Advertiser and

Buyer's Guide
Buyers Guide

Chip Shots blog

Greatest Hits of 2005
Greatest Hits of 2005

Featured Series
Featured Series

Web Sightings

Media Kit

Comments? Suggestions? Send us your feedback.


X-Architecture Update

Diagonal design to yield
production chips in 2005

X is no longer an unknown quantity for backers of a collaborative effort to develop a workable interconnect chipmaking process with diagonal routing. Participants in the X Initiative endeavor have spent the past four years trying to answer the manufacturing, cost, and yield questions about the technology, which uses strategic diagonal interconnect routing as its focus.

REAL X: An example of X architecture layout (left) alongside a real-world Toshiba digital TV chip using the diagonal design, a device Toshiba says was 11% fatser, with less logic area, than a comparable IC.

And 2005, they insist, is the year that the responses will free the concept from its test phase to deliver actual product. In 2004 Toshiba, an initiative cosponsor, used the technology on the fourth- and fifth-metal layers to produce a 180-MHz chip at the 130-nm process node. Designed for digital television, that device was 11% faster and had a 10% smaller logic area than a comparable IC. In a major announcement on April 11, UMC said it is the first pure-play foundry to qualify design rules for chips based on X architecture at the 90-nm process node. The Taiwan-based foundry is prepared to process IC designs using the architecture.

X architecture bypasses the semiconductor industry's traditional search for improvements in processes and materials to focus on metal layers four and five of multilevel chip architecture.

The X design retains the common Manhattan-style 90° angles on layers one through three. Although it's not a new concept, the technology always lacked the proper industry infrastructure until Simplex Solutions, a software and design service company that is now part of Cadence Design Systems (San Jose), teamed with Toshiba four years ago to start the ball rolling.

Placing the interconnects at a 45° angle on the two uppermost metal layers reduces device wiring by more than 20% because wires can be sent in eight different directions. Device performance shows increases of more than 10%, proponents of the technology say. In addition, the diagonal routing produces 30% fewer vias and 30% more semiconductors per wafer for system-on-chip (SOC) devices.

Both process and timing yield will show improvements, according to supporters. A study conducted in mid-2004 by PDF Solutions, an early initiative supporter, used yield prediction software to determine all the critical areas for opens and shorts in X architecture. A second finding determined that overall hot spots and the intensity of hot spots also decreased.

This information comes from the X Initiative, a consortium of more than 40 companies that was launched in 2001. Most observers intuitively understand that diagonal wiring "makes sense," insists Ketan Joshi, director of marketing for the initiative. But the technique has always raised questions. "Why aren't people doing this? Is there something wrong with it. Can you actually manufacture this?"

Cadence Design Systems and Toshiba cosponsored the X Initiative, which set out to establish a "design chain" in order to prove the worthiness of the technology to the engineering community "in a more rigorous fashion," notes Joshi. When Aki Fujimura, X Initiative steering group member and CTO, new business incubation at Cadence, began laying the initiative's groundwork, "he visited each piece of the supply chain—for example, maskmakers, wafer exposure tools, and so on. Every person told Aki, 'I don't see any issues in my area. I can make it work, but I don't think others can.' At the end of the day we pretty much realized we can do this if all the pieces of the supply chain collaborate."

Getting the collaboration started under the X Initiative umbrella brought a dawning realization to each disparate link in the supply chain. Participants discovered that they had the wrong perceptions about whether diagonal routing's technical issues were surmountable, Joshi says.

In 2003 the consortium confirmed first-silicon design rules. Joshi notes that member companies such as Applied Materials, STMicroelectronics, Infineon, UMC, and TSMC "all have consistently shown that test chips, in terms of design rules, are quite comparable to Manhattan. Obviously, the next step is complete design and production."

Nevertheless, a skeptical Toshiba kept its options open during development of the 130-nm production chip. "They wanted to have an option," Joshi says, "so they ran a shadow project. They had two teams, one on X and one on Manhattan." Toshiba's goal was to design a 162-MHz chip, but a couple of months into the project the manufacturer "reached 180 MHz with X. At that time they were so confident, they canceled the Manhattan project."

Before 2003, Patrick Lin, chief SOC architect at UMC, says it was difficult for the foundry to assess the level of customer interest. In particular, the management needed to determine "from the design side whether the charter of the X Initiative would be accepted by the public or not."

After UMC confirmed customer interest in diagonal routing, the foundry turned its focus to technological matters. "We had to make sure that the architecture was compatible with our existing process," Lin explains. Specifically, UMC wanted to ensure that it would need to make only minor process modifications and then qualify the process. "We had to design the test structure to make sure that the design rules meet our quality standards. So that's where we spent most of our time and effort."

Cadence's design rules target their X architecture designs to UMC's process technologies, Lin points out. UMC did modify specific design rules, particularly "the via that goes outside the metal line. We wanted to see if it always does that or not because of its 45° nature."

On the manufacturing side, he says that a test chip put through stress testing at 225°C revealed "very little degradation in the resistance shift." In addition, UMC wanted to ensure that electrical migration reliability was not a concern. The test chip results enabled the foundry to verify "that our process is quite suitable for the X architecture."

