How Advanced Are We?

John Baliga, Associate Editor -- Semiconductor International, 7/1/1999

The semiconductor industry keeps advancing its technology at an increasing rate. The technology nodes given in the 1997 National Technology Roadmap for Semiconductors occur on a faster schedule than shown in the 1994 Roadmap. The 1998 ITRS update has them coming even faster, and Motorola went well ahead of the pace when it made devices with 100 nm gates earlier this year.

The industry's roadmapping activity looks 15 years down the road, and updates come out at least once every two years. Clearly, companies in the industry are keeping an eye on the future, and this will help keep upcoming trends from surprising them. They must also be flexible enough to handle big changes because quantum leaps in technology can happen at any time (for example, see 'Gas Dome System Provides Unity-k Dielectric,' p. 125).

As for flexibility on the fab floor, fabs do look to buy equipment that will last two or three technology generations. In other words, they find tools that are flexible. There are a lot of advantages to this tool-focused approach, but when it comes to productivity and profitability, processing technology is not the critical path. Large advantages can be gained by implementing advanced process control (APC) across the whole fab. I have to wonder if semiconductor manufacturers and their suppliers have the flexibility to make that move.

APC has two main categories: run-to-run (R2R) control, and fault detection and classification (FDC). Run-to-run control helps keep wafer-state parameters, like film thicknesses and CD sizes, more tighly controlled to their nominal values. This helps tighten performance parameters more closely to their optimum values. Fault detection and classification predicts when a tool is going to malfunction and identifies corrective action. Avoiding tool crashes with FDC can save a lot of money in scrap wafers.

APC and its principles have been implemented to a limited degree, but APC will have to be used on a much wider basis in the near future (see 'Advanced Process Control: Soon to be a Must,' p. 76). The main barrier to implementing APC probably will be deciding who owns the controller. The 1997 NTRS Roadmap discussion on sensor driven model-based integrated manufacturing (SDMBIM) states that the tool suppliers need to own them, and they eventually will. The question is whether they will have them ready in time, and what to do if they don't.

There are some who think the tool suppliers are behind the curve by not having APC capabilities built into their tools right now. Fab engineers have been implementing their own APC. How will suppliers migrate toward supplying APC when fab engineers have become used to taking and analyzing tool data? Will the tool suppliers have to convert internal tool data into generic parameters to maintain propriety while cooperating with an external controller? When fab engineers hook sensors up to tools and something goes wrong, who pays?

Controlling processes more closely to nominal will be a must, as will avoiding scrap. Advanced process control on a fabwide basis is the way to do both. There will be hurdles to clear and fights to fight to make this happen. With all the benefit that is just waiting to be had, clearing these hurdles and fighting the fights will be well worth the trouble.