Most of the foundary provide DFM deck and all user use it blindly . Most of the time its increase are and decrease random yield
So what is your question about this?
should we follow DFM deck blindly or do lfd on our design . When shoul we run DFM deck after design complted or in prallel . most of the layout size increse with DFM deck .
what a fabless company should do to get better yield rather than DFM
Hi. With DFM, you get a spread of the results that shows which areas are the worst. So you're right, if you always follow the DFM deck, then your chip will blow up. But if you do analysis on it, you can maybe correct the worst 10%, and get good yield improvement. I think that doing DFM analysis and correction is the only way to get yield improvement on the design side. On the fab side, they work to improve their yield by changing the manufacturing process.
I assume when you say DFM deck that you mean a DRC style checker set to flag recommended rule violations. You are bringing up a big issue with recommended rules. If you flag everything that fails recommended rules you will get thousands or even millions of errors (data overload). You are also right that if you follow the recommeded rules as if they were DRC rules then it will force you to make your chip bigger. The right approach to recommended rules requires a few new capabilities:
Instead of just flagging every place that doesn't meet recommended rule each violation should be "scored" based on where it falls between recommended value and DRC value. The impact of a violation changes non-linearly as you move from DR to RR value. The impact is maximum at DR and approaches zero at RR but half way between DR and RR is not 50% impact.
Each recommended rule has a different relative impact on the design vs another. Each rule should be weighted relatively to each other. Combined with the right within rule model from #1 you can propertly prioritize each violation in relation to each other.
The quality of your design in regards to recommended rules is a statistical roll up of each of the uniquely weighted violations as discribed in #1 & #2. Your objective in regards to improving DFM should be to systematically improve this statistical score over time.
You should strive to fix violations in order of the priority above using available "empty space" in the layout without growing your die size. In other words, no layout is 100% dense so a good DFM design makes the optimal use of the empty space by optimizing as many of the high impacting RR violations as possible using the available space
We have developed a tool called YieldAnalyzer that supports two DFM improvement analysis flows (Critical Area Analysis & Critical Feature Analysis). CAA analyzes your designs sensitivity to random particles and CFA analyzes your designs sensitivity to recommended rule violations using the ideas in #1-4 above. If you want more detailed information feel free to contact your Mentor sales team or come see us a DAC next week.
ps. you mentioned LFD. Litho Friendly Design is a process simulation based tool we provide for looking specifically at litho issues. Some recommended rules do relate to litho issues but they cover other process issues as well. The advantage of LFD is very high accuracy and the potential to catch issues that rule based solutions might miss. The advantage of CFA is high speed analysis of a broad range of DFM issues. Our most advanced customers use the combination of CAA, CFA, LFD and CMP analysis tools to characterize the full breadth of DFM issues at advanced nodes.
Retrieving data ...