Not a week goes by without a panel, a viewpoint, or an analysis of DFM and the DFM market. This brings to my mind some fundamental questions. If DFM is so necessary, why the debate? If every design team needs DFM why the uncertainty about DFM startups viability? And can we all agree to one definition of DFM?
I have in the past dealt with the last question, but I will repeat my position here for the sake of completeness. We all understand the "For Manufacturing" part but there is confusion as to what "Design" means. The definition ranges from" everything an engineer does to generate a GDS II file", to "the design representation at RTL and above". I like the latter definition, since the steps taken to transform a RTL design into a fabricatable representation fall in the definition of "implementation". The term "design" is too frequently used and, as recently stated by Peter Rabkin of Xilinx, means "everything"(see my DesignCon Morsels blog).
Chenmin Hu, president of Anchor Semiconductor, took another point of view in his recent Viewpoint (see www.designline.com/news/197005159) by stating that one of the goals of DFM is "to feed design intent forward to manufacturing." I like this approach, but his opinion is in the minority, even with his colleagues in DFM companies.
The second question is a matter of discipline, not just in the EDA industry, but in the semiconductor industry. The foundries have taken the easier attitude that yield is a customer problem, and thus the customer should be responsible to use any and all available tools to insure proper yield for the design. This attitude is akin to the old "throw it over the wall" approach to release to manufacturing. Below 100 nm we need a team approach to succeed, and pointing fingers does not build a team.
The reluctance of foundries to make available process information to DFM tools is understandable, but maybe not excusable if this turns out to be the only way to produce working silicon. Under these circumstances, Chenmin's plan makes even more sense since it is almost guaranteed that to transmit design intent one does not require process details. In fact requiring them would complicate the lives of RTL designers so much that the tool would not be useful.
As far as the first question is concerned, I think that some type of DFM has always been necessary. I remember in the early days of ASIC design, highly skilled designers trying to push geometries to the limit and hearing complaints from the foundries that designers were "tampering" with cell libraries. Then as now, my conclusion is that DFM is just as much about design methods as it is about manufacturing constraints. The most important ingredient in a design is "common sense".
There have been attempts to go around manufacturing difficulties in VDSM (Very Deep Sub-Micron): structured ASIC and other cell based design methods are the most common approaches. What is interesting to me is the rise in the use of FPGA for medium size production runs. What this trend says to me is that the higher cost of each die is more than compensated by the lower design cost and faster time-to-market that this solution provides. Altera's HardCopy family of products is an example of what can be achieved when leveraging the tight collaboration between an FPGA house and its foundry partners to provide lower cost non-programmable devices to those who do not need to use a specific process to the limit.
Push the envelope only when absolutely necessary and then invest in all the available tools to make sure you did it safely. But in general designers must remember that, with the usual exceptions, most designs need less area than what is available and thus there is room to play it safe.
My conclusion is that a reasonably small percentage of designers must adopt and use state-of-the-art DFM tools. But the number of companies that need to purchase DFM tools, especially in large quantities is quite limited (certainly less than fifty at the present). This is too small a market size to support all the present DFM companies, and therefore only few will survive: it is just the law of natural selection and we all know that it is not nice to fool Mother Nature.