Comparing GDS-II Files and Mask Sets in IC Design & Manufacturing

GDS-II (Graphic Database System II) and mask sets are two related but distinct concepts in the semiconductor design and manufacturing flow. GDS-II refers to the digital layout files used at the end of the design phase, whereas a mask set refers to the physical photomasks used in the fabrication phase. Below is a detailed overview of each, their roles in the IC workflow, their key differences, and how a GDS-II file is transformed into a mask set.

GDS-II Files: Definition and Role in Design

A GDS-II file is a binary database file that stores the complete physical layout of an integrated circuit. It is the de facto standard format for exchanging IC layout data in electronic design automation (EDA) . A GDS-II file encodes all the geometric patterns (polygons, paths, etc.), text labels, and other layout information in a hierarchical manner (cells within cells) representing the chip’s layers. This format allows efficient sharing and transferring of layout artwork between different tools and teams, and ultimately serves as the source for creating photomasks.

In practice, the GDS-II file is the final output of the IC design process (often called the tape-out database). After all design checks are passed, the layout data is exported to GDS-II and delivered to the semiconductor foundry . In other words, the GDS-II file represents the completed chip design ready for manufacturing. Foundries accept this file as the blueprint from which they will prepare the physical masks and fabricate the wafers. GDS-II files are handed off at tape-out and contain the intended patterns for every layer of the chip, but in an idealized form (without manufacturing enhancements applied).

Mask Sets: Definition and Role in Fabrication

A mask set is the collection of all photomasks needed to fabricate a given IC design. A photomask (or simply “mask”) is a quartz or glass plate with an opaque pattern (typically chrome) that corresponds to a single layer of the chip . Each mask acts like a stencil that projects the layer’s pattern onto the wafer during photolithography. In modern processes, dozens of masks are used in sequence – one mask for each fabrication step/layer (e.g., active areas, polysilicon, contacts, metal layers, etc.). Together, these multiple masks form a mask set that collectively contains the entire IC layout one layer at a time.

For any given chip, the mask set might consist of anywhere from a handful to dozens of individual masks, depending on process complexity. In fact, a single device may require between 5 to 40 or more individual photomasks (one per layer), which all together constitute the mask set . Advanced nodes and multipatterning techniques can drive the number of masks even higher (for example, a 10 nm process might use 70+ masks) . The mask set is used on the fab floor: during wafer fabrication, each mask is loaded into a lithography scanner and used to expose its pattern onto the photoresist-coated wafers, thereby transferring the circuit patterns layer by layer . In summary, the mask set is the physical realization of the IC layout, enabling the designed patterns from the GDS-II file to be printed on silicon.

Transformation from GDS-II to Mask Set (MDP and RET)

After tape-out, the foundry must convert the GDS-II data into actual photomasks through a series of postprocessing steps. This overall procedure is called Mask Data Preparation (MDP), which translates the designer’s layout file into the detailed instructions and patterns that mask-making equipment (pattern generators) use to write the masks. In essence, MDP bridges the gap between design and manufacturing by taking an idealized layout and creating manufacturable mask data from it. Key steps in transforming a GDS-II file into a mask set include:

1. Data Import and Verification: The foundry receives the GDS-II (or OASIS) layout file and performs final checks. The GDS-II data is read into the mask tooling software to ensure it matches the intended design and that no errors were introduced in transfer. This step confirms the design layers will be correctly mapped to the masks (often guided by a mask order form).
2. Chip Finishing: Certain structures may be added or adjusted to improve manufacturability before mask generation. This can include inserting a seal ring around the chip periphery, adding fill patterns (dummy metal or polysilicon shapes that ensure uniform density), and other custom features required by the foundry for yield or process control. These additions do not change the circuit functionality but help the chip fabricate correctly.
3. Reticle Composition: The mask writer arranges one or more chip layouts onto the area of a mask reticle. If the die is small, the same die pattern might be repeated in an array on one mask (to maximize efficiency in wafer exposure). Alignment marks, identification text, and test patterns are also added to the mask layout at this stage . The result is a reticle layout for each layer, which includes all instances of the die and necessary alignment features on that mask.
4. Resolution Enhancement (RET): Before mask writing, the layout data is modified with resolution enhancement techniques to improve lithography fidelity. One common RET method is Optical Proximity Correction (OPC), which adjusts the shapes on the mask (e.g., adding small “serifs” or bumps, hammerheads, and biasing line widths) so that the patterns will print on the wafer as intended despite optical distortions. More advanced RET can include inverse lithography technology (ILT) or adding sub-resolution assist features (SRAFs) and phase shifting elements, all aimed at achieving better pattern transfer for deep-submicron features. These modifications are applied to the mask data only – the original GDS-II remains unaltered, serving as the design intent reference.
5. Mask Fracturing: The complex polygons and curves in the corrected layout must be translated into a format that the mask writing hardware can handle. Fracturing is the process of breaking down the geometry into simpler shapes (typically rectangles or trapezoids) that the electron-beam mask writer will draw. For example, a circular or diagonal shape in GDS might be approximated by many small trapezoids in the mask file. This step often results in a huge volume of data because hierarchical repetition is lost – every pattern instance on the mask is explicitly written out. (Special care is taken to manage data volume so that mask write times remain reasonable.) The output of fracturing is a set of mask writer files (often in a proprietary format like MEBES or JEOL formats) containing all the low-level shapes to be exposed on each mask.
6. Mask Fabrication: Using the fractured mask data, the photomask manufacturer creates the physical masks. A blank mask (quartz plate with an opaque chrome layer) is loaded into a pattern generator (usually a laser or e-beam writer). The machine uses the prepared data to expose the pattern onto the mask’s resist. The mask is then developed and etched, removing chrome in the exposed areas to leave the intended transparent vs. opaque pattern. After etching, the mask pattern matches the post-RET layout for that layer. Each mask is inspected for defects and may undergo repairs if necessary. Finally, a protective pellicle (thin transparent film) is mounted over the mask to keep it clean during wafer exposures. The complete set of masks is thus manufactured and ready for the wafer fabrication process.

