ga0.6N) provides “solar blindness”: cut-off below 280 nm, which is the lowest wavelength produced by the Sun to reach the surface of Earth (sea level) through the atmosphere In solar science this is important considering the fact that the solar spectrum is several orders of magnitude more intense in the EUV than in the visible range. In practice, this means that additional filters are needed for removing unnecessary visible and infrared wavelengths. Adding filters brings major, unwanted consequences. In the EUV, all materials are highly absorbing, so any layer between the radiation source and the detector is highly undesirable because it considerably limits incoming flux. What’s more, if these filters degrade – for example, contamination and/or pinhole formation – this can degrade instrument’s performance. Rectifying this in a silicon foundry may not be a major issue, but it certainly is in a telescope operating in the EUV that is attached to a satellite or onboard a space station. Compounding all these issues, non-standard processing procedures are needed to make ultraviolet silicon based imagers. III-Ns strengths Turning to a wide bandgap semiconductor based on nitride alloys promises to improve the design and performance of EUV detectors. The binary compound GaN has already been used to make flame detectors operating in furnaces that can detect signals in the presence of hot backgrounds with no saturation, contrary to devices based on other technologies. These detectors are commonly referred to as ‘visible blind’, and have a cutoff wavelength – the upper limit of absorption – of 365 nm. Far shorter cut-off wavelengths are possible by increasing the aluminum content in AlGaN. Take this to the extreme, AlN, and the bandgap hits 6.2 eV, translating to the highest detectable wavelength of only 200 nm, which is in the vacuum ultraviolet range. If an aluminum composition of at least 40 percent is applied, it’s possible to fabricate what is known as a ‘solar blind’ detector. This device is completely insensitive to solar radiation reaching the Earth’s surface, because the ozone layer in our atmosphere absorbs all radiation below 280 nm, the cutoff wavelength for Al0.4Ga0.6N. Thanks to this complete absence of sensitivity above a certain UV wavelength, imagers and detectors can be built with far fewer filters that offer superior detectivity, due to a reduction in background signal. Degradation under high doses of UV radiation is also diminished by switching from silicon to AlGaN. This equips the detecting systems with greater robustness and better long-term stability. Additionally, AlGaN-based imagers do not require cooling, facilitating the instrument design. Efforts to develop GaN photodetectors for various applications have been going on for a couple of decades, with single-pixel photodetectors receiving the most attention. Some work has also been directed at the development of two-dimensional imagers, which are challenging to make, because the imager has to be integrated with the readout circuit. Uniting these two is not easy because the nitride active layers come with unwanted baggage – the substrate that provides a platform for their growth. ‘Face-up’ or ‘face-down’? One option for integrating nitride layers and the read-out circuit is a ‘face-up’ approach, which requires the fabrication of through-wafer-vias to contact the readout through the substrate. This is possible for relatively large pixel-to-pixel pitches and thin substrates, but at the cost of a decreasing fill-factor - a part of the active pixel has to be used for interconnection. Complicating matters, fabricating vias with a high aspect ratio is tough in silicon, and would be even more challenging with the sapphire and SiC substrates that are commonly used for AlGaN heteroepitaxy. A ‘face-down’ approach also has its downsides. First and foremost, you have to work in the backside illumination configuration. Several groups, who all have faced challenges associated with substrate absorption, have used this geometry. Building a detector of low-energy ultraviolet radiation is relatively easy, because sapphire is transparent in this spectral range. But this material is opaque in the EUV regime. If silicon is used for AlGaN epitaxy, backside illumination is impossible, so the substrate has to be completely removed. Do this, and you are left holding an epitaxial stack that is less than a micron thick. Handling this without damaging it is tricky, but even so, this is still the most promising way to make AlGaN detectors. At Imec, which is based in Leuven, Belgium, we have adopted this approach for an EUV imager that has been developed in the framework of the Blind to Optical Light Detectors (BOLD) project from the European Space Agency (ESA). The concept was established together with CRHEA-CNRS, based in Valbonne, France, responsible for the AlGaN epitaxy. Another partner in the project was Royal Observatory of Belgium, based in Brussels, working on system specifications. The project’s primary goal is fabrication of a EUV imaging instrument that can be placed onboard the Solar Orbiter spacecraft and used to study the Sun’s atmosphere. Technical specifications for the imager are very challenging: It should provide EUV images not only with a 10 μm pixel-to-pixel pitch, but also with a rejection ratio of visible radiation of several orders of magnitude. To meet these goals, we and our project partners have agreed on a design that involves an AlGaN-on-silicon active layer on which the 256x256 pixel focal plane array (FPA) is fabricated (see Figures 2 and 3). A flip-chip bonding technique integrates the AlGaN 2D pixel array with the custom-design readout, manufactured in a commercial CMOS technology. Fig. 2. Exploded schematic of the hybrid imager, with AlGaN-on-silicon detector chip (top) integrated by flip-chip bonding with the CMOS readout (bottom) using 10 