New processing concepts needed for short-run minifabs*
12/01/2000
Rethinking the production process should reduce the cost of a minifab by about 50%, raising the return on investment from a mini-100-lot/month line enough to match that of a high-volume fab running 1000 lots/month. This will help chipmakers turn a profit on short runs of system chips, but could also attract a crowd of new competitors now able to afford the modest investment to make their own specialty chips.
The future of the semiconductor industry is in system chips for digital consumer electronics gear. And that means short runs of many specialized products, requiring minifabs running only a tenth the volume of the current standard 1000-lots/month plants — but without sacrificing return on investment. This new paradigm will require completely rethinking the production process.
Specialized systems chips for digital consumer electronics gear will soon replace general purpose chips for computers as the driver for the semiconductor industry. All these electronic games, wireless phones, digital cameras, digital video players, and digital video cameras will need their own different combinations of audio and video, logic and memory. The short product cycles of consumer goods, from six months to a year, mean production of these chips must be ramped up fast and changed over quickly, but in small product runs of a few hundred thousand units at a time. And the consumer market absolutely demands the lowest possible prices.
Current semiconductor production lines, geared to stable high-volume production of a standard product, are completely unsuited for doing these sorts of small runs and quick product changeovers at low cost as demanded by the consumer market. A totally new approach is needed, a new kind of minifab that can do low-volume production at a profit. This minifab will have to produce at low volume as efficiently as current plants do at high volume, and be able to change over to making a new product quickly, by focusing on reducing production time as much as possible.
The big problem with making money on small-volume production is the wasted capacity of all that expensive equipment. The smaller line still has to do the same number of production steps, and still needs a machine for each. And most equipment on the market comes with capacity to run 400-500 lots/month, not 100. Building a line to run 10% of the volume takes 20% of the investment of a 1000-lot/month line, cutting return on investment 50%.
But we figure that rethinking the production process can reduce cost of the minifab by about half, bringing its return on investment up to match that for the standard high-volume plant. Making this low-volume production practical will require the following seven steps:
1. The first key to making practical minifabs is to standardize production processes as much as possible, so the same equipment can be used for more process steps. If, say, microprocessors and SRAMs and DRAMs all used the same shallow trench isolation and gate insulation film processes, they could all be made with the same equipment. Each of the different processes for making SiO2 and Si3N4 films could be standardized to run on the same equipment. The more processes that can be done with the same equipment, the better the utilization rate. If each unit could combine at least four or five, or even better, eight or 10 different processes into one standard recipe, the minifab could make use of the full 400-500 lot/month capacity of the equipment.
2. Second, equipment makers need to make multifunctional units that can each handle several different process steps to further improve utilization and bring down costs. Most likely candidates for combination are the diffusion, CVD, and reactive ion-etching processes. A single machine could do both oxidation and annealing and perhaps even nitridization, or could both remove oxidation and make the polysilicon layer, or could both make and remove the antireflective film or the hard mask. CVD equipment in particular could be made to handle several different process steps.
3. Third, the equipment for the minifab should be made in modules for easy maintenance and upgrading. With so few pieces of equipment in a minifab, and such frequent product changeovers, the equipment's capabilities need to be easy to change. By putting each function in a separate module, performance could be upgraded or capacity increased just by replacing a module. It's especially important to make the gas supply in CVD and RIE equipment in modules for easy replacement.
4. Fourth, it will be necessary to rethink production technology for low-volume production to bring down costs as much as possible. Chipmakers and equipment makers will have to work together to reconsider the best ways to do low-volume production, balancing reducing the cost of ownership against reducing the raw process time. Low-cost coating with new low-k solgels looks like it may replace plasma CVD processes. And the best approach will probably be to use direct-write e-beam lithography instead of increasingly costly masks.
 |
5. Fifth, the key to the profitable minifab is to reduce raw process time, especially by focusing on reducing the transition time between processes. We recommend the use of process and in situ monitors. These can reduce the time needed for quality control and for checking initial conditions. We also recommend continuous process manufacturing through multiple steps, such as the diffusion and CVD processes. Our research has found that more than half of production steps can be combined into continuous processes, greatly reducing raw process time, and also reducing the number of machines needed (see table). Continuous processing also cuts production time and labor, reducing the time spent moving the wafers around, and the time the equipment has to wait.
6. Sixth, production equipment must be made more reliable. With so few pieces of equipment in a minifab, there is no redundancy, so if one unit has a problem or even needs regularly scheduled maintenance, the entire line stops. Instead, the minifab will need to use monitoring systems to reduce malfunctions, and to rely on easily replaceable modules to make repairs and maintenance fast. Regular short maintenance adjustments between runs may replace longer, regularly scheduled maintenance work. It will be especially necessary to reduce the dry-cleaning time of RIE and CVD equipment, and simplify recalibration after cleaning.
7. Finally, plant management must focus on reducing production time not just by trimming raw process time, but also by reducing cleanroom steps between processing, and by better handling of the many changeovers between different products. With the minifab running so few lots, batch processing of multiple lots at once is not efficient.
But processing one lot at a time can increase cost of ownership. Again continuous processing looks like the best solution.
Our simulation of low-volume production of a six-layer, 0.13µm microprocessor using the approach outlined above suggests a minifab could produce 100 lots/month with the same return on investment as a standard high-volume line making 1000 lots/month (see figure on p. 66). The new approach to organizing small lot production could cut investment for a 100-lot advanced system chip line down to <1/10 that of a 1000-lot line, bringing it down to under $100 million.
Such low-cost, low-volume, minifab production would bring a major shakeup to the semiconductor industry. Companies that make computers and consumer electronics gear would be able to start making their own system chips instead of buying them from chipmakers. And companies producing everything from automobiles to precision equipment to robots could consider making their own custom semiconductors as well. With investment in chip production more affordable, venture businesses with new ideas could start making their own system chips, too.
Traditional chipmakers would find it easier to manage their plants more efficiently. Instead of having to make small runs of system chips on a high-volume line, they'd be able to use one of their many separate minifabs, simply turning on production at more minifabs in order to ramp up output for any particular product.
Semiconductor equipment makers, however, could see sales to their traditional customers drop by as much as half. Their sales to all the new entrants to the system chip business, however, would probably increase more than enough to make up for the decline. And since the new entrants have no chipmaking experience of their own, equipment suppliers would be able to sell their expertise as well, marketing their process recipes along with their equipment in a higher value-added package. — Katsuya Okumura, Process and Manufacturing Engineering Center, Toshiba Corp. Semiconductor Co.
*This story was translated for SST from the July 2000 issue of Nikkei Microdevices, our partner in Japan.