Through these steps, the GDS-II layout database is transformed into the physical mask set. The MDP process ensures that by the time the patterns are on the masks, they are not only an accurate rendition of the original design but also optimized for the limitations of the lithography process. The end result is that the wafer, when exposed with these masks, will reproduce the intended IC patterns as closely as possible to the original design.

Key Differences Between GDS-II Files and Mask Sets

While a GDS-II file and a mask set are directly related (the mask set is created from the GDS-II data), they differ in format, usage, and stage in the chip lifecycle. Below are some key differences:

• Nature and Format: A GDS-II is a digital file (binary data format) used for IC layout representation. It is software-readable and meant for data exchange between design tools. In contrast, a mask set consists of physical artifacts – the actual quartz/chrome plates for each layer (or the corresponding mask pattern files for mask writers). The mask data is typically in a machine-specific format (not human-readable) and ultimately realized as physical plates used in lithography.
• Content and Structure: The GDS-II file contains the entire chip layout in one package, usually organized hierarchically (with repeated sub-circuits defined once and instanced many times). All layers of the design are present in the GDS-II database. A mask set, however, is organized by layer: each mask corresponds to one layer of the process. Thus, the mask set breaks the unified design into separate pieces – one mask per layer. The patterns on masks are generally flat geometries (post-fracturing) with no hierarchical info, since the mask writer must explicitly draw every feature.
• Stage in Lifecycle: GDS-II is at the design sign-off stage (tape-out). It is the output of the design cycle and is handed over to manufacturing. The mask set is at the manufacturing stage – it is produced after tape-out and used during wafer fabrication (in photolithography steps). In other words, GDS-II lives in the realm of EDA and CAD tools, whereas masks live in the cleanroom and fab equipment.
• Use Cases/Purpose: Designers and EDA tools use GDS-II files for final layout verification, IP exchange, and as the contract between design house and foundry. Its purpose is to accurately describe what needs to be manufactured. The mask set’s purpose is to physically implement that description: masks are used in scanners/steppers to project the patterns onto silicon wafers. Without the mask set, the patterns in the GDS-II file cannot be transferred to the wafer – the mask set is the tooling that makes fabrication of the IC possible.
• Modifications and Fidelity: The GDS-II layout is generally the ideal pattern drawn by designers (or by automated place-and-route tools) – it does not inherently include manufacturing distortions or corrections. By contrast, the patterns on the masks include resolution enhancements and adjustments. For example, masks will have OPC features and biasing that are not present in the original GDS-II data. These mask modifications ensure that the silicon result matches the intended shapes from the GDS. Thus, the mask set reflects a modified version of the design geometries tailored for the lithography process, whereas the GDS-II reflects the as-designed geometries.
• Number of Items: Typically, there is one GDS-II file per chip design (occasionally one per chip version or configuration). In contrast, a mask set comprises many individual masks for that design – one per layer, often numbering in the dozens for modern chips. All masks together correspond to the single GDS-II database (after processing). Managing one GDS file vs. handling 20– 70 masks is a significant practical difference in terms of data management and logistics.
• Data Volume: The GDS-II file (with hierarchy) is relatively compact for what it represents, and can be processed by design tools. The mask writing data, after fracturing and RET, is much larger (often orders of magnitude larger in file size) and is handled by mask writers and lithography simulation tools. This explosion in data is due to the flattening of repeated structures and addition of fine correction features. Consequently, mask data requires specialized processing and storage, unlike the more design-friendly GDS-II format.

The table below summarizes these differences between GDS-II files and mask sets:

Summary Table: GDS-II vs. Mask Set