A new definition of
03/01/2004
Until recently, mask data preparation (MDP; including fracturing) was considered a commodity service conducted after tape-out that facilitated the transition of the design data into a maskwriting format for mask manufacturing. MDP typically required only a day or two, but as circuit dimensions and design rules shrink, the role of MDP is changing significantly.
The International Technology Roadmap for Semiconductors (ITRS) predicts that flat data volumes will continue to grow exponentially. In test cases, file sizes >54Gb have already been encountered when output to a flat format. For the 130nm node, file sizes up to 64Gb are predicted. The main reason for file size explosion is growing design complexity, e.g., a 512MB DRAM chip has more than 109 transistors. Such a design realized in a 100nm process requires ~100mm2 of chip area.
In systems-on-chip (SoC), standard logic components are combined with analog and high-frequency blocks to create more complex and powerful devices. The mix of components differing in design style, density, hierarchy, and processing requirements results in enlarged chip sizes. These designs have led to microprocessor chips that are as large as stepper field sizes (~400mm2).
But while the push for more aggressive design rules continues to accelerate, the wavelength of the exposure tools in lithography shrinks at a much slower pace. The corresponding decline in k1 factor indicates that photolithography is approaching the theoretical limit of resolution. To help resolve subwavelength pattern features and achieve necessary feature sizes within a reasonable process window, resolution-enhancement techniques (RET) are required, but add to the data's growing size and complexity.
Historically, MDP and fracturing are conducted in tape-out groups, which send the data to the mask house with the expectation it will not be altered. Tighter requirements on critical dimension (CD) and registration, however, have forced maskmakers to deploy additional enhancement techniques (e.g., geometry processing).
Analysis of the common data flow
Typically, a series of steps is conducted in the common data flow between tape-out and mask manufacturing (Fig. 1). Classical physical verification is performed, including design rule check (DRC) and layout vs. schematic (LVS). If required, manufacturing aids like metal fill and slots are introduced on the appropriate layers. The foundry applies RET, such as scatter bars with OPC, phase-shift mask (PSM) with OPC, or simple OPC. Next is a "fracturing" of the design features into maskwriter and inspection formats that are passed on to the mask shop. At each interface between tools in the flow, copies of the original data are generated, stored, and exchanged in differing formats, resulting in a large number of data representations of a single design, each growing in data volume.
In complex algorithms, iterations may be required to achieve final data configuration, forcing a refracturing of data along the flow. A side effect of complexity can be a partial or full flattening of data. Flat processing is time-consuming and contributes to larger file sizes that can create a serious data bottleneck.
A different approach to fracturing
Maskwriting machines require data in a dedicated format. The most commonly used raster scan machines expect the data to be organized in stripes and segments with all rectangles and polygons broken apart; that is, fractured into elementary rectangles and trapezoids. Projecting the functions currently embedded in the fracture command onto the prior steps in the flow reveals a significant overlap with functions conducted during verification, RET, and introduction of manufacturability aids. By exploiting this overlap, a benefit can be derived by separating the geometry-processing steps from the fracturing functions.
This new approach defines fracturing as an export step at the very end of the data flow, avoiding iterative fracturing and postponing the flattening (if required) to the last possible moment. The original, common flow is reconfigured to include the revised fracture definition, producing a new distribution that decouples the format requirements of the maskmaking equipment from geometry processing. Any data flattening required will occur at the very end of the flow. Design data hierarchy is preserved, speeding up processing and saving precious time.
A new flow for mask data preparation
While hierarchical processing can help the individual functions in a complex flow to perform faster, the whole flow should be considered with respect to data exchange, file handling, and storage. Geometry-processing functions extracted from the fracture step can be absorbed by steps that already exist as geometry-processing sequences. A new scheme can be proposed by extending this approach throughout the flow (Fig. 2).
This new approach advances a solution integrating all functions that manipulate or operate on geometries into one tool, which facilitates efficient application of hierarchical processing algorithms, reducing processing or data-handling overhead. It also allows the generation of hierarchical output data that can be fed to advanced maskmaking equipment capable of direct input from hierarchical database formats (e.g., GDSII). Otherwise, data can be exported into maskwriter formats when all processing is concluded. Ideally, this flow has only two data representations: original tape-out and final data set passed on to the mask house. If the need arises to split the flow between groups or organizations, the exchange could be conducted based on an open hierarchical format such as GDSII.
Conclusion
Revision of the post-tape-out data-preparation task flow to incorporate efficient hierarchical processing tools integrates geometry processing with other flow steps, and enables a single-tool post-tape-out process that reduces tool interfaces and minimizes data-management overhead. Design-independent optimization of the design hierarchy for shape manipulation enables dramatic improvement in TAT. Further enhancements can be achieved by improving the current exchange format (GDSII). Preliminary experiments have shown that a more efficient structure can reduce the file size of hierarchical databases up to a factor of 17 and flat fractured databases up to a factor of 2.
For more information, contact Steffen Schulze, Mentor Graphics Corp., ph 503/685-7000, e-mail [email protected].