David Grosjean, Rachel Elwell, and Jody Keller, Analog Devices; and Mike Clay, August Technology

Reducing defects and increasing yields in MEMS manufacturing involves the use of the same tools and methods as in standard IC manufacturing. However, in MEMS applications, deep-submicron inspection technologies are not critical because typical geometries are >1 µm. Automated inspection technology that detects >0.5-µm features can therefore locate defects of interest and their origins in MEMS processes. This article describes how such an inspection tool was used at Analog Devices (Cambridge, MA) to locate the source of a defect that caused yield loss at probe in MEMS accelerometers.

The MEMS accelerometers manufactured by Analog Devices' micromachined products division are primarily used as automobile crash sensors to trigger airbag deployment. Consequently, they must meet the highest quality and reliability standards, yet be inexpensive. For the manufacturer to achieve those goals, die sizes must shrink from one product generation to the next and yields must be maximized. Since most systematic yield-loss problems have been solved, current work focuses on eliminating the rare random defects that are more than spurious in nature.

Accelerometer Functioning and Fabrication

The accelerometer sensor is a single layer of suspended structural polysilicon, from which a movable proof mass and fixed interdigitated fingers for capacitive sensing are formed. The proof mass moves in the sensing axis in response to an acceleration, causing a change in distance between it and the fixed fingers. This distance change can be measured as a very small change in capacitance between the fixed fingers and proof mass. By including the circuit on the same chip as the sensor, extremely small changes in capacitance and, accordingly, small accelerations can be measured. Figure 1 shows a scanning electron microscopy (SEM) image of how this is implemented in silicon. The functioning of the accelerometer is described in greater detail elsewhere.^1,2

Figure 1: SEM image of the sensor element in the ADXL150 accelerometer.

By forming the sensor elements within a mature BiCMOS process, the integrated MEMS (iMEMS) fabrication process employed by Analog enables micromechanical structures and circuitry to be combined on one chip. That fabrication process involves four general steps:

1. All BiCMOS circuitry is created through the poly gates, with an area left empty in the center of the die for the MEMS sensor.

2. The sensor is created in successive steps: a sacrificial oxide and then a polysilicon layer are deposited, and then the sensor elements are patterned.

3. Aluminum interconnects are formed and the circuit is passivated.

4. The sensor elements are released while the circuit remains protected.

Defect Morphology

Because accelerometer fabrication combines MEMS and BiCMOS processes, defects formed in the one process can affect the other. However, defects in the circuitry and MEMS structures often have quite different morphology.³ During many circuit-processing steps, the sensor area and defects that may be present are covered. In subsequent processing steps, thick sacrificial oxide and sensor poly are deposited in that area, covering or decorating existing defects.

For example, when one particle lands in the circuit area of a die and another lands in the sensor area, the particle in the circuit area will probably be covered with various passivation layers. By the end of processing, the particle may appear as a small dot. Meanwhile, the particle in the sensor area will be covered by thick spacer oxide and sensor poly, resulting in a much larger, bubblelike defect. Although both defects have a common origin, they will appear to be different. Further investigation, involving focused ion beam cross-sectional analysis, energy-dispersive x-ray compositional analysis, or line partitions, is required to decipher the defects.

Defects in the sensor area that prevent movement of the proof mass are of particular concern. A test at probe deflects the proof mass with an applied bias to simulate acceleration and then measures the capacitance change. If a defect interferes with the movement of the proof mass, the die will fail. At Analog, a defect created on the ground plane below the sensor impeded sensor motion. While the same defect was also present in the circuit area, it had a very different appearance.

Finding a Killer Defect in the Sensor Portion of an Accelerometer

Choosing an Automated Inspection Tool. To locate the source of the defect that impeded sensor motion, investigators at Analog determined that the ability to detect defects smaller than 0.5 µm was not required for iMEMS production and that the use of tools with that capacity has distinct disadvantages. Consequently, after careful evaluation of several automated inspection units, the group decided to use a system that is less expensive and better suited to iMEMS inspection tasks than other systems commonly utilized in IC manufacturing: the NSX-105 macrodefect inspection tool from August Technology (Bloomington, MN). With its long-working-distance objective, the system is particularly well suited for inspecting the tall, 3-D topography of MEMS devices. It can discern a wide range of feature sizes and has a powerful feature for capturing, storing, and correlating pictures. The tool can also inspect defined regions of interest and program them for various inspection routines. It can differentiate among very small die and has very high throughput.

In addition to its ability to accommodate tall structures, the tool has a high depth of field that enables it to inspect die on bowed wafers. In contrast, other instruments can maintain focus only if wafers are entirely flat on the chuck. Moreover, the tool's wafer-edge-handling feature is useful for inspecting wafers with features on both sides.

