Search

Software construction has typically drawn on engineering metaphors like building bridges or cathedrals, which emphasize architecture, specification, central planning, and determinism. Approaches to correctness have drawn on metaphors from mathematics, like formal proofs. However, these approaches have failed to scale to modern software systems, and the problem keeps getting worse.We believe that the time has come to completely re-imagine the creation of complex software, drawing on systems in which behavior is decentralized, self-regulating, non-deterministic, and emergent---like economies.In this paper we describe our vision for, and prelimary work on, the creation of software economies for both open systems and internal corporate development, and our plans to deploy these ideas within one of the largest developer communities at IBM.

The validity of a photolithographic develop process in a manufacturing environment is measured by critical-dimension control, photo defects, chemical usage, cycle time, functional test yield, and rework. Due to the rapid pace of change in the semiconductor industry and the trend toward ever tighter ground rules, the demands that these parameters place on the develop process are never constant. A process that worked in the past may suddenly become inadequate for a new product design with specific performance, throughput and cost requirements. This paper details the techniques used for engineering support of the photocluster single-wafer develop processes in the IBM Microelectronics Division's semiconductor production facility near Burlington, Vermont. Unique analysis procedures have been established for baselining spray- and puddle-develop processes and for identifying the sources of sudden process deviations or defect occurrences. Effective engineering support can maintain the required process performance while also ensuring that all other production considerations are met.

The environmental specifications for submicron photolithographic processes on photocluster tools have traditionally demanded strict setpoint control with very little variation in temperature and humidity. In an effort to reduce support costs, the need for such tight tolerances on relative humidity is being examined. This paper discusses the effects of relative- humidity variation on photoresist thickness after spin-coating and on the final printed-image linewidth after full processing through the photocluster. Two different photoresists are compared using spray- and puddle-develop processing. When the resist thickness is varied by changing the humidity setpoint, the swing-curve correlation between printed-image linewidth and resist thickness is equivalent to one typically generated by changing the photoresist thickness with different casting speeds.