Mining high-precision CD-STEM data for TFH manufacturing

by Haifeng Wang, Western Digital Corp.; Jason Arjavac, FEI Co.

Executive overview

Scanning transmission electron microscope (STEM) metrology of thin-film magnetic head wafer production processes has demonstrated its ability to improve frontend-to-backend correlation and thus production yield forecasting capability. CD measurement data has shown correlation as high as 88% to certain head-level electrical testing parameters. Also, strong contributors to yield loss in the wafer manufacturing process have been identified for in-line monitoring and process optimization. Early detection of manufacturing defects and the strong correlation to the backend test results are the major advantages of the in-line CD-STEM.

The high-precision data created by STEM-based process metrology provides not only sub-nanometer control of critical dimensions, but also a rich basis for correlation with functional test results that can reveal and prioritize yield-critical structural parameters for subsequent optimization and control. The relentless demand for more data storage and computer power in less space has driven many of the critical dimensions of microelectronic devices beyond the measurement capability of scanning electron microscopes (SEM). STEM offers at least an order-of-magnitude better resolution than SEM, and recent developments in microscope technology and automated focused ion beam (FIB) based sample preparation have improved analytical speed to the point that STEM now compares favorably with SEM in throughput, turnaround time, and even cost-of-ownership. STEM has become increasingly practical and necessary for control of many advanced wafer fabrication processes. In addition to its immediate value in process control, the accumulated data admits readily to interrogation with standard data mining techniques, such as correlation with performance measurements of the finished device that can discover which structural characteristics have the greatest impact on final yield.

High precision CD-STEM for TFH

Driven by consumer demand for increased areal data storage density, the physical dimensions of magnetic recording heads have decreased rapidly for both the reading and writing elements. As the dimensions approach those comparable to the latest semiconductor nodes, magnetic performance of the head becomes sensitive to very subtle Ångström-order structural variation caused by various processing equipment. This effect becomes increasingly prominent due to the geometric and compositional complexity of the reader/writer structures at sub-nanometer level.

Historically, the reader has been the major focus as it is the smallest magnetically functional device, particularly with the latest technological transition from giant magnetoresistive (GMR) to tunneling magnetoresistive (TMR) technology. Now, with the introduction of perpendicular recording technology, the writer is also approaching dimensions similar to the reader. Conventional SEM-based critical dimension metrology is challenged in both sensitivity and precision in this sub-nanometer regime.

The technical details of applying STEM imaging to wafer-level CD monitoring, demonstrating Ångström-order CD precision and operational feasibility in a volume production environment were previously described [1]. The technique has value in a variety of advanced microelectronics device wafer manufacturing applications, including semiconductor devices, thin-film magnetic heads, and microelectromechanical systems (MEMS). It has shown the ability to achieve throughput practical for in-line monitoring, greatly reduced cycle time that ensures nearly immediate feedback, and most importantly, Ångström-level precision, an order-of-magnitude better than SEM-based CD metrology.

In order to retain competitive advantage in both technology and profitability, it is critical to focus monitoring and control efforts on those key manufacturing processes that contribute predominantly to the final product performance, and thus effectively control final production yield. The lifecycle of a magnetic head wafer can be as long as several months, from the frontend wafer production to the backend assembly and test. The reader structure is built in the early stages of fabrication, within the first 10%-20% of the entire wafer production process flow. Any defects or process excursions that are not detected by conventional wafer level testing, usually performed on test structures or monitor wafers, would have unpredictable yield impact on the backend head manufacturing several months in the future. Moreover, all structural parameters monitored in the wafer manufacturing process do not contribute equally to backend head performance. Knowing the key contributors and focusing on controlling those parameters promotes optimal resource allocation and yield predictability.

Unlike most semiconductor wafers, recording head wafers contain tens of thousands of die, each one destined to become a head in the hard disk drives. Sampling active devices on live wafers immediately following the process not only allows immediate action to correct any manufacturing anomaly without continuing to invest resources on bad wafers, but also provides essential process variation information potentially correlating to backend test results. Although the CD-STEM process is destructive, the impact is small relative to the large number of heads on the wafer and insignificant in the cumulative yield.

The information gained has proven extremely valuable and effective in identifying process excursions, validating new design changes, and the most importantly, establishing frontend-to-backend correlation and yield impact.


Focused ion beam (FIB) systems (in this case, FEI Co.’s Model 875 dual-beam FIB) prepare the ultrathin STEM samples directly from in-line production wafers, which are returned to the production line immediately after samples are lifted-out of the wafer. Two highly-automated FIB preparation tools operate continuously so that sample preparation and sample transfer can be staged simultaneously to achieve minimum overall turnaround time. The total capacity and throughput of the sample preparation are matched to support one STEM (FEI’s Tecnai G2 S/TEM with 200kV FEG) to achieve a balanced workflow.