The NSX-105 uses multipass software, enabling users to inspect different areas on the die at different settings. For example, in accelerometer inspection routines, the circuit portion of the die can be inspected at one magnification level while the sensor portion can be inspected at a much higher level, after which the captured images can be merged to form a single defect map. That ability can have a dramatic impact on inspection throughput.

Finally, the tool can incorporate a backlight beneath the wafer chuck. Although backlighting is not used in Analog's accelerometer inspection routines, it is useful for inspecting MEMS containing through-wafer etches for vias or moving structures, which are present in pressure sensors, micromirror arrays, and ink-jet heads.

Figure 2: SEM image of the defect located under a sensor that impeded sensor motion.

Defect Detection. The type of defect that caused yield loss at probe is shown in Figure 2. On the beam it was approximately 5 µm wide, but clearly it had been generated before the beam was deposited. Since the initial morphology of the defect was unknown and likely had changed during processing, the defect had to be found at the end of the line and then traced back to its origin through a sequence of reverse inspection steps. To accomplish that task, a line-partition inspection methodology was employed.

Investigators performed a line partition on two wafer lots and inspected the wafers at various phases of the fabrication process. The inspection tool was set to take pictures of all defects at each step in the line partition and to store them as jpeg files linked to defect maps. When the offending defect was discovered during final quality control, the investigators used the pictures to track the changing appearance of the defect in reverse and locate its source. Figure 3a illustrates the defect, while Figure 3b shows its position on a wafer map.

Knowing the location of the defect, the investigators could view it on maps derived from earlier layers. The ability of the inspection tool to take all the required pictures and organize them enabled the investigating team to identify the source of the defect quickly without having to speculate about its original appearance.

Determining the Source of the Defect

Defect Evolution. Figure 4 shows the evolution of the defect in the sensor area. In Figure 4d, a defect similar to that in Figure 2 is presented. Although the defect in Figure 4d was in the sensor area and not on the sensor itself, and although it should not have caused the device to fail, it was the defect the investigators were looking for. Once it had been found, pictures from the tool were analyzed to understand its evolution. The defect first appeared after the gate poly deposition step, as shown in Figure 4a. Figure 4b depicts the defect after the gate poly was etched, and Figure 4c shows it after the sensor poly was deposited.

After determining that the defect had formed around the gate poly deposition step, the investigators looked in the circuit area for the same type of defect. Figure 5a, another defect that first appeared at the gate poly deposition step, resembled the defect in the sensor area. Figures 5b–5d trace the evolution of the defect in the circuit area through the rest of the line—after gate poly etch, interlayer dielectric etch, and passivation.

A comparison between the image in Figure 4d with that in Figure 5d indicates that the defect had a very different appearance in the sensor area after sensor poly etch than it did in the circuit area after metal passivation. Had it not been for the picture-taking capabilities of the inspection tool, the investigating team would not have associated the two defects as originating from the same source. The effort to improve yields in the sensor portion of the process therefore improved yields in the the circuit portion of the process as well.

Defect Formation. During the circuit-formation process, a 5500-Å layer of gate poly is deposited over the entire wafer and then doped with POCl₃. The POCl₃ glass is removed from the surface using a dilute–hydrofluoric acid (HF) deglazing step.

The defect appeared after the HF deglazing step. Samples of the defect as it appeared immediately after that step were analyzed using Auger electron spectroscopy (AES), which showed that the defect was composed of silicon dioxide. Since the phosphorus content of the defect was below the AES detection limit, the defect probably was not POCl₃ residue but an oxide formed as a result of a water spot.

Water spots resulting from the POCl₃/HF deglazing step and the corresponding failure mechanisms are well documented in the literature.^4,5 However, the evolution of water spots into killer defects in the iMEMS process differs from that in the documented cases. In a standard iMEMS process, the gate poly is completely removed from the sensor area, after which a sacrificial oxide is deposited using plasma-enhanced chemical vapor deposition (PECVD). Anchors are etched in the oxide to provide contact for the sensor poly. The sensor poly is then deposited using low-pressure physical vapor deposition, and the sacrificial oxide is removed with dilute HF to release the sensor.^6,7

As illustrated in Figure 6a, the presence of oxide residue from the deglazing step prevents the gate poly from being removed during poly etch. Sacrificial oxide is then deposited over the defect using PECVD. The poor step coverage of the oxide deposition results in a "breadloafing" effect over the topography of the defect, as illustrated in Figure 6b. The sacrificial oxide is then exposed to HF during subsequent processing, which opens a slight seam in the sacrificial oxide around the defect created by the breadloafing effect. A conformal 2-µm layer of sensor poly is then deposited over the sacrificial oxide, which fills the seam around the defect. When the sacrificial oxide is etched and the sensor structure is released, the thin layer of polysilicon that filled the seam in the sacrificial oxide remains. The remaining polysilicon protrudes from the underside of the sensor and can inhibit its motion.

