Automatic wafer inspection system replaces eyeballs with cameras


Christopher Eric Brannon, Texas Instruments, Inc., Dallas, TX

A fully automated RDS inspection system that replaces human inspectors is a game changer.

The chemical-mechanical polishing (CMP) process in the fabrication of silicon wafers is exceedingly labor intense. As a result, automating a critical aspect of CMP ??? namely, the physical inspection of wafers for metal residue following CMP ??? cannot only improve CMP cycle times, but it also can be integrated into a quality control system that improves wafer yields and maintains those yields at a high level.

The typical method for inspecting wafers as they emerge from CMP has involved a technician visually examining one wafer at a time as it is raised on an H-bar wafer elevation tool (FIGURE 1). The inspector would try to detect metal residue left on the wafer following an incomplete polishing process. If undetected, this metal residue can damage the devices made from the silicon. In other words, undetected metal residue can cause wafer and chip yields to plummet.

Figure 1. Manually inspecting a wafer on an H-bar tool.
Figure 1. Manually inspecting a wafer on an H-bar tool.

Of course, manual inspection processes have inherent limitations. Technicians may not perform a thorough inspection of the entire wafer or miniscule remnants of residue may elude human vision. Microscopes are also difficult to employ at today's technology nodes. And, expanding wafer sizes, from 200mm to 300mm, doubles the area that must be inspected, increasing the chances that residue somewhere on the wafer will elude the human eye.

As a result of this situation, research was begun to develop a low-cost, single-step automated process that would inspect 100 percent of all wafers after the CMP polishing process. Improved yields would be achieved by detecting residue more effectively and by incorporating the results of the inspection into a closed-loop feedback system that would control and dynamically fine tune the CMP process. At the heart of this automated residue detection system (RDS) would be a high-resolution imaging capture and compare process.

Capturing the golden wafer

First the RDS's camera must scan a fully polished and fully metalized wafer. This image will become the golden wafer or reference wafer for subsequent inspections. A grayscale map of the golden wafer is stored and reference values are associated with various characteristics of the wafer, such as the thickness variations in metals and dielectrics, the number and orientation of detectors, edge exclusions and the metals detected, usually tungsten or copper.

Figure 2. A grayscale image of a wafer.
Figure 2. A grayscale image of a wafer.

With the image of the golden wafer stored, the RDS can begin inspecting production wafers. Each production wafer is scanned and a grayscale image captured (FIGURE 2) which is compared to the golden wafer. The differences in the gray levels between the image of the production and that of the golden wafer will reveal any residual metal remaining on the production wafer (FIGURE 3).

Figure 3. RDS system flow diagram.
Figure 3. RDS system flow diagram.

As mentioned, the data gathered by the RDS following CMP polishing can be fed back into a control mechanism that can adjust the parameters of the CMP process itself and thereby improve its effectiveness. This is best illustrated by citing several brief case studies:

Erroneous CMP recipe

A set of 25 wafers was inspected with the RDS tool and residual metal was detected on every wafer (FIGURE 4). Manual inspection with a microscope confirmed the presence of the metal residue.

Figure 4. The RDS detected residual metal on the wafer
Figure 4. The RDS detected residual metal on the wafer.

The consistent location of the residue and the fact that it was present on each wafer indicated a systemic fault of some sort. This knowledge assisted in the troubleshooting exercise that ensued. Eventually, it was determined that the software controlling the CMP process had employed an incorrect polish recipe with shorter polish times and other erroneous process parameters.

Flow problems

In another case, the polish used in the CMP process began to exhibit random or non-linear behavior result patterns. Specifically, some, but not all, of the wafers inspected by RDS had a considerable amount of residual metal at the center of the wafer (FIGURE 5). The erratic polishing patterns led to an examination of the CMP slurry flow on the polishing pad. It was concluded that the flow of slurry to the pad was interrupted intermittently; hence, the erratic behavior patterns of the polishing process.

Figure 5. The RDS detected residual metal at the center of the wafer.
Figure 5. The RDS detected residual metal at the center of the wafer.

Operator error

Subsequent to the polishing of a batch of wafers, inspection by the RDS system revealed that five wafers were inexplicably left unpolished. Fortunately, the relatively fast cycle time of RDS allows for 100 percent inspection of all wafers. Eventually, the mysterious five unpolished wafers were attributed to operator error. Had RDS not been capable of inspecting every wafer, the five unpolished wafers might have been missed completely.

Benefits all around

Improving manufacturing yields on silicon wafers and semiconductor devices is the foremost goal of every chip manufacturer. Improving yields will increase operating efficiencies and profitability. Although the CMP metal polishing process is only one segment of the entire fabrication process, it is particularly crucial in terms of cycle times and yields because it has been an intensely manual part of the overall process.

The development of a fully automated RDS inspection system to replace human inspectors who can be error prone and less than comprehensive is a game changer on several fronts. First, as a standalone inspection tool, it is fully automated. This means RDS has a robust and fast cycle time, which gives the system the ability to inspect every wafer as it emerges from the CMP polishing slurry. This 100 percent sampling rate is hugely beneficial to the manufacturer because a much higher percentage of residue is detected when compared to a manual inspection process which would likely involve a sampling rate that would be much less than 100 percent.

Second, because of the small footprint of the RDS system, it can be integrated into a CMP process tool and provide a means to continuously monitor the CMP process, providing an early detection scheme for residual metal left on wafers. The data output by RDS can allow for tight control of the CMP process within pre-defined reference limits and it can give insight into the development of more effective CMP processing procedures.

CHRISTOPHER ERIC BRANNON, is a DMOS5 Copper CMP Manufacturing Engineer Texas Instruments, Inc., Dallas, TX.

Solid State Technology | Volume 56 | Issue 2 | February | 2013