Backside Fault Isolation Using A Magnetic-Field Imaging System On Srams With Indirect Shorts

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Backside Fault Isolation Using a Magnetic-Field Imaging System on SRAMs with Indirect Shorts

L. A. Knauss, B. M. Frazier, and A. B. Cawthorne Neocera Inc., 10000 Virginia Manor Rd. Beltsville, MD 20705 E. Budiarto, R. Crandall, S. Melnik, and C. Bennett Intel Corporation, 5200 N.E. Elam Young Parkway Hillsboro, OR 97124

Abstract
With the arrival of flip-chip packaging, present tools and techniques are having increasing difficulty meeting failure-analysis needs. Recently a magneticfield imaging system has been used to localize shorts in buried layers of both packages and dies. Until now, these shorts have been powered directly through simple connections at the package. Power shorts are examples of direct shorts that can be powered through connections to V and Vss at the package dd level. While power shorts are common types of failure, equally important are defects such as logic shorts, which cannot be powered through simple package connections. These defects must be indirectly activated by driving the part through a set of vectors. This makes the magnetic-field imaging process more complicated due to the large background currents present along with the defect current. Magnetic-field imaging is made possible through the use of a SQUID (Superconducting Quantum Interference Device), which is a very sensitive magnetic sensor that can image magnetic fields generated by magnetic materials or currents (such as those in an integrated circuit). The current-density distribution in the sample can then be calculated from the magnetic-field image revealing the locations of shorts and other current anomalies. Presented here is the application of a SQUID-based magnetic-field imaging system for isolation of indirect shorts. This system has been used to investigate shorts in two flip-chip-packaged SRAMs. Defect currents as small as 38 A were imaged in a background of 1 A. The measurements were made using a lock-in thechnique and image subtraction. The magnetic-field image from one sample is compared with the results from a corresponding infrared-microscope image.

Introduction
As integrated-circuit process technologies become more complex and the industry moves toward flipchip packaging, present tools and techniques are having increasing difficulty in meeting failureanalysis needs [1]. In particular, flip-chip packaging requires that nondestructive measurements be made through the silicon substrate. Package substrates for these new integrated circuits are also becoming more complex with finer-pitch dimensions and many layers of metallization, often with several ground and power planes that complicate nondestructive analysis. To meet the needs of failure analysis for some present and most future applications, new techniques are needed. Some of these techniques will come through enhancements of conventional approaches like backside emission microscopy, and others will come from entirely new technologies like backside laser voltage probes and techniques like PICA.

Magnetic-Field Imaging System


One way to overcome some of these difficulties is a new technique that enables magnetic-field imaging. From magnetic-field images, the source currents can be calculated providing the failure analyst the ability to see a map of current in the device. By mapping the current in an integrated circuit or a package, short circuits can be localized and designs can be verified to see that current is flowing where expected. Unlike thermal, optical, ion, or electron beam techniques, magnetic fields are not affected by the materials in an IC or package, therefore imaging can be performed from both the front and backside of a device through many layers of metal or packaging materials. The only difficulty is that the strength of the magnetic field decreases with current magnitude and increasesing separation between the sensor and source currents. The rate of decrease depends on the nature of the current source, but for ICs the magneticfield strength is typically inversely proportional to the distance between the sensor and t he source.

The ability of imaging effectively at a distance depends on the sensitivity of the magnetic sensor. While there are many magnetic sensors available, from induction coils to Hall probes to MFM tips, the most sensitive magnetic sensor known is a superconducting quantum interference device or SQUID [2]. SQUIDs can be designed to measure fields as small as 1 femtotesla (10-15 tesla) or 40 billion times smaller than the Earths magnetic field. In order to operate, a SQUID needs to be cooled to cryogenic temperatures of less than 90 K (liquid nitrogen temperatures) for a high-temperature SQUID and less than 9 K (liquid helium temperatures) for a low-temperature SQUID.

