Time for new thinking on fabs?
03/01/2002
by Robert Haavind
Editor in Chief
 |
Semiconductor manufacturing is a driven industry, continuously updating itself as it chases the elusive rabbit defined by Moore's Law — double the circuitry on a chip every 18 months, and halve the price per function. Competition is what keeps the chase going, not just among companies across the globe but even among nations. Any chipmaker who doesn't keep up will be left behind. With the cost of tools and fabs rising rapidly, it gets tougher and tougher to make the investments required. Today some of the integrated device manufacturers (IDMs) are selling or abandoning fabs that would be very costly to upgrade, choosing instead to go to foundries for the chips they need. In 2001, a record down year, Intel still spent $7.5 billion to keep out front, and even then IBM remains a leader in some of the key chip technologies. In Taiwan, TSMC and UMC, the world's largest foundries, also pressed forward with technology upgrades and a start toward 300mm wafers, not just to attract new business from IDTs and fabless design companies, but also to be ready for the upturn that may be just around the corner.
This incredible industry must constantly reinvent itself on a massive scale. Major changes such as copper interconnects may sweep through the industry over periods of 3-4 years. A process such as chemical mechanical polishing progressed from doubts and even derision to a core technology in a span of a just few years. The concept of subwavelength lithography with reticles enhanced to sculpt the wavefronts of light went from being seemingly impossible to commonplace in less than a decade.
Yet to outsiders trying to introduce new ideas to semiconductor fabs, they sometimes seem like the most rigid factories in the industrial world. This is another, less obvious, part of the nature of the industry. Chipmaking is essentially a chemical process, perhaps the most demanding one of all time, and rigidity is an important aspect of complex chemical processing. Kodak, for example, has factories that stretch for miles in Rochester, NY. Standard photographic film is produced in huge roll-to-roll machines where myriad chemicals are combined and processing steps performed. Managers there tell war stories about hitches in film production that stymied all attempts to fix them until they learned that one errant vendor had "upgraded" or "improved" its chemical plant. As a result, companies such as Kodak demand that their chemical suppliers keep all the steps used to make their products exactly the same.
Changes in large-scale chemical processes do get implemented, but only after exhaustive tool characterization and experimentation, to find the optimum mix of chemicals, temperatures, flow rates, and scores of other key variables. A company such as Intel has a fab specializing in process development, where new process steps can be simulated on computers and then tried out and optimized. Imagine someone with a new tool — or a twist on tool design or metrology — knocking on the door while dozens of potential new process steps are going through this rigorous procedure. Unless what is being offered fits right in with the program, it is unlikely to get much notice. Once all the wrinkles are ironed out, Intel, like some Japanese chipmakers, propagates the new process around its fabs in a "copy exact" fashion, to prevent any slight variations from throwing off the results (even tool spacing and piping layouts are replicated). This approach is aimed at saving a lot of expensive headaches, of which other chemical processors like Kodak have bitter experience.
The large chipmakers look to Sematech (or Selete in Japan, or similar organizations in other nations) to help them probe the potential for future processes and tools. Again, smaller companies and those with something new and different are often left out in the cold. As long as chipmaking proceeds in a predictable way — with Roadmap efforts to define future routes — these orderly development procedures keep efforts on track while saving a lot of duplication and chasing down blind alleys.
But what about disruptive ideas? They have little place in this paradigm, but there may be several reasons why a broader view is becoming necessary. Product life is shortening while development costs rise. As process margins shrink and device models fail to scale smoothly to smaller structures, new thinking may be needed to provide workable solutions.
In Japan, there is recognition that system-on-a-chip (SOC) designs may change the economics of chipmaking, perhaps requiring agile mini-fabs with more flexible, quick-set-up tools that are less costly. Such companies as Tokyo Electron (TEL) and Toshiba have collaborated on designing some multifunction, lower-cost tools that may be suitable for such quick-reaction, smaller-run chipmaking (see Minifab key to low-cost chip production, SST, Feb. 2002, p. 34).
While there is general agreement that 300mm wafer processing will require much more in-line metrology, to tune processes as well as to boost yield, it is not clear where these new capabilities will originate. Metrology designed for laboratory work is too costly to add to every process cell. Some chipmakers — NEC in Japan, for example — are developing their own quick-look metrology for tasks such as checking particles and monitoring etching of contact holes and vias (see AsiaFocus, p. 30). Some observers believe that toolmakers need to do more development to add measurements to their tools (see Letters, "Quick-and-dirty metrology earns kudos," p. 14).
Innovators we talk to say that it is just as hard as ever to get the attention of fabs, or toolmakers, even though the need for major new initiatives may be growing more pressing. Perhaps it is time to be a little less rigid, at least in examining new approaches. Those who take the trouble to check out new ideas now may be very glad they did so a few months from now.