News, views and contacts from the global Semiconductor industry

Maskless Lithography


1 September 2004


The growing cost of lithography masks is raising concerns for future technology generations. In response, semiconductor manufacturers are exploring maskless lithography as a possible solution. Walt Trybula, senior fellow of SEMATECH, assesses how far the technology has come and where the gaps are.


The increasing cost of lithography masks is raising concerns for future technology generations. For example, a mask set for the 65nm device generation is projected to cost $3 million. As a result, semiconductor manufacturers are beginning to look for ways to reduce the need for masks - maskless lithography is one such approach.

In a sense, one variation of maskless lithography is already in use – Electron Beam Direct Write (EBDW) has been used for many years to develop new devices and circuit prototypes. Charged Particle Maskless Lithography (CP-ML2) is a derivative of EBDW that promises much higher throughput, and Optical (or Photonic) Maskless Lithography (O-ML2 or P-ML2) employs light and programmable optics to create circuit patterns. But what is the potential for these applications in semiconductor manufacturing, and what is the current state of their development?

EBDW LITHOGRAPHY

Early EBDW systems derive from scanning electron microscopy and electron beam mask writer tools and are used to create much finer patterns than were possible using conventional optical projection lithography. EBDW systems expose an electron-sensitive resist using a vector- or raster-scanned beam of electrons. The areas exposed to the electron beam undergo a chemical reaction that allows the resist to be developed to create the desired circuit patterns.

"A mask set for the 65nm device generation is projected to cost $3 million."

The typical equipment configuration consists of a vertical e-beam column in a vacuum chamber with the electron source at the top and the exposure working area below. The working area contains a stage for accurately positioning and moving the wafer.

The main advantages of EBDW are that it is capable of a high resolution (single-digit nanometer structures have been fabricated) and can work with numerous resist materials. Because it is a maskless technology, a small number of prototypes can be produced without the time and expense involved in fabricating photomasks.

However, compared with optical lithography tools, EBDW tools are slow, expensive and require significant maintenance. The key throughput detractor in the case of EBDW arises from the fact that these systems deliver individual pixels, compared with the terapixel delivery of projection tools. Thus, full 300mm wafer writing is measured in hours per wafer, not the wafers per hour rate required for high-volume manufacturing. An additional disadvantage of building prototypes with EBDW is that a new process must be developed and qualified using optical lithography tools and resists for volume manufacturing of the final product (some work has been done to develop multi-functional e-beam / optical to provide an easier transition to volume manufacturing.)

MASKLESS LITHOGRAPHY

Maskless Lithography (ML2) is used to describe technologies for creating circuit features without the use of a mask, but at significantly higher throughput than EBDW. CP-ML2 and O-ML2 (or P-ML2) are the two major technologies. In order to be commercially viable for volume semiconductor manufacturing, these technologies will need to achieve throughput rates of at least 20 wafers per hour, and perhaps as high as 60 wafers per hour. Thus, the key question for any maskless lithography approach is whether it will be capable of achieving such throughput at reasonable cost, and at the feature sizes of interest for current and future device generations.

It should be noted that some current maskless lithography development efforts are targeting prototype tools with throughput capability in the region of five wafers per hour. Although these approaches may be extendable to higher throughputs, it is also possible that a five wafer per hour tool could be used for device prototyping and very low-volume manufacturing applications.

OPTICAL MASKLESS LITHOGRAPHY

Typical O-ML2 approaches are based on conventional optical lithography scanner architectures, but with the significant difference that the photomask is replaced with an addressable array of light modulating elements that are used to generate the mask pattern in real-time. Micro-mechanical Spatial Light Modulators (SLMs) are examples of such modulators. These can be of several basic types – pistons, tilting mirrors, or sinusoidal height modulation – and they can be diffractive or specular. They all have different operational principles, but their common function is to produce an image to be projected on the wafer.

In high-end performance systems, the optical maskless systems are based on piston or tilting micro-mirror SLMs. Similar performance can be achieved with both mirror types. Micronic laser systems have favoured tilting mirrors, which are reportedly less complex to manufacture, enable the use of larger mirrors and less complex rasterising algorithms, and make fewer demands on the data path. Consequently, a tilting mirror arrangement might be more appealing for high-performance, high-capacity and eco-nomical optical maskless lithography.

CHARGED PARTICLE MASKLESS LITHOGRAPHY

"Electron Beam Direct Write (EBDW) has been used for many years to develop new devices and circuit prototypes."

Charged particle maskless lithography includes cell projection, mini-column and parallel arrays (field emitters, blanker array, Negative Electron Affinity [NEA] cathodes or proximal probes). EBDW and CP-ML2 (Cell Projection) were used by some integrated device manufacturers in the 1980s and 1990s, but were replaced by optical lithography for patterning device masks. Optical lithography has been predominant since then.

With the cost of a 90nm mask set near $900,000 and much higher figures being projected for the 65nm features, making small-volume or customised products using optical masked lithography tools is becoming unaffordable.

CP-ML2 may offer less expensive lithography for custom applications and for low-volume products. CP-ML2 tools may hold promise for more rapid prototyping and for reducing the risk of fabricating expensive masks to prove processes and designs. CP-ML2 may also be a solution for manufacturing ICs that are not required in high volume, significantly lowering the cost of lithography.

MIX AND MATCH

Another approach being used in prototyping is coupling a maskless lithography system with projection or other lithography tools to expose a single level. This configuration, which can result in significantly higher throughputs than a maskless system by itself, is typically called mix and match lithography.

"Making small-volume or customised products using optical masked lithography tools is becoming unaffordable."

Currently, no tool manufacturers offer a mix and match system. Instead, manufacturers that employ this technique physically move wafers from one type of machine to another. The development of an e-beam and 193nm or 248nm resist that would allow one material to be used for both exposures would enhance this technique. It would also provide the mechanism to develop prototypes with e-beam exposure and transfer it into production with optical exposures with the same resist formulations. There is an ongoing development effort in this area, with potential for higher throughput.

ARE ML2 SYSTEMS FAST MASK WRITERS?

Some see a fast ML2 system as a panacea: the same system could become a fast mask writer, resulting in lower mask costs. If the system can write a wafer with 100 or more images, then creating a single image for a mask should be easy and quick. The lower masks costs would in turn obviate the development of an ML2 system for production. However, this is not the case.

The area being written on the wafer is small compared with the area on the mask. The mask requires image control over the entire 6in by 6in area, while the wafer writer only needs to maintain control over an area roughly 2cm by 2cm. Moreover, the mask requires more image data than direct writing because the mask needs assist features. However, a fast ML2 system could create the templates for nano-imprint, since these images are identical to the image written on the wafer and do not require all the assist features of an optical mask. Thus, it would be possible to have a fast nano-imprint template writer.