DFM: What’s real now?
10/01/2007
A fter nearly half a decade, design for manufacturing (DFM) has not provided the expected value to erstwhile users, in spite of the perceived need and vaunted advantages. For most DFM start-ups, life is still difficult; few have achieved profitability or widespread adoption. However, there’s hope.
To properly comprehend the DFM dilemma, we have to view it from two sides: DFM is design-centric because it improves yield and minimizes manufacturing errors by bringing manufacturing information up into chip design flow. Meanwhile, DFM is manufacturing-centric because it brings design intent down to the silicon manufacturing flow. DFM is actually thriving-providing value, achieving profit, garnering users and getting into manufacturing as well as design flows.
First and foremost, when we talk about the value of DFM, we have to realize that the ultimate goal is to get silicon out with acceptable (or improved) yield. Right now, those DFM tools that directly address the limitations of lithography equipment and make subwavelength manufacturing possible (i.e., OPC and PSM) provide the most value. Long before designers started their 65nm projects (with or without DFM tools), foundries and IDMs were already hard at work developing 45nm and even 32nm processes. At these nodes, there is a considerable need to address newly identified manufacturing issues in the patterning loop, among other challenges.
Even though foundries and IDMs are reluctant to turn over proprietary process data to DFM-tool developers, they are willing to utilize manufacturing-targeted DFM tools by themselves. Large IDM-employed engineers would relish seeing design intent pushed into the manufacturing process. Unfortunately, given what we’ve already seen, the poor understanding of design intent has started to seriously hurt yield on the manufacturing side as we get down to 65nm and 45nm. Manufacturing-intended DFM can prevent systematic yield loss, which has been becoming dominant. One real benefit accounts for the quicker adoption of manufacturing-intended DFM: it’s much easier to measure improvement of yield loss in the fab than on the design side. Given the huge cost to build and operate a fab line, the value of DFM tools is easier to recognize and justifies revenue to DFM venders.
The EDA industry has already demonstrated great success in the adoption of OPC/RET application and flow methodology. Therefore, great market opportunities do exist for some DFM startups that provide DFM solutions that push design intent into manufacturing. These DFM tools can offer initial revenue streams to the startups until DFM tools take hold.
However, the DFM toolset has great near-term potential . Design closure on timing and power for advanced technology nodes is already a challenge, consuming huge amounts of a designer’s time. Individual users are being mandated by their design constraints to get closure with these tools as the first priority. At 65nm, no foundry has made design-intended DFM tool usage mandatory at design tapeout. Below 65nm, we may see widespread adoption of DFM tools.
Designers will migrate to DFM-oriented flows only when they can’t get to design and manufacturing closure. Even designers (or their managers) who see the need for DFM don’t want their flows to be upset with radical changes to the methodology. Of course, these design flows come from the large, full-line EDA vendors. Interfacing with these flows without adding much application complexity for designers is a challenge for DFM tool developers.
So what’s the crux of this issue? In my mind, it’s whether the design-centric or manufacturing-centric approach will bridge the gap between IC design and manufacturing that slows the improvement of chip yield. My answer is that they both will. What’s most important? The industry needs tools that can communicate design intent and large amounts of data (in an appropriate format) among the design, OPC, mask, and wafer manufacturing steps. Right now, such infrastructure changes on the manufacturing side are manageable.
Make no mistake: design-centric DFM will become a crucial tool for improving yield. Layout patterns are originated by designers. The necessary close DFM alliances and partnerships between DFM start-ups, major EDA venders, semiconductor manufacturers, and equipment venders are already forming. But for now, most DFM value comes from the manufacturing-centric approach, which has already yielded results and enabled advanced technology nodes.
 Chenmin Hu, president Anchor Semiconductor Inc. |
Contact Chenmin Hu at Anchor Semiconductor Inc., 5403 Betsy Ross Dr., Santa Clara, CA US 95051; ph 408/986-8969.