New tools needed for improving cost-of-ownership
01/01/2006
In its listing of the Top Technical Challenges facing the industry in 2006, Sematech recently included several aspects of metrology and fault detection. Specific areas of concern cited include e-manufacturing, advanced equipment and process control, advanced equipment software testing, short cycle time and short ramp-ups, equipment and fab agility, and standards development. The consortium further noted that its research on the manufacturing front will focus on commercial defect inspection and redetection tools for the 45nm technology node and beyond. A series of factory and equipment-related projects to improve productivity and reduce costs at fabs and foundries were also announced.
While tools and techniques to monitor and control processes to detect faults and improve yields always seem to make the wish list for a new facility, they often get cut from the orders when discussions about controlling spending get into the final stages. This short-term response to the long-term problem of reducing equipment cost-of-ownership (COO) is no longer prudent, given noteworthy advances in process control and fault detection.
Fabs are witnessing an explosion of data, generated by on-tool sensors and in-line metrology. At the same time, traditional statistical process control (SPC) techniques have been overwhelmed by the number of materials in use and the complexity of devices being manufactured. Today, fab managers and process engineers who rely on SPC may be faced with dozens of control charts to check, plus a troubling number of false alarms, because traditional systems cannot categorize abnormal events, let alone evaluate their significance. For example, process parameters for plasma etch tools often tolerate only a <2% deviation margin before they affect yield. Traditional SPC methods to control etch processes have relied on both test wafer and in-line metrology measurements to detect process chamber faults. Not only are the data sometimes unreliable, most of today’s tools don’t offer prescribed fixes to the excursion, requiring the process engineer to ferret out the root cause.
In recent years, fault detection and classification (FDC), an advanced process control (APC) technology with real-time evaluation and response, has emerged as an effective technology for device makers to detect and fix manufacturing problems before they affect fab yield. FDC is considered next-generation SPC; it is a compilation of library of fault patterns that are known to affect the process and only report a fault when a match with the library is detected. This process overcomes the false alarm problem that has overwhelmed traditional SPC. Moreover, the process engineer has to monitor only one index to see if the process is running in a healthy state.
Implicit in Sematech’s list are improved approaches to COO, a model the industry has been fine-tuning for some 15 years. In the early years, tool-purchase decisions were primarily driven by the obvious considerations of the cost of buying and installing equipment; considerations of longer-term factors such as tool uptime and yield were missing from the calculations. That has changed, of course, because fab and foundry owners recognize that, over time, equipment reliability and yield may be more significant considerations than initial costs. In addition, FDC’s potential contribution to improved COO is growing because of the many new materials and extremely tight process parameters moving into general use on semiconductor production lines.
Improving yield is one of the few remaining opportunities for chipmakers and foundries to make a sustained and systematic effort to control costs and improve efficiency. The move to 300mm production and sub-100nm geometry underscores the wisdom of focusing on detect inspection and other productivity techniques, especially given recent estimates that a processed 300mm wafer can be valued at $40,000 or more. A commonly used COO model for fabs and foundries incorporates calculations of tool cost, annual operating cost, annual cost of yield loss, tool lifetime, annual tool throughput, average per-wafer die yield, and tool utilization. For process tools, the model assumes an average COO of nearly $7.50/wafer pass. If a typical fab runs 30,000 wafers/month, with 200 tool passes/wafer, detecting and preventing faults early and often will be the key for unlocking improved CoO.
 |
For more information, contact Brendan Coyle, CEO, Straatum Processware Ltd., Dublin, Ireland; e-mail [email protected].