Are Design Rules Broken?
Are Design Rules Broken?
It is no mystery that the number of design rules has exploded over the past few technology nodes. It’s impossible for any human designer to “remember” them all, much less follow them all. It’s also a problem for the CAD engineer. We extracted some data from a spectrum of DRC decks that our customers have in production and the graph below shows the results.
The rule count is increasing, but maybe more surprising the Calibre operation count has gone up even faster. This increase in operation count is due to the fact that the rules themselves are more complicated. Despite all these rules with all this complexity, we are hearing from customers that design issues are slipping through the net and causing problems in manufacturing.
I believe that two primary issues are driving this problem. The first is a very practical issue of overspecification. There are so many rules with so much interaction that no one can even tape out DRC clean any more. We have heard from a large number of our customers that they no longer even expect to tape out DRC clean. Essentially, it is so complicated they can’t deal with the situation. They have developed informal systems for requesting waivers from the foundry. It seems like some kind of heresy to even suggest that you might consider taping out with DRC errors and even more heresy that the foundry would say ok. However, it happens because of the second bigger issue which is that the rules themselves don’t work in the first place.
In today’s advanced technology nodes, the robustness of a particular layout feature can no longer be determined accurately by the traditional single dimensional pass/fail constructs that design rules and design rule checkers have depended on. For example, how far apart should two metal 1 shapes be? In the “good old days” then answer would have been some minimum dimension X. However, today the answer would be best summed up with two words; “It depends”. Is it a line end? How wide are the metal shapes? what is the concurrent run length of the adjacent shapes? See the example drawing below. As you can tell, the answer is becoming multi-dimensional and sometimes X is ok, sometimes X is tool small causing yield loss and sometimes X is too big causing loss of area compaction.
The result is that the foundries attempt to compensate by creating 100’s of separate single dimensional rules to cover all the corner cases. This, in turn, creates severe over constraint that the designer can’t meet, and since the foundry knows they have overconstrained the problem they are prone to giving in to waiver requests. The simplicity of single dimensional DRC rules forces us into following the “letter of the law” without capturing the “spirit of the law”. It starts to feel like we let a room full of lawyers into the design center.
The question for you is do you agree or am I over reacting? Have you encountered these issues? What types of things have you seen? What do you think is wrong? How much time are you wasting on rules or violations that you end up waiving anyway? Have you had a design pass DRC and fail in manufacturing for some layout feature? How much do you think your layout area is being impacted by these “bad” rules?
Should we stage a strike or lauch rockets?
More Blog Posts
- Battle of Fins and BOXes
- TSMC 28nm yield (SemiWiki)
- DAC 2011 is upon us!
- Mentor Graphics User to User (U2U)
- Gate Oxide Breakdown Failures Highlight Industry Need for New Electrical Rule Checking Tools
- Dawn at the OASIS
- Layout Density and the Analog Cell
- Effects of Inception
- On-line session covering the DAC presentation for Calibre xACT 3D
- You can't give stuff away fast enough
- December, 2012
- March, 2012
- May, 2011
- April, 2011
- February, 2011
- January, 2011
- November, 2010
- August, 2010
- June, 2010
- May, 2010
- April, 2010
- March, 2010
- February, 2010
- January, 2010
- December, 2009
- November, 2009
- October, 2009
- September, 2009
- August, 2009
- July, 2009
- June, 2009
- "Waive" of the Future?
- How do you debug LVS?
- DFM for Non-PhD's: Part 2 - Reliability
- Mixed-Signal SoC Verification
- Process Variation: The Use of In-Die Variation
- DFM for Non-PhDs
- Calibre Everywhere -- the customer value of universal integration
- So, why not just write better rules?
- To be the man, you've gotta beat the man!
- Power in need, Power indeed
- May, 2009