Latest from DAC: Xyalis research charts the gradual merging of design, mask making technologies

Once the interface between the IC back-end design team and the mask shop was very simple. GDS-II files went to the mask shop, and if there were no errors in the data format, masks came back. But the complexities of sub-wavelength lithography have changed all that forever. Now the level of bidirectional data traffic between the design team and the mask makers is increasing on several levels, to the point that it is more accurately a conversation rather than a transmission. The mask shop is asking for more data on design intent, and providing more feedback to the design team.

A research project jointly conducted by Xyalis, ST Microelectronics, Atmel, and Topan is now examining just how those deeper links might work in practice. As Eric Beisser, CEO of Xyalis, explains it, the program goal is in effect to create the equivalent of design-for-manufacturing tools not for the chip manufacturing process, but for the wafer manufacturing process. "We believe that there are issues that complicate or slow down mask production that have their roots in the chip design flow," Beisser said. "These issues may not impact the final silicon at all, but because they give the mask shop trouble, they do impact the schedule. We believe that some new constraints on the back-end design process can potentially reduce mask cycle time." Beisser said that there appear to be similar research projects going on within Synopsys in the USA, and between other companies in Japan.

Initially this may just represent a data flow from the mask shop back to the mask data preparation step. But eventually—and this is the subject of some longer-range study that Xyalis is doing with Atmel—the data may percolate all the way back into the chip design flow. Just as it is possible to make changes to the physical design that do not alter the function of the chip but improve manufacturability, it may be possible to make non-functional changes that simplify the mask data preparation, mask-making, and mask inspection processes.

One easy example is to link the flat fractured mask data base back to the hierarchical data structures that define the chip design to the design team. This would allow a design team to mark certain design features—such as dummy metal or the corporate logo—as non-critical areas for lithography rule checking, phase-shift feature generation and OPC decoration. This is just a matter of preserving a bit of design knowledge that already exists in the chip designers’ minds, but that is lost by the time the data gets to the mask shop. Yet knowing these details could save considerable unnecessary work for the mask makers.

As the links between back-end design, mask data preparations and mask-making grow more bidirectional, the importance of formal transfer methods such as SEMI P10—and the importance of supporting them with commercial tools—will grow, Beisser stated. At some point, he suggested, we would not be thinking of this as a file transfer process at all. "All of this is really data in one relational database," he said. And so in principle the needs of all three groups can be met by queries against a common data set.