Side-stepping the wavelength barrier

17 Jun 2002

New equipment for building silicon chips will soon be available, but problems with the optics and laser systems still exist. Nick Flaherty finds that these are becoming more difficult to solve with the increasing complexity of the chips and the wavelengths of light that are required to make ever smaller devices.

From Opto & Laser Europe April 2001

The next generation of silicon chips with feature sizes of 0.13 µm will be manufactured in bulk using optics that operate at 193 nm. Equipment makers are currently shipping the systems that can build such devices. Now, the race is on to develop equipment to make chips with feature sizes of less than 0.10 µm. However, this will require a move to 157 nm systems where a number of key challenges still remain.

To meet the rise in demand for chips, the industry is also having to look at what it will do beyond 157 nm. The answer isn't clear, so the problem is being addressed by an international consortium, which is based at Sematech in the US.

Malcolm Gower, chairman and chief technical officer of Exitech in Oxford, in the UK, said: "We have shipped a 157 nm microstepper tool [to Sematech], but not all of the optics problems have been solved. However, most of the difficulties have been understood and solutions look possible."The stepper is used to shine light though a mask onto a silicon wafer for processing. The mask is "stepped" over the wafer and illuminated, producing multiple images of the mask. Each layer is processed using different chemicals and mask patterns to build up the transistors that make up each silicon chip on the wafer.

"A production stepper has a longer lifetime than a microstepper, so the demands on the optics are more severe," said Gower. "Its imaging performance is better than in the optics and equipment that we have been selling. However, the system has shown that from an optics and laser subsystems point of view there are no major show stoppers.

"A big problem is contamination, such as water vapour and trace amounts of hydrocarbons, in the beam lines. Ultraviolet light dissociates the molecules, which can then be deposited on the optics that attenuate the beam. We use purge gases in the lines to remove contaminants, but purging is not a trivial task because the concentration of oxygen is about 1 ppm and that of the hydrocarbons is even less."

Another challenge is to make reflective coatings that have a reasonably long lifetime and transmissive materials that don't degrade or build up colour centres. "You can't use fused silica as a transmissive material for optics any more, so you have to move to calcium fluoride. The problem is getting good-quality material," said Gower.

One of the early issues was manufacturing 157 nm masks. Mask producers have been working with silica doped with fluorine to produce a material that is sufficiently transmissive, even if it isn't good enough for the optics of the system.

The problem with calcium fluoride, which is needed for 157 nm equipment, is that the majority of the supply is currently going into the 193 nm generation devices, such as the PAS 5500/1100 stepper from Dutch equipment maker ASML.This system is aimed at building the 100 nm generation of chips and combines a 0.75 numerical aperture (NA) lens - the industry's largest commercial NA - with an exposure capacity of 90 (200 mm) wafers per hour at a dose of 20 mJ/cm². It uses Carl Zeiss's Starlith 1100 lens and a 2 kHz, 10 W laser with a bandwidth of 0.35 pm.

ASML entered the 193 nm lithography-equipment market two years ago when it introduced its PAS 5500/900 tool, but this move to commercial 193 nm systems is eating up the source of calcium fluoride for lenses. "The supply is scarce for large lenses," said Gower at Exitech. "It's fine for the smaller lenses that we use, but not for the larger production systems."

Laser manufacturers have been trying to get round the supply problem. The 193 nm technology has significant consequences for optics design and laser parameters, says US laser manufacturer Cymer. These are accelerated, non-reversible damage of coatings and SiO₂ optics - which represent a large cost concern in a production environment - and a higher dispersion at 193 nm, which requires the lasers to operate at lower bandwidths or use chromatic correction on the projection optics.

While optical-material manufacturers work to reduce the absorption, Cymer has been developing ArF sources that have a stretched-pulse duration to get round the problem. The first production-generation ArF-excimer lasers operated at 1 kHz and 5 W with pulses of more than 30 ns. Experiments have shown that the pulse duration can be extended to 60 ns.

As new generations of ArF lasers are developed with higher repetition rates, pulse-stretching techniques will become more common. Currently, the fluence limit for ArF laser radiation through fused silica has been set to 0.1 mJ/cm² in compliance with a 10 year optics life target to be achieved in a production system.

Laser-system maker Lambda Physik in Göttingen, Germany, has also developed an ultra-line-narrowed litho laser for 193 nm systems, which is due to ship this quarter.

