Current Imaging using Magnetic Field Sensors

Transcription

1 Current Imaging using Magnetic Field Sensors L.A. Knauss, S.I. Woods and A. Orozco Neocera, Inc., Beltsville, Maryland, USA Introduction As process technologies of integrated circuits become more complex and the industry moves toward advanced packaging like flip-chip and stacked die, present tools and techniques are having increasing difficulty in meeting failure analysis needs [1]. With gate sizes approaching 65 nm, killer defects may only be a few nanometers in size. In some cases, the defects are nonvisible, i.e. there is no particle that can be imaged by optical microscope or SEM. The increasing number of transistors on a die is also requiring more levels of metal interconnect, which can limit thermal and optical techniques. The more complex devices today have 6 levels of metal, but many companies see 10 to 12 levels in the near future. Further complicating die level analysis are the trends in packaging technology. Flip-chip packaging requires that nondestructive measurements be made through the silicon substrate, and stacked die packaging can require that data is taken through multiple die and packaging materials. The package substrates for these new integrated circuits are also becoming more complex with finer line dimensions approaching 10 µm 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, techniques are needed that are not obstructed by these complications. To some extent this can be accomplished in electrical test through scan architectures once adopted. However, diagnosis of defects using such methods is limited to one logical node or wire, which can often be greater than 200 µm in length and traverse many levels. Further, such diagnostic methods are often non-existent for high current failures and faults in analog devices. Magnetic current imaging is one such technique that can provide failure analysts with a tool to help overcome some of the hurdles involved in fault isolation of present and next generation semiconductor devices. Through the use of a sensitive magnetic sensor, currents in integrated circuits can be imaged via the magnetic fields they produce. Unlike thermal, optical, ion or electron beam techniques, low frequency magnetic fields are not affected by the materials in an IC or package. Therefore, imaging can be performed from both the front or backside of a device through many layers of metal or packaging materials. These images can reveal the locations of shorts and other current anomalies at both the die and package levels. This technique has applications in fault isolation, design verification, and defective component isolation in full assemblies. A description of this technique, the sensor technology used, and a summary of the various applications of this tool at the die, package, and assembly levels are presented in this chapter. Physical Principle Magnetic current imaging uses the magnetic fields produced by currents in electronic devices to obtain images of those currents. This is accomplished though the fundamental physics relationship between magnetic fields and current, the Biot-Savart Law, v v v µ 0 Idl r db =, (1) 2 4 π r where B is the magnetic induction, Idl is an element of the current, the constant µ 0 is the permeability of free space, and r is the distance between the current and the sensor. As a result, the current can be directly calculated from the magnetic field knowing only the separation between the current and the magnetic field sensor. The details of this mathematical calculation can be found elsewhere [2,3], but what is important to know here is that this is a direct calculation that is not influenced by other materials or effects, and that through the use of Fast Fourier Transforms these calculations can be performed very quickly. A magnetic field image can be converted to a current density image in about 1 or 2 seconds. Once the current density image is obtained, the information can be used to localize shorts in the packaging, the interconnect, or the die. High resistance defects like cracked traces, delaminated vias and cracked or non-wet bumps can also be localized by looking for small perturbations in the magnetic field between good and bad parts. The current distributions can also be used to verify designs or hunt down I DDQ leakage. In principle, the current density images have the potential to find any type of current related defect or provide any type of current related information. System Principles The basic components of a magnetic current imaging system are shown in Figure 1. The magnetic field produced by a sample can be imaged by rastering the sample or the magnetic

