Issue

Enabling yield in automated material handling systems

09/01/2007

The future of the semiconductor industry will depend on the ability of companies to introduce new solutions that improve fab productivity and maximize the return on investment in existing wafer fabs. Companies committed to the 300mm Prime Initiative see a great opportunity to help fabs be more flexible by reducing bottlenecks so wafers can move through fabs faster and more efficiently.

In today’s highly automated fabs, tool companies are developing wireless devices that keep process equipment aligned within very tight parameters to reduce wafer damage, and that make system data available in real time so factory personnel can take immediate corrective action. These devices offer greater consistency during the manufacturing process while significantly reducing human interference and contamination for optimum fab performance.

Causes of particle generation

When wafer transfers occur, the cooperating supports must be as parallel to one another as possible. When they are not parallel, the wafer may slide as the weight of the wafer shifts from the first to the second support. In addition, the transfer coordinates need to coincide with the centers of the two supports, because when wafers are transferred slightly off center, nonuniform processing may occur. Larger transfer coordinate errors may cause the wafer edge to rub on the shoulders of one support or the other, dislodging particles or breaking the wafer. Although transfers may be completed without system-stopping faults, misaligned transfers may lead to intermittent and difficult-to-diagnose faults.

It’s all too easy for tools that were precisely aligned by their manufacturers to become misaligned during maintenance, when modules are replaced or when technicians-operating under tight time constraints or with limited training-neglect to recheck and adjust each critical alignment. In addition, geometric relationships can change over time as parts wear. For instance, SCARA (selective compliant assembly robot arm) robot joints use very stiff rotary bearings to limit vertical compliance while permitting movement in the horizontal plane. When these bearings wear, robot arms droop. End-effectors, too, can twist or sag, especially after repeated heating and cooling.

Finally, when cooperating wafer supports are not parallel, repeated raising and lowering of misaligned transfer pins will cause the wafer center to move away from the intended location. Figure 1 shows this effect. Even when the transfer coordinate is taught using the traditional dowel and hole method, the final position of the wafer after transfer to the chuck can shift due to wafer walk. This phenomenon may explain the common practice of iteratively changing wafer transfer coordinates, processing wafers, and inspecting for process uniformity.

Alignment methods
obot teaching

It is difficult and time-consuming to check and adjust alignment in the fab. These maintenance operations often require the technician to install various jigs, read dial indicators or meters, and directly view the interactions among various parts of the automation system as it moves through its sequence. Fixture installation and removal, coupled with technician intrusions into the ultra-clean tool environment, create contamination that compounds downtime and jeopardizes yields. Direct viewing also risks injury to the technician. In addition, most current alignment methods do not allow for the collection of data that can be trended and used to control the automation process.

Figure 1. Misaligned transfer pins cause wafer centers to move (wafer walk).Click here to enlarge image

Traditionally, inclination has been measured using spirit levels. These devices use a bubble that rises to the top of a cylindrical or curved glass vial. When the edges of the bubble are entirely contained within the lines, the surface supporting the gauge is said to be “level.” This attribute data, however, cannot be used for standard process control, because the inclination can only be recorded as “level” or “not level.”

Circular spirit levels are most commonly used in semiconductor automation leveling since they are small and lightweight. Unfortunately, these advantages are usually accompanied by relatively low accuracy of 0.2-0.3°. A 300mm wafer tilted by 0.3° rises more than 1.5mm over the diameter-more than enough to cause significant wafer transfer problems. In addition, circular spirit levels cannot be placed directly on wafer carrier shelves or transfer pins. Placing a circular bubble level on a dummy wafer, which itself may be warped and will sag ~0.5mm under its own weight, further degrading inclination measurement accuracy. Finally, viewing the bubble image reflected by an inspection mirror through a narrow opening further degrades the inclination measurement accuracy.

High accuracy spirit levels, called “machinists’ levels,” can provide up to 0.003° measurement resolution and can be read over a small range of angles to obtain scalar inclination data. This data can be used to implement statistical process control, but the large size, heavy weight, and considerable contamination shed by these instruments limit their use mainly to mainframe installation.

Figure 2. Evidence of chamber deformation.Click here to enlarge image

Most transfer coordinates are taught using the dowel pin method, whose steps are well known; but this method shares many of the drawbacks of the circular bubble level check: the chamber must be opened, the technician must enter the chamber to view alignment, and only attribute data are generated. It is not possible to measure the transfer coordinates using the dowel pin method-they can only be reset.

Alternative approaches

An alternative to traditional inclination measurement instruments is the “wafer-like” level, a concept that made its appearance in wired form almost a decade ago. This consists of a wafer-shaped platform that supports two uni-directional electronic spirit levels, plus a cable that connects the electronics capsule to a handheld display unit. These devices display scalar data, but cannot be moved through the tool automatically or operate within an evacuated chamber.

The advent of the 300mm Prime Initiative has raised the bar for wafer productivity. It makes it imperative for equipment suppliers and vendors to look harder at current methods of tool alignment, and research new methods for not only reducing wafer contamination and eliminating tool-to-tool process variances, but also significantly shortening the time it takes for fab engineers to set-up equipment and troubleshoot problems. This has led to the development of wireless tools that decrease the need for human interference and simplify the movement of wafers through process equipment so that it is more predictable, reliable, and controllable for greater throughput and yield.

More recently, an array of wireless wafer-like sensors have reached the market, including the WaferSense ALS inclinometer and software tools developed by CyberOptics Semiconductor. These wireless systems are adapted for automatic handling and for use under vacuum and utilize Bluetooth communication and a high-accuracy two-dimensional inclinometer in a low contamination package to enable measurements while the chamber is being pumped down, revealing for the first time the magnitude of chamber deformation (Fig. 2).

In the auto teach system, an on-board camera captures live video from inside semiconductor equipment as an on-board image processor reports the xyz offset from the teaching wafer to a target present inside the equipment. The data produced by this procedure is recorded and logged in real-time so critical handoff positions can be set, compared, adjusted, and repeated. These auto teaching systems have been shown to be accurate to ±0.1mm (±0.004 in.) in the x and y position; and ±0.5mm (±0.02 in.) in the z position, and to reduce teaching time by several hours. In a coater/developer teaching application, it was reported that robot teaching time was reduced from two shifts to <2 hrs.

Conclusion

At the 18th Annual IEEE/SEMI Advanced Semiconductor Manufacturing Conference, keynote speaker Areih Lev Greenberg, senior principle at Qimonda AG stated, “Problems such as low equipment availability and reliability, variation in parameters, high loss from batching and the low efficiency associated with inherently complicated tools...can no longer be tolerated.”

As the semiconductor industry strives to improve 300mm processing, it will need innovative approaches to optimize the automatic handling systems that have become indispensable. Accurate and reliable control of wafer support inclinations and wafer transfer coordinates are a natural outgrowth of the 300mm Prime initiative. The new wireless sensor solutions enable effective control of critical parameters that can minimize wafer damage and particulate contamination, reduce downtime, and lead to greater productivity, higher yield, and lower total cost of ownership.

Acknowledgment

Bluetooth is a registered trademark of the Bluetooth SIG Inc.