The company has already dispatched a 193 nm, 40 W, 10 mJ ArF laser, the 4020, for imaging sub-100 nm linewidths. This has a bandwidth of less than 25 pm and the 4005 is intended to improve on that performance with a spectral bandwidth of less than 0.35 pm. This will depend on lens designs that have an NA of 0.78.Cymer has also just started shipping its first ELX-6500F₂ fluorine laser for 157 nm systems. "As integrated circuits continue to shrink, the ability of semiconductor manufacturers to image the circuitry associated with the devices becomes increasingly critical," said Pascal Didier, Cymer's president and chief operating officer.

"Lithography light sources are the key to the ongoing shrink of the devices and to chip makers maintaining their competitive edge. Cymer is enabling the industry to move to the next generation of lithography at 193 nm, 157 nm and beyond."

But where does the industry go from here? For years the next step has been seen as 126 nm. However, there are a number of fundamental problems to overcome.

The 126 nm technology currently suffers from not having an optical material that is transparent without being birefringing, such as MgF₂. Therefore, 126 nm technology will need to rely on all reflective designs that are limited to NAs of 0.5 or less via modelling. The cost of developing manufacturable systems with a limited life span is high for 157 nm and prohibitive for 126 nm technology.

As a result, the next-generation lithography technologies are being developed. Out of the three most popular options, Exitech is working on an extreme-ultraviolet (EUV) source that uses a 13.7 nm radiation device rather than the alternative Scalpel electron beam and ion-beam systems.

EUV is a reflective scheme that uses mirrors rather than lenses and masks, and this is key. If there has to be a move to a reflective rather than a transmissive system for 126 nm anyway, then moving to EUV would provide headroom to move to future generations of manufacturing technology.

One of the challenges here is the radiation source. At the moment the cleanest source comes from a laser-produced plasma in a gas jet, says Gower. "But it is not very bright - not enough for a stepper that has to process 80 wafers per hour."Other possible sources are capillary discharge devices, which are not very bright, or synchrotrons, which have been moving more and more into the X-ray region rather than the EUV range. "In a world without politics the synchrotron has some technical advantages, but the initial cost is high so the synchrotron makers have fallen off the lithography roadmap," said Gower.

One of the first companies to sell low-power EUV lamps is AIXUV in Aachen, Germany. The firm was set up last year as a spin-off from the Fraunhofer Institute for Laser Technology and backed by Lambda Physik (OLE December 2000 p6).

AIXUV is looking at radiation sources that are in the 10 to 15 nm range, with production planned for this spring. It is using a plasma that is generated by a 10 kA current in a tube with a patented discharge geometry. This results in the creation of a zone that is 0.5 mm in diameter and a few millimetres in length, from which the plasma emits radiation at 100 pulses per second.

However, it is still not known how much longer 157 nm systems will be used for the production of smaller and smaller chips, because new design techniques and other technologies can be used to push back the need for more aggressive lithography. "Lithographers have a habit of getting cleverer all the time," said Gower. "With phase shift masks, our 157 nm stepper has been used for sub-100 nm linewidths."Scientists at the Massachusetts Institute of Technology, in the US, have built transistors that have gate lengths of 25 nm using existing light-based lithography. They used a silicon-on-insulator substrate, patented phase-shifting technology from Numerical Technologies and deep-ultraviolet (DUV) 248 nm optical lithography.

"The fabrication of 25 nm gates is a dramatic step forward from our announcement in February that 50 nm transistors had been manufactured using our technology," said Buno Pati, CEO of Numerical Technologies. "With Numerical Technologies' phase shifting, optical lithography will no longer be the limiting factor in advancing integrated circuits."

Other researchers have reported results for devices with physical gate widths of less than 50 nm at the International Electronic Devices conference, with some at 20 nm, and all of them used existing technologies.

Perhaps the most promising work is that of R Chau et al. of Intel in Hillsboro, Oregon, in the US, which produced a high-performance CMOS transistor with a gate width of 30 nm, compared with 0.13 µm - which is state of the art and just starting production. The 30 nm device was built with conventional 248 nm processing, not even using the leading-edge 193 nm technology that is being used for 0.13 µm systems.

In the meantime, researchers at Bell Labs in New Jersey, in the US, have come up with a scheme to use the existing optical techniques and materials for devices with 25 nm gate widths. The self-aligned local-channel V-gate with optical lithography uses only current production tools with 248 nm KrF lithography.

These are all clear pointers that the move to the smaller geometries does not necessarily push lithography past the point of no return. New circuit techniques could well propel the need for 13.7 nm EUV systems further into the future, and optics has still got a long way to go.