μm pixel-to-pixel indium solder bumps. Image not to scale Fig. 3. AlGaN Focal Plane Array with 10 μm pitch: a) schematic showing consequent processing steps together; b) array fragment after MESA etching and deposition of the ohmic and Schottky contacts; c) array after indium bumps deposition, with one bump per one Schottky diode pixel; and d) the same array seen from the backside after silicon substrate removal (through the optically transparent AlGaN layer), showing all levels from a) from the other side of the wafer Indium bumps are used as high-density interconnects. These must have good uniformity to ensure reliable connection. Meeting these criteria is tough, because the distance between two adjacent bumps must be less than half-pitch, which equates to 5 μm, and that the height of the bump must be at least 3 μm. Another challenge is removing the silicon substrate. This is done using an SF6-based, inductively coupled plasma reactive-ion etching, which is highly selective to the AlN at the interface of the epitaxial layer and the substrate. A submicron membrane of the active material that is supported by an array of indium bumps is left after this step has been completed. There are several critical requirements for this AlGaN layer: It must be thin enough for optimum performance in the EUV range, where the penetration depth is very small; it must be completely free from cracks; and it must contain as few defects as possible, to minimize the recombination losses after carrier generation. To avoid curling of the AlGaN layer due to relaxation and the resulting debonding of the detector chip, a 1-mm-wide frame of silicon is left around the active area, which is supported by an array of dummy interconnects (see the shape of the silicon substrate in Fig. 2). Integrated chips are then encapsulated in a package, wire-bonded and characterized in any facility equipped with the EUV radiation source. Spanning the UV Relatively simple systems can be used to perform measurements under ultraviolet illumination – commercially available lamps in combination with filter wheels or with a monochromator serve the purpose perfectly. However, characterization below 200 nm requires a much more complicated, vacuum configuration, because high-energy radiation is strongly absorbed in air. Our imagers were characterized with the radiation from a synchrotron located at PTB/BESSY II in Berlin, Germany. Using a dedicated readout system with vacuum feedthroughs, the imagers were placed in the beamline and measured after aligning the beam to the active area of the detector chip. Images from a 100x25 pixels region were analyzed as a representative part of the 256x256 array (see figures 4 and 5). Both these samples had a silicon substrate grid patterned on top of the entire AlGaN active layer, apart from the 1-mm-wide frame around it. Thanks to this approach, it is possible to have a fixed pattern in the image and to explore different postprocessing schemes. Fig. 4. A 100x25 pixel fragment of the array after integration and substrate removal, showing the AlGaN layer (yellow) exposed to radiation with the silicon substrate frame around (brown). The response pattern under illumination with wavelengths below the cut-off wavelength (280 nm) corresponds to the shape of the substrate opening. Sensitivity down to the wavelength of 13 nm is demonstrated with the synchrotron radiation Fig. 5 A 50x25 pixel fragment of the array after integration and substrate removal, showing the AlGaN layer (yellow) exposed to radiation with the silicon substrate frame around (brown). The response pattern under illumination with wavelengths below the cut-off wavelength (280 nm) corresponds to the shape of the substrate opening. Sensitivity down to the wavelength of 1 nm is demonstrated with the synchrotron radiation. At such a low wavelength silicon becomes transparent, which is visible as more pixels are activated at the edges of the opening, where the Si substrate is thinner Samples had a cut-off wavelength of 280 nm, due to the deployment of an Al0.4Ga0.6N active layer. At longer wavelengths no response was registered, and at lower wavelengths the pixels produced a response corresponding to the area in the substrate opening. Since the substrate is not transparent, it acts as a shadow mask, allowing better distinction of the photogenerated signal. Reducing the wavelength even further, it’s possible to obtain images at the Lyman-α line (121.6 nm) and in the EUV range – the target spectral span for our device. Excitation of our devices with radiation as short as 13 nm produces a response, which is very promising for applications in the EUV lithography. It is worth noting that at this excitation wavelength more pixels at the edge of the opening show response. This is because at the edge the silicon substrate is thin enough to allow penetration of the high-energy photons. Even though not all the pixels in these examples are operational, this experimental data provides a proof-of-concept for this imager: It is possible to demonstrate a solar-blind response in a twodimensional array with 10 μm pixel-to pixel pitch. Process investigation and optimization is ongoing, and the next goals are to obtain a uniform response and good fabrication repeatability. The results from the first batches of demonstrators reveal many possible improvements. Concerning applications, these AlGaN imagers promise to serve many areas outside the original EUV solar observation. Not only could they have an impact in EUV lithography; they may also serve other scientific applications requiring long-term stability and intrinsic blindness to visible and infrared light. © 2011 Angel Business Communications. Permission required. Further reading P.E. Malinowski et al. IEEE Electr. Device Lett. 30 1308 (2009). F. Barkusky et al. Rev. Sci. Instrum. 80 093102 (2009). P.E. Malinowski et al. IEDM 14.5 (2010).