|
Electron beam lithography has been used in the production of integrated circuits for decades. However, due to the
limitation of throughput it was not a viable solution for high volume manufacturing and its biggest application is the
production of semiconductor masks. For many considerations it has particularly now become desirable to eliminate the
semiconductor mask and introduce maskless lithography for semiconductor fabrication. Multiple Electron Beam
Maskless Lithography (MEBML2) has been proposed as a solution to overcome the traditional source current limitation
of an electron beam system by using many thousands of parallel electron beamlets to write a pattern directly on the
wafer.
In developing the MEBML2 tool the challenges have shifted and, in absence of the mask, the system data path has
emerged as one of the central challenges. The main theme in the data path development is bandwidth. The required raw
bandwidth at the patterning beams is determined by throughput and resolution, i.e. pixel size and number of intensity
modulation levels. To achieve a production worthy throughput at 10 wafers per hour in a Gaussian-beam-based maskless
lithography system, by writing 3.5-nm pixels at 2 levels (on/off) which is required for the 22-nm lithography node, the
required aggregate bandwidth at the beam blanker array is up to 45 Tbit/s. Such a large bandwidth requirement means
that the data path architecture is mainly characterized by the bandwidth of the data streams in the system. Compression
techniques can be used to reduce the intermediate data stream bandwidth requirements and consequently lead to
simplifying the system design, reducing power consumption and footprint, but come at the cost of increased data
processing complexity and possible limitations on throughput.
In this paper we will show results from the development of a prototype data path for the Gaussian-beam-based maskless
lithography system. A new concept for data processing and storage is proposed. The vertex-based processing and storage
technique is adopted to reduce memory usage considerably, with only modest requirements on the hardware resources. It
reveals that a realistically implementable data path system for the maskless lithography tool in high volume
manufacturing is feasible.
|
