Alignment impacts semiconductor manufacturing

Semiconductor manufacturing depends heavily on the ability of machine-vision systems to correctly locate, identify, and align wafers and circuits for testing. Inaccuracy in any of these activities can reduce yields significantly.

George Blackwell, director of product marketing for PC vision at Cognex (Natick, MA; www.cognex.com), says alignment accuracy is crucial to steps such as wafer probing, and its failure can cause expensive delays and downtime. Incorrect pattern identification during alignment, for example, can cause a prober to shut down completely and require assistance from a human technician. The technician may even have to retrain the vision system with the appropriate image. Poor-performing machine-vision systems can require thousands of such assists and produce thousands of damaged wafers over a 5- to 15-year life. Blackwell contends that inadequately performing vision subsystems have cost equipment companies market share and have added substantially to their support costs.

Considering its long history in semiconductor manufacturing, why should machine vision ever misalign or fail to find a wafer? The answer lies with relentless changes in manufacturing processes that force vision technology to evolve to identify ever-more-complex alignment patterns. Chemical mechanical planarization (CMP), metalization, new low-k dielectrics, continuously shrinking chip features, and similar developments add to the challenge.

Blackwell notes that examining machine-vision systems from several companies reveals that machine vision is not a commodity product. Analysis algorithms are often proprietary, and therefore differ from vendor to vendor. In addition, different companies implement even publicly available algorithms in different ways, so each company's experience with a particular algorithm can determine its success.

High-performing machine-vision systems generally exhibit the lowest costs when calculated over the system's life. Consider the simple example (from a real semiconductor-process tool company) in the accompanying table. A machine aligns and processes wafers at the rate of 100/hr.

System A achieves 99.90% correct alignments, while System B boasts 99.99%. As a result, System A experiences 0.1 incorrect alignments per hour—2.4 failures per day, assuming three 8-hour shifts in a 350-day year, or 840 alignment failures per year. System B, on the other hand, fails to properly align only 0.01 wafers per hour, 0.24 per day, and only 84 wafers per year, a difference of 756 alignments. The cost of those additional failures and other performance parameters must be taken into account when evaluating vision systems from different manufacturers.

The best choice for any operation is the one that contributes the lowest overall costs, a factor that does not necessarily correspond to the initial purchase price. Blackwell contends that semiconductor manufacturers understand this, weighing vision performance at 50% or more of the reason for selecting one piece of fab equipment over another.

No one will contend that the vision subsystem of a piece of semiconductor fabrication equipment is the most complex part of the system, but its performance can profoundly affect the overall economics of the fab operation..