Lin says that UMC has heard from potential customers interested in the X-architecture devices. "We began receiving requests before our 130-nm process became available. They keep on migrating to smaller and smaller geometries, and that's why we set out to do this qualification task." Most of the interest has come from customers in the United States and Japan.

Asked whether the diagonal routing will cause an increase in photomask costs, Lin replies, "As far as I know, there is no extra cost whatsoever. But in preparing the mask itself we've taken care of the OPC (optical proximity correction) issues when we're doing the 45° [routing]. We don't see any problem doing that."

Yields, he insists, "should be comparable with yields of lines routed with the Manhattan architecture. We hope that fabless semiconductor companies will  realize the advantage of this X architecture and start taking advantage of it." Given UMC's broad customer base, Lin predicted that the first type of customers to announce that they are taking advantage of the X architecture will be those that design digital consumer electronics products.

When the X Initiative was launched in 2001, early participants raised concerns that the photomask realm could face challenges unique to X architecture. A chief worry was that mask-writing tools—and vector scan tools in particular—may need longer operating times than raster scanning machines. After four years of work, most of the concerns have been addressed, according to a member of the initiative's steering committee.

Jim Jordan, the Toppan Photomasks representative with the steering group, says that the work with the mask-writing tools has shown encouraging results. In X-mask tests, the company has been working with laser writing tools that use a raster-scan technique. The results have shown equivalent write times.

"We've had no negative write times compared with Manhattan geometries," Jordan explains. He adds that a couple of test runs on vector scan systems by other X Initiative members also have shown equivalent write times. "In theory, there's some impact with vector tools when you get into an angled geometry. Overall, though, the impact of Manhattan to X has not really been substantial."

The steering committee member asserts that the consortium has made the case that X is production ready. "At this point we've pretty much demonstrated manufacturability of 65 nm. In fact, we demonstrated manufacturability through 65 nm, and that really refers to all aspects of maskmaking, whether it's data preparation, writing, inspection, defect inspection and repair, or metrology and design rules. The ability to hit specs has been demonstrated as equivalent also."

Jordan emphasizes that throughputs using advanced masks in the X-architecture tests were three times better than normal. Testing has shown that the costs associated with photomask use have been comparable with non-X processes. "We've found that the production flow and processes for masks, including the major costs for write time and inspection, are equivalent."

Some participating members did have doubts that were dispelled as the X project progressed. Mike Smayling, the CTO of Applied Materials' Maydan Technology Center, admits that there "were always concerns at each step in the design chain. Will the layout tools work properly? Will the verification tools work properly?"

Applied Materials produced the industry's first 65-nm interconnect test chip using X architecture at the Maydan facility in Sunnyvale, CA, in early 2004. Cadence provided the design and validation tools, while Canon provided its ArF lithography system. The device featured multilayer copper/low-k interconnects on 300-mm wafers. Applied used its wafer inspection and metrology tools to measure critical dimensions and defects. It helped that Applied's Etec unit had been involved with the consortium "from a mask-writing standpoint" from the beginning, Smayling notes.

"We thought there would be a low risk at each step," he says. "We thought the tools were in good shape, since we had already tried a design flow at 90 nm, and it worked. So we were fairly confident that our methodology should continue to work, and that fortunately was the case."

Yes, there was a concern about the diagonal routing's impact on process tools. "Certainly, we had some concerns with things like CD-SEMs, the inspection tools and also some of the other tools like our etchers and our CMP systems, which have in situ monitors that could be fooled by some pattern. We wanted to confirm that they would in fact work fine and not be fooled by diagonal wiring."

Smayling and others at the technology center were concerned in particular about inspection tools or in situ monitors set up for Manhattan-style features. "Diffraction patterns or interference patterns coming off of a standard layout would be oriented in a very predictable way, and if you had a detector and sensor set up only for that style of layout, you would completely miss signals coming off of X patterns. It proved not to be the case, due to the system designs with multiple detectors. The inspection tool did not have any particular spatial resonance because of diagonal routing." For metrology, Applied's CD-SEM tools have the software "to allow measuring critical dimensions at 45° [angles].

From an EDA standpoint, Smayling believes specialized tools exist now that "more efficiently handle X-type layouts." He and his colleagues were heartened to see that OPC tools were able to handle diagonal layout correctly, although some had to be fixed.

"We ran across a bug in one of the tools that ended up putting one of the edges on a Manhattan layout instead of a diagonal one. That was fairly quickly resolved by the tool vendor."

Those sorts of solutions appear to be a prominent feature of the X Initiative endeavor. Some members say the consortium is a good example of industry cooperation, where various supply-chain partners work together to figure out all the angles, so to speak.

"It's good news, and any time you see a steady stream of good news it's usually because of some malice aforethought," Toppan's Jordan muses wryly. "There are always little problems along the way, and fortunately with X we were able to get far enough ahead of the design introduction to really smooth the path quite a bit."  — JC

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit

Questions/comments about MICRO Magazine? E-mail us at

© 2007 Tom Cheyney
All rights reserved.