Figure 1: A typical STEM image output from IC3D metrology recipe with labels showing definitions of two typical measurements (junction angle and hard bias layer thickness) on a GMR reader.
CLICK HERE to view larger image

Depending on the wafer fabrication tools and manufacturing workflow, an optimized sampling plan can be created to maximize within-the-wafer precision of certain critical dimensions directly associated with manufacturing control parameters. Different plans can be created to reflect known correlations of sampling location distribution on wafer, e.g., different processing equipment generates unique distribution patterns of given parameters. Sampling plans can also be designed to accommodate the different requirements of engineering/development and manufacturing applications. Sampling plans are frequently updated as more understanding is acquired from the correlation between process parameters and reader characteristics. In the work described here, a site plan with a specific layout is formulated to establish such correlation at two different critical steps of reader buildup. For GMR readers, 20-30 measurements are extracted from each STEM image using automated metrology recipes, as shown in Figure 1 above.

A typical example of 2D contour map showing junction angle variation on the wafer is presented in Figure 2. Many within-the-wafer variations in two orthogonal directions have been confirmed and quantified for the first time, allowing precisely measurable control of the fabrication process to achieve the desired wafer uniformity. New STEM sampling steps in the wafer process flow can be carefully designed and inserted to ensure the least impact on normal workflow and negligible level of wafer contamination.

Figure 2: 2D wafer map contour plot of left junction angle in typical GMR reader structure.

The resulting measurement data, derived from STEM images, can be converted and imported to many popular statistical process control formats to support daily monitoring and process trend charting.

Results and discussion

Reader structure is tightly associated with ion beam milling and ion beam deposition conditions as well as with the variation of those processing tools controlling photo resist profiles, milling, and deposition. Understanding the impact of each processing variable on control parameters and identifying major contributors is critical to manufacturing optimization. CD-STEM technology quantitatively reveals the characteristics of those processing variables. Typically, as shown in Figure 3, hard bias (HB) thickness has ~4nm variation across the wafer from left to right, which accounts for a significant portion of the average HB thickness (left and right junctions behave inversely). Figure 4 shows a 2D wafer map contour (the data is plotted as function of X/Y locations) illustrating the distribution of HB thickness for both left and right junctions.

Figure 3: Cross-wafer distribution of hard bias layer thickness. Triangles stand for left junction and squares for right junction.

Figure 4: 2D wafer map contour plot of hard bias layer thickness for left and right junctions.

The ultimate benefit of the newly-acquired data is to establish statistically significant correlations between the dimensional measurement given by STEM metrology and various backend test results, including amplitude, asymmetry, Barkhausen noise, and hysteresis. This work employed a multivariable correlation algorithm on over 50 wafers worth of data from production wafers to identify strongly correlated parameters. For example, the correlation between the predicted and actual tested hysteresis is as high as 0.88 (R2), as shown in Figure 5.

Many process parameters in all three areas monitored heavily by STEM, i.e., photo profile, milling, and deposition, showed strong correlation to backend quasi-static test results. Controlling those parameters is essential to yield improvement and consistency. A typical example is that backend yield is usually closely associated a junction geometry parameter, junction angle distribution. In a controlled environment where two populations of wafers processed by two different processing tools responsible for junction angle formation show almost identical junction angle distribution across the wafer, except that the average junction angle from these two wafers has a consistent 3° shift in the STEM measurements. It was later found that this difference in junction angle mean resulted in a consistent difference in reader magnetic performance that directly associated with the observed difference in yield in the backend. This finding prompted quick and appropriate changes to processing conditions to reclaim the yield loss.

Figure 5: An example showing 88% correlation between backend test result and predicted hysteresis based on the CD-STEM measurements.

Short cycle time and high throughput permit the collection of a large volume of statistically significant data with reasonable turnaround in manufacturing environment. However, the benefits go far beyond the remarkably high data quantity. The precision of the measurements has confirmed that many nanometer scale variations are intrinsic to the process and not the result on variability in the inspection methodology. High sensitivity and high precision of CD-STEM allow the identification and control of those subtle intrinsic variations responsible for final product yield. Based on this information, resources can be focused on optimizing the key processes, yield predictions become more accurate and reliable, and design improvements can be made much early in the process flow.


CD-STEM technology has been successfully deployed in the thin-film magnetic head wafer production environment where it has generated immediate, measurable benefits in effective yield management. The significant volume of high quality reader CD data from in-line wafers has promoted a better understanding of the impact of processing variables on backend test results and product yield. High-impact variables are identified and become the focus of the wafer process control and design improvement to achieve high and consistent backend yield.

[1] Haifeng Wang, et. Al., “Scanning Transmission Electron Microscopy for Critical Dimension Monitoring in Wafer Manufacturing,” Microscopy Today 16:1, January 2008.


This work has received strong support within Western Digital’s magnetic head operations, as well as FEI Co. for the technical development. In particular, we greatly appreciate Jason Fang and Yingbo Zhang for continuing process optimization and data analysis and collaborative team work with thin-film, reader design, and product engineering organizations.


Haifeng Wang is a senior engineering manager, materials characterization group, at Western Digital Corp., 44100 Osgood Road, Fremont, CA 94539 USA; ph 510/683-7448; fax 510/683-7666; e-mail [email protected].

Jason Arjavac is a senior applications development engineer at FEI Co., 5350 NE Dawson Creek Drive, Hillsboro, OR 97124 USA; ph 503/799-7921; fax 503/726-7509; email [email protected].


Easily post a comment below using your Linkedin, Twitter, Google or Facebook account. Comments won't automatically be posted to your social media accounts unless you select to share.