Experimental Results
Until now, all fault isolation with a magnetic-field imaging system has been of direct shorts where the defect can be activated by applying power to a pair of pins or pads on the package. Some published examples [3-5] include power-to-ground shorts, signal-line-to-ground shorts, signal-line-to-power shorts, or high leakage currents. Many shorts, however, cannot be directly accessed through connections at the package terminals. Logic shorts, for examp le, require a set of vectors to activate the defect in many cases. This provides a new challenge: because the defect is only active after driving the part to a particular state, so the defect current must be imaged in the background of full device power. It also may not be possible to generate an alternating current at the defect, further complicating detection of the defect current in a large background current. This was recently explored using an SRAM with a known short between a local bitline and ground. Indirect Short: AC Measurement To perform this measurement, a tester was connected via an umbilical cord to an interface card as shown in Figure 2. The interface card had a significant amount of magnetic material due largely to stainless steel comp onents. While this did not inhibit the measurements, reducing background magnetic fields can further improve image quality. Prior to imaging, the SRAM was supplied a set of vectors to activate the defective local bitline. This resulted in a DC background current of 1 A. By turning on the defective local bitline, a connection is made between the defect and an IO on the package. Through this

Figure 1. Photograph of the commercial scanning SQUID microscope, MAGMA-C1, with electronics rack and probe station. MAGMA-C1 (see Figure 1), a commercial magneticfield imaging instrument designed for failure analysis of ICs, packages, and boards has been developed using a high-temperature SQUID with a sensitivity of 20 picotesla or two million times smaller than the Earths magnetic field. By using a high-temperature SQUID, the microscope has been designed to keep the SQUID cold and in vacuum, while the device under test (DUT) is at room temperature and in air. The unique design of MAGMA-C1 also facilitates positioning the SQUID as close as 50 m from the DUT. Sensitivity is high enough to detect currents as small as 10 nA at a 100-m working distance with 1-second averaging, but low enough to enable the instrument to function in an unshielded environment contrary to the case of a low-temperature SQUID. The SQUID is held stationary while the DUT is raster scanned under the SQUID tip to acquire the magnetic-field image. The current supplied to the DUT is typically alternating at a frequency less than 20 kHz. By using a lock-in technique, an image of just the current can be acquired while static background fields are ignored. 2

Figure 2. Interface card with SRAM positioned under magnetic-field imaging system, MAGMA-C1.

(a)

Figure 4. Current image results overlaid on the CAD layout for the SRAM package. current travels 4200 m on the global bitline before switching to the local bitline and shorting to ground. The length of the current path on the global bitline agrees within the error of the image to the distance to the local bitline as measured in the die-level CAD layout. Indirect Short: DC Measurement (b) Figure 3. (a) Magnetic-field image of shorted current path in SRAM. (b) Current image of same defect. IO, a 15-kHz AC current of 900 A was applied. By using the built-in lock-in amplifier, the magnetic field of the defect current could be selectively imaged, as shown in Figure 3(a), even in the presence of the large background current. Figure 3(b) shows the resulting current-density image after inversion. The intesity changes correspond to a transition from current in the package at the bottom of the image to current in the die in the top half of the image. These changes in intensity are not surprising considering that the die was mounted in a flip-chip configuration and was therefore closer to the SQUID sensor than the package. The separation between the SQUID sensor and the current in the die was around 375 m and the separation between the SQUID sensor and the current in the package was approximately 500 m. Figure 4 shows an illustration of the current path as determined by the magnetic field imaging system overlaid on the package-level CAD layout. The path in the die is also illustrated, but the CAD layout does not contain the die-level metallization. Once in the die, the current goes through several switching transistors that put it onto the global bitline that carries the current to the defective local bitline. The 3 In the previous example, once the defect was activated, it was possible to apply an AC signal to the defect through an IO accessible at the package. This approach worked very well, but this convenient situation may not always be possible. It may be necessary to do all imaging with DC currents only. To explore this possibility, the same defect was investigated by using only DC currents. In Figure 5(a), the static magnetic-field image is shown for the background magnetic fields produced by magnetic materials and the 1 A background current used to activate the defect. Figure 5(b) shows the magnetic-field image with the addition of 900 A of direct current to the defect. By subtracting these images, an image of just the defect current can be obtained and is shown in Figure 6(a). This image is noisier than in Figure 3(a), but still has a high enough signal-to-noise ratio to allow an accurate inversion to produce the current image in Figure 6(b). This is an exceptional result considering that the subtracted result is of a current 4 orders of magnitude weaker than the background currents. The success of this measurement can be attributed to the very low noise of the SQUID sensor.