Figure 7: Cross section of a typical defect resulting from water spots.

This defect mechanism was confirmed when the sensor was removed from the wafer, flipped over, and cross-sectioned using a focused ion beam. An image of the cross-sectioned sensor is presented in Figure 7.

Conclusion

There are many ways to solve the problem of water-spot defects. One option is to use a dry etch process that has lower selectivity to oxide to remove the gate poly. Making the poly etch process less selective prevents residual oxide after the HF deglazing step from masking the gate poly. Another option is to improve the HF deglazing process by using a different cleaning chemistry, as documented in the literature.⁴ Yet another solution involves using a dry oxide etch to perform the deglazing process. That option prevents gate poly from being exposed to wet HF chemistry and therefore prevents water-spot defects from forming in the first place.

Figure 8. Water-spot defect density before and after a diffusion cleaning step.

Given the constraints of the toolset and process at Analog, the investigators decided to add a diffusion clean (HF/SC-1/SC-2) after the POCl₃/HF deglazing step. In a diffusion clean process, an HF dip step removes the water spots. Figure 8 shows that water-spot defect levels were reduced after the implementation of the extra cleaning step.

Acknowledgments

This article is a revised version of a paper presented at the IEEE/SEMI Advanced Semiconductor Manufacturing Conference and Workshop (ASMC) in Boston, May 4–6, 2004.

The authors gratefully acknowledge the assistance of several people who contributed to this project: Terry Egan for defect inspection work; Dana Northcutt and Bruce Wachtmann for useful discussions about the deglazing step; Brad Waterson for failure analysis; Tony Tolis for FIB work; Eric Lent, Matt Wilson, and Tom Bentz for applications help; and Barbara Berthold for SEM work.

References

1. KH-L Chau and RE Soulouff, "Technology for the High-Volume Manufacturing of Integrated Surface-Micromachined Accelerometer Products," Microelectronics Journal 29, no. 9 (1998): 579–586.

2. KH-L Chau, "An Integrated Force-Balanced Capacitive Accelerometer for Low-g Applications," Sensors and Actuators A, vol. 54, nos. 1–3 (1996): 472–476.

3. D Grosjean, "Reducing Defects in Integrated Surface-Micromachined Accelerometers," MICRO 21, no. 2 (2003): 43–54.

4. L Lowell, "New Deglaze Process for Doped Polysilicon," Solid State Technology 34, no. 4 (1991): 149–153.

5. L Peters, "Water Spots: The Scourge of Wafer Dryers," Semiconductor International 21, no. 9 (1998): 83–90.

6. TA Core, WK Tsang, and SJ Sherman, "Fabrication Technology for an Integrated Surface-Micromachined Sensor," Solid State Technology 36, no. 10 (1993): 39–47.

7. K Nunan et al., "Developing a Manufacturable Process for the Deposition of Thick Polysilicon Films for Micromachined Devices," in Proceedings of the Advanced Semiconductor Manufacturing Conference (Piscataway, NJ: IEEE, 2000), 357–366.

---

David Grosjean, PhD, is a staff yield engineer in the micromachined products division of Analog Devices (Cambridge, MA), He is responsible for enhancing yield and reducing defects throughout the manufacturing process. Before joining the company in 1999, he was an NSF-CGP postdoctoral fellow at NEC in Tsukuba, Japan, where he researched electromigration in aluminum/copper interconnects. He is a member of the Materials Research Society, IEEE, and Tau Beta Pi. He received a BS in applied math from Yale University in New Haven, CT, and a PhD in engineering physics from the University of Virginia in Charlottesville. (Grosjean can be reached at 617/761-7123 or david.grosjean@analog.com.)

Rachel Elwell is a senior process engineer in the micromachined products division of Analog Devices. Since joining the company in 2000, she has been responsible for diffusion and ion implant processes. Before joining the company, she worked at Kokusai Semiconductor Equipment as part of the process development group. She received a BS in chemical engineering from Northeastern University in Boston. (Elwell can be reached at 617/761-7148 or rachel.elwell@analog.com.)

Jody Keller is a yield engineering technician in the micromachined products division of Analog Devices, where she is responsible for defect reduction and microcontamination. Before joining the company in 1996, she worked on defect reduction and photolithography at Digital Equipment Corp. (Keller can be reached at 617/761-7159 or jody.keller@analog.com.)

Mike Clay is regional sales manager for August Technology (Bloomington, MN). He is responsible for sales in the western United States. He joined the company in 2003 as part of its acquisition of Semiconductor Technologies and Instruments, where he had been director of applications engineering and technical marketing. He has more than 16 years of industry experience, supporting monitoring, metrology, and inspection tools used in all front-end processes. Since 1997, he has specialized in automated wafer inspection technology applications and product marketing. (Clay can be reached at 214/252-9397 or mike.clay@augusttech.com.)