2 sensor in close proximity to one another. If the sample contains a permanent magnetic field, as in many land grid arrays and lead frames, the system will image this constant or DC magnetic field. More importantly, current in the device produces a magnetic field. Fields produced by constant currents can be imaged along with permanent magnetic materials, whereas fields produced by alternating currents can be isolated from DC fields and result in a higher signal-tonoise ratio. In AC mode, the signal from the magnetic sensor is sent through a lock-in amplifier, which detects signals with the same frequency and phase as the applied current. This removes the effect of any magnetic materials or magnetic fields produced by constant currents. Interference from ambient time-varying magnetic fields is likewise negated. Thus, the AC magnetic image measures only the field produced by the applied alternating current through the part. In most systems, the magnetic sensor is oriented to detect the z-component of the magnetic field, i.e. the component perpendicular to the scanning plane. To understand the image generated by the instrument, consider the case of a long straight wire carrying a current I (see Figure 2). As the sensor moves over the wire, the z component of the magnetic field will be first negative, then zero, then positive, as seen in Figure 2. A two dimensional image of B z (the 'z' component of the magnetic field) for a simple wire is shown in Figure 3. The two regions on either side of the wihte line correspond to the negative and positive B z and the white line corresponds to the location of the current where B z = 0. To best locate a short in a buried layer, however, the magnetic field image is converted to a current density image. The resulting current map can then be used to determine the fault location either directly or by comparing to a circuit diagram or reference part image. Sensors Magnetic sensors have been around for a long time. The earliest sensor was a crude compass that used a lodestone, a magnetic rock, to sense the earth s magnetic field for navigational purposes. It is not known when this was first done, but there are references to this kind of use as early as the 12th century. Since then magnetic sensors have become much more sensitive and sophisticated. Some magnetic sensors of Computer and Electronics Lock-in Amp Magnetic Field Sensor Sample Staging Sample Stimulus -Funct. gen. -Tester -Other Figure 1: Block diagram of a magnetic current imaging system. Figure 2: Illustration of a magnetic sensor in the magnetic field of a long straight wire. Wire at center is producing circular magnetic field. Large up and down arrows indicate z component of magnetic field through sensor. Figure 3: Two-dimensional false color representation of B z of a long straight wire. Each side of the white line corresponds to oppositely directed flux and white corresponds to B z = 0. today include Hall effect sensors, flux gates, induction coils, magnetic force microscope tips, magnetoresistive sensors and SQUIDs (superconducting quantum interference devices). A table of the sensitivities of these sensors is shown in Figure 4 [4]. Of these sensors, only SQUIDs and magnetoresistive devices are practical for magnetic current imaging of integrated circuits today. The differences between these sensors correspond simply to sensitivity 1 and resolution 2. In a typical IC failure, currents are often much less than 1 ma and the sensor is separated from the sample by a distance greater than 150 µm. The ability to effectively image weak currents at a such distances depends on the sensitivity of the magnetic sensor. SQUIDs are the most sensitive magnetic sensors known [5]. They can be designed to measure fields as small as 1 femtotesla (10-15 tesla), which is 40 billion times smaller 1 Sensitivity for these sensors corresponds to the capability to detect weak magnetic fields. 2 Resolution here corresponds to the ability to spatially separate two current paths in a current density image.

3 Magnetic Sensor Technology Squid Fiber-Optic Optically Pumped Search-Coil Nuclear Precession Anisotropic Magnetoresistance Flux-Gate Magnetotransistor Magnetodiode Magneto-Optical Sensor Giant Magnetoresistance/SDT Hall-Effect Sensor * Note: 1gauss = 10-4 Tesla = 10 5 gamma than the Earth s magnetic field. SQUIDs for electronic fault isolation are typically designed to have sensitivity around 20 picotesla (10-12 tesla). This provides capability to image 500 na of current at a working distance of 400 µm under typical operating conditions. The best magnetoresistive sensors are about 2-5 orders of magnitude less sensitive, but they are easier to scale down in size to provide resolution less than 300 nm. SQUIDs As the name implies, SQUIDs are made from superconducting material. As a result, they need to be cooled to cryogenic temperatures of less than 90 K (liquid nitrogen temperatures) for high temperature SQUIDs and less than 9 K (liquid helium temperatures) for low temperature SQUIDs. For magnetic current imaging systems, a small (about 30 µm wide) high temperature SQUID is used. This system has been designed to keep a high temperature SQUID, made from YBa 2 Cu 3 O 7, cooled below 80K and in vacuum while the device under test is at room temperature and in air. A SQUID consists of two Josephson tunnel junctions that are connected together in a superconducting loop (see Figure 5). A Josephson junction is formed by two superconducting I b Detectable Field Range (gauss)* Figure 4: Sensitivities of common magnetic sensors. Solid lines represent demonstrated performance and dashed lines represent anticipated performance [4]. (Last updated by NVE 2004) I 0 Φ I 0 Figure 5: Electrical schematic of a SQUID where I b is the bias current, I 0 is the critical current of the SQUID, Φ is the flux threading the SQUID and V is the voltage response to that flux. V regions that are separated by a thin insulating barrier. Current exists in the junction without any voltage drop, up to a maximum value, called the critical current, I o. When the SQUID is biased with a constant current that exceeds the critical current of the junction, then changes in the magnetic flux, Φ, threading the SQUID loop produce changes in the voltage drop across the SQUID (see Figure 5). Figure 6(a) shows the I-V characteristic of a SQUID where V is the modulation depth of the SQUID due to external magnetic fields. The voltage across a SQUID is a nonlinear periodic function of the applied magnetic field, with a periodicity of one flux quantum, Φ 0 =2.07x10-15 Tm 2 (see Figure 6(b)). In order to convert this nonlinear response to a linear response, a negative feedback circuit is used to apply a feedback flux to the SQUID so as to keep the total flux through the SQUID constant. In such a flux locked loop, the magnitude of this feedback flux is proportional to the external magnetic field applied to the SQUID. Further description of the physics of SQUIDs and SQUID microscopy can be found elsewhere [6-8]. I 2I 0 V nφ 0 (n+1/2)φ 0 I b V Φ/Φ 0 Magnetoresistive Sensors The magnetoresistive microscope depends upon a sensor which is intrinsically less sensitive to magnetic fields than the SQUID, but is more readily miniaturized to the nanoscale. These magnetoresistive sensors are routinely fabricated with dimensions less than 100 nm. These devices are commonly used to sense submicron bits in hard drives, and recent papers have demonstrated magnetoresistive-sensor microscopes with submicron resolution [9,10]. If such a sensor can be brought within about 1 micron of currents to be measured, it has the magnetic sensitivity to map out submicron current lines carrying about 1 ma or more of current. Magnetoresistive devices are fabricated from materials whose resistance changes significantly in the presence of magnetic field. Before the 1990's these devices were of limited use, depending on anisotropic magnetoresistance (AMR) in thin ferromagnetic films, which only exhibit resistance changes of a few percent. In the last 15 years, magnetoresistive sensors have been revolutionized by the discovery of giant magnetoresistance (GMR) in multilayers composed of ferromagnetic and non-magnetic materials, with resistance V V 0 Φ (a) (b) Figure 6: a) Plot of current vs. voltage for a SQUID. Upper and lower curves correspond to nφ 0 and (n+1/2)φ 0 respectively. b) Periodic voltage response due to flux through a SQUID. The periodicity is equal to one flux quantum, Φ 0.