(a)

(a)

(b) Figure 5. (a) Magnetic-field image produced by static materials and background current. (b) Magnetic-field image produced by static materials, background current and defect current.

(b) Figure 6. (a) Magnetic-field image after subtracting Figure 5(a) from Figure 5(b). (b) Current image obtained from magnetic-field image.

Defective Component Isolation A second SRAM device was studied. This device had a short that generated a hot spot detected by an infrared microscope. This hot spot was right on the edge of the die. As a result it was not possible to determine if the short was in the die or the package. Using the same set of vectors to activate the defect (1 A background current), an AC current was supplied to the IO connected to the defect. The current draw was only 38 A (5 orders of magnitude smaller than the background), however, the signal-to-noise ratio is still very good. Figure 7 (a) shows the magnetic-field image of the defective current and Figure 7 (b) shows the current-density image. The current path shown in Figure 7 (b) provides two indications that the defect is in the package and not the die. First, the current is observed to travel away from the die at the point where it should be entering the die (the edge of the die is indicated by a white line). Secondly, in the previous example, the 4

intensity of the current increased when it passed from the package into the die. In this case, the current image is most intense in the path in the package indicating that the current does not move into a layer closer than the package and therefore not into the die. This example indicates the capability of the magnetic field imaging system to isolate the defective component in an assembly.

Conclusions
As the semiconductor industry faces new challenges in failure analysis due to less accessible defects, it will be interesting to see the new techniques that emerge to address these needs. From the preceding examples, it can be seen that magnetic-field imaging using a SQUID sensor can be applied to indirect shorts in addition to direct shorts (previously demonstrated). It has also been shown that this tool can be a very effective means of isolating the faulty component in a fully assembled part. We have also seen that the high sensitivity of the technique

References
[1] J. Ouellette, The Industrial Physicist, p. 11, June 1998. [2] H. Weinstock (ed.), SQUID Sensors: Fundamentals, Fabrication, and Applications, (Kluwer Academic Publishers, The Netherlands), (1996). [3] L.A. Knauss et al., Detecting Power Shorts from Front and Backside of IC Packages Using Scanning SQUID Microscopy, ISTFA 99: Proc. 25th Intl. Symp. On Testing and Failure Anal. (Materials Park, OH: ASM International), p. 11 (1999). [4] E.F. Fleet et al., HTS Scanning SQUID Microscopy of Active Circuits, IEEE Transactions on Applied Superconductivity, 9(2) 4103 (1999). [5] W.E. Vanderlinde et al., Localizing Power to Ground Shorts in a Chips-First MCM by Scanning SQUID Microscopy, IRPS 00: Proc. 38th Intl. Reliability Physics Symp. (IEEE cat.# 00CH37059), p 413 (2000).

(a)

(b) Figure 7. (a) Magnetic-field image of shorted current path in package of SRAM. (b) Current image of same defect. White line indicates the perimeter of the die. facilitates imaging very weak currents, even in the presence of a background current five orders of magnitude stronger. This high sensitivity also facilitates effective imaging with much lower currents than thermal techniques, and it provides additional information in situations where thermal imaging can be applied.

Acknowledgements
The authors would like to thank S. Chatraphorn, E. Fleet, and F. Wellstood for their assistance and insightful discussions.

You might also like