EUV lithography technology is attracting keen interest. With mass production commencing around 2011, it promises to make possible 64- to 256-Gbit NAND Flash memory, microprocessors with 1.6 to 6.4 billion transistors, and more.
Nikkei Electronics Asia -- October 2007
"We want it right now!" said a lithography engineer at Toshiba Corp of Japan working in NAND Flash memory manufacture. Extreme ultraviolet (EUV) lithography technology is attracting keen interest from semiconductor engineers today. Long referred to as the ultimate lithography technology, it has never entered commercial production - but finally it seems to be on the way. Considering the current state of development, it could enter partial use as early as 2010, and many people in the industry believe it will be on the volume production line in 2011.
EUV technology would make it possible to achieve the integration levels needed for NAND Flash memory of 64- to 256-Gbit capacities, and microprocessors with 1.6 to 6.4 billion transistors. This is four to 16 times higher than the integration levels of cutting-edge integrated circuits (IC) today (Fig 1).
In the last few years, many semiconductor engineers have felt that the design rule would have trouble getting below about 30nm, given the limits of existing photolith technology. The immersion ArF photolith used on volume production lines now can't make patterns much smaller than 40nm even when modified to achieve half-pitch performance. The term "half-pitch" refers to reducing the space between parallel pattern lines to half the normal width, so that lines and spaces are of identical widths.
Immersion lithography boosts the numerical aperture (NA) by injecting pure water between the lithography lens and the wafer, increasing resolution. The max NA with pure water is 1.35, and the resolution limit about 40nm. Resolution could be further increased by using a material with a higher index of refraction than water, but this is not viewed as very practical currently. Double-exposure lithography, which splits the process for a single layer into two steps, will also increase resolution, but cost rises accordingly. In response to the apparent slowdown in resolution shrink, there is active development of ICs using 3D structures.
EUV lithography stands an excellent chance of achieving a breakthrough. If EUV lithography can be made practical, it will mean half-pitch performance between 10nm and 30nm, which is equivalent to manufacturing technology between 10nm and 30nm generations. If EUV lithography is perfected on-schedule, it will make it possible to preserve Moore's Law for at least this generation, as far as circuit size is concerned.
EUV lithography makes it possible to shrink chip geometry because of the extremely short wavelength used. In general, the shorter the exposure wavelength the higher the resolution. A look back at the last 30 years shows that wavelength has grown steadily shorter, from 436nm to 365nm, 248nm, and 193nm. EUV lithography would drop this down to only 13.5nm.
Makers Get Serious
EUV lithography has suddenly become much more likely because concrete results for actual lithography are beginning to be released. The three major stepper/scanner manufacturers - ASML Netherlands BV of the Netherlands, Canon Inc of Japan, and Nikon Corp of Japan - have all announced results paving the way for commercial use. ASML seems to be in the lead, successfully forming dense 32nm contact holes at half-pitch in February 2007, using the prototype Alpha Demo Tool (ADT). Most steppers begin as experimental systems with small fields, evolving into prototypes (alpha), evaluation models (beta) and finally production models. The ADT is the world's first alpha model, and currently the design closest to volume production capability. ASML plans to ship the beta model in 2009. Japanese manufacturers are close on their heels, though, with Canon skipping an alpha model shipment to ship beta in 2009. Nikon has already shipped the EUV1 alpha system, and plans to release the beta EUV2 to the market at the end of 2009.
Data are beginning to appear that indicate development projects by Japanese manufacturers are on schedule. In May 2007 Canon used its Small Field Exposure Tool (SFET) to form a half-pitch 26nm pattern (Fig 2). Exposure was handled by Semiconductor Leading Edge Technologies Inc (Selete) of Japan, on research consigned by New Energy & Industrial Technology Development Organization (NEDO) of Japan. This was the first successful 26nm pattern mixing dense lines and arcs. There are a number of systems like the SFET, but none has reached this resolution.
Following the SFET, Selete has also installed Nikon EUV1 alpha system. The first exposure resolutions are scheduled to be available in September 2009, and presented at the 2007 International EUVL Symposium on EUV lithography to be held in Sapporo this month (October).
Cost Stability
The biggest worry of the semiconductor manufacturers when it comes to EUV lithography is climbing cost, which would result from price hikes in manufacturing equipment, masks and more. Manufacturing equipment manufacturers, however, say there won't be any major price increases, and it might even be possible for them to hold down price rises in the future.
According to estimates by Extreme Ultraviolet Lithography System Development Association (EUVA) of Japan, which is developing EUV lithography technology, EUV lithography offers the lowest per-wafer cost of all for 32nm-generation ICs. Other methods require two exposure steps to achieve the required resolution, and the single-step design of EUV lithography is the primary reason for the lower cost. In two-step exposure two masks are needed to expose each layer, which also means twice as many resist develop processes are needed.
An EUV lithography system is not cheap, at between Yen6 billion and Yen7 billion, but this is not such a leap from the Yen4 billion to Yen5 billion price tag of existing immersion ArF systems. Any attempt to use the low-resolution immersion ArF systems at high resolutions would result in complex processing. EUVA estimates cite this as a major reason for the high cost of immersion ArF.
It is apparent in the intensity distribution of light on the wafer during exposure (Fig 3). When both systems are used to form half-pitch 45nm contact patterns, the patterns are distinct with EUV exposure, but imagery is blurred with immersion ArF. This blurriness can be fixed, but only with an immense number of correction patterns on the mask, which boosts mask cost.
Issues to be Resolved
There are still a number of problems with EUV lithography that must be resolved before it can be used in commercial production. The key difference is that now the way to solve most of them is known.
At the International EUVL Symposium every year a vote is taken to determine the technical difficulty of unresolved issues. A look at the latest results, 2006, shows that the "most difficult" problem is the light source, followed in order by resist in second place, mask issues in third and fourth, and the optics in fifth (Fig 4).
Using CO2 Lasers
The light source was cited as the most difficult problem, but a method has been developed that promises the requisite output in volume production.
Light source output is a crucial element in determining EUV lithography processing speed. The higher the output, the less time it takes to expose the resist, which boosts stepper/scanner throughput and lowers cost. Volume production of ICs is generally said to demand at least 100 wafers/hour performance, which works out to 115W EUV light source output for a resist sensitivity of 5mJ/cm2, and 180W for 10mJ/cm2
There are several EUV light sources, but a dramatic improvement in output has been achieved with a CO2 laser in the laser-producer plasma (LPP) design that uses laser light illumination of a source to create plasma, and extracts EUV light from the plasma. Output had been several W until February 2007, when Cymer Inc of the US announced 25W output, and EUVA announced 40W. At Semicon West 2007 in July, Cymer reported 50W output. The firm plans to build a 100W system, close to commercial level, before the end of 2007, and ship 100W EUV light sources to ASML and other manufacturers in 2008.
With LPP, an increase in laser output fundamentally means an increase in EUV light. Output has been low until recently because of flying debris from the plasma, especially with the Sn plasma that has such a high EUV light emission efficiency.
EUVA experimentally demonstrated that a CO2 laser would reduce Sn plasma debris. CO2 lasers have longer wavelengths than the usual YAG lasers, reflecting in low-density regions close to the plasma surface instead of penetrating to the plasma core. As a result, the molten Sn in the core is not splashed significantly by the laser light.
Rotating Electrode Design
Another type of EUV light source uses the discharge-produced plasma (DPP) method, creating plasma through an arc between electrodes. DPP designs are thought capable of 115W even with present-day technology.
The biggest obstacle to boosting the output of the DPP light source is that the electrodes meld under the heat of discharge. Rotating electrodes, with excellent thermal radiation performance, are being evaluated as a way to avoid the problem. The light source from Philips Extreme UV of the Netherlands, used in ASML's alpha system, is of this type. A rotating circular electrode discharges from one point, and then that point has one full revolution to cool off until it discharges again.
Xtreme Technologies GmbH of Germany and EUVA are working on a modified rotating electrode design in combination with a laser. Technology development has only just begun, but there is good reason to think high output will be achieved. Xtreme is aiming for 100W output in 2009 through its tie-up with EUVA. The group illuminates molten Sn droplets with the laser to create a gas concentration optimal for EUV light generation before discharge. The conventional rotating electrode design uses the laser to start discharge, and the plasma gasifies the source after discharge starts, making it difficult to optimize gas concentration.
The DPP light source also has a problem with debris, for which reason an opaque foil trap is positioned between plasma and the confocal mirror for Sn deposition.
Tweaking Molecule Size
The resist needed will satisfy three conditions simultaneously: 22nm to 32nm resolution half-pitch, sensitivity of 5mJ/cm2 to 15mJ/cm2, and line width roughness (LWR) of 1.2nm to 1.7nm (3). LWR is an index of variation in line width due to irregularities in resist walls. Companies like Fujifilm Corp of Japan and Tokyo Ohka Kogyo Co Ltd of Japan have resists offering 25nm resolution, but none that also fulfills the sensitivity, LWR and other requirements simultaneously. Resist performance will be boosted in the future primarily through two methods - namely, improving existing polymer resists and utilizing new low-molecular resists. Polymer resists have been in volume production for a long time now, and their mature technology makes them most likely to succeed. Fujifilm, for example, has improved its polymer resist for electron beams to work with EUV, achieving 25nm resolution (Fig 5). The polymer molecules are large, however, which can often result in surface irregularities after developing, making it difficult to reduce LWR.
Low-molecular resist, on the other hand, has small molecules likely to assist in reducing LWR. Material strength, heat resistance and other properties have been thought insufficient in the past, but they are improving.