4 changes of up to 40% in fields as small as 10 Oe 3 at room temperature. The increased magnetic sensitivity of these magnetoresistive multilayers has made them the material of choice for hard drive read heads, enabling the rapid reduction of bit sizes in the last decade. Giant magnetoresistance multilayers are composed of alternating layers of ferromagnetic and non-magnetic layers. In the simplest design, a non-magnetic layer (only several atomic layers thick), is sandwiched between two ferromagnetic layers, as shown in Figure 7. One ferromagnetic layer is made magnetically hard, often pinned by exchange coupling to an antiferromagnetic layer, and one ferromagnetic layer is magnetically soft and its magnetization direction is free to rotate in small magnetic fields. The resistance of the multilayer depends upon the angle between the magnetization directions of the ferromagnetic layers, with minimum resistance when these directions are parallel and maximum resistance when they are antiparallel, as seen in Figure 8a. To make a device with high sensitivity and a linear, bipolar response, the hard layer is pinned perpendicular to the free layer. Then the response of the device to a magnetic field is approximately linear over the largest range with saturation at fields where the magnetizations of the layers become parallel or antiparallel, as shown in Figure 8b. Applying a constant current through such a device allows a simple measurement of the voltage across the device to yield its resistance change, which is directly proportional to the field at the device's free layer. Noise levels as low as 1 nt/hz 0.5 have been reported in magnetoresistive devices [11]. Antiferromagnetic (pinning layer) Ferromagnetic Layer 1 (hard) M Non-magnetic Layer I Ferromagnetic Layer 2 (soft) B from sample M Figure 7: Typical construction of a giant magnetoresistance sensor. The magnetization directions of the magnetic layers are shown with arrows (M) where the soft layer is free to move under the influence of the external field, B. The indicating current (I) is shown on the non-magnetic layer. 3 1 Oe (oersted) = 1 G (gauss) = 10-4 T (tesla) This relationship is only true for the oersted when the magnetic permeability is equal to µ 0. This is usually true for fields in and around semiconductor devices R 180 θ -10 Applications Magnetic current imaging began as a technique to image power shorts in integrated circuits and packages. While it is still heavily used for this, the capability has expanded to include I DD leakage, in-line test structure leakage, I/O leakage, internal circuit (isolated from any I/Os) leakage, hard logic fails, high resistance defects (sometimes called resistive opens), power distribution mapping, and isolating defective components in assembled devices. In the following examples section, we will show a few of these defect types. Magnetic current imaging is always used after electrical test as with most other fault isolation techniques. It can be used to isolate any electrical defect involving any anomalous current. This can be measured as a leakage or a resistance change in the device. For a resistance change, it can be either higher or lower than normal. At the time of this writing, magnetic current imaging is only applicable to static defects (i.e. those that can be held in their failing state). It is also possible to use this technique to image current in a good device for the purpose of verifying a design. This review, however, will focus on fault isolation only. The next step in the process is to decide which sensor type to use. This choice is made based on allowable working distance and how much current the circuit or defect will tolerate. If the device is packaged, the working distance between the sensor and the current is usually > 100 µm. In this situation, the sensitivity of the SQUID will produce the best results and require the least amount of deprocessing (potentially none at all). Starting with the SQUID can also be helpful in preserving a defect that might be lost through any deprocessing. If the device is not packaged or the front side of the chip is accessible, then a magnetoresistive sensor can be used to obtain higher resolution. The higher resolution is accomplished by the smaller sensor size and the closer working distance. It is always important to remember that for a near-field scanning technique like this, the resolution is limited by the sensor size or the working distance, whichever is larger. The sensitivity for magnetoresistive sensors is lower, but since they are used much closer to the source current, images of currents as small as 300 µa have been obtained [10]. This capability is still improving. In the most general case of a fully assembled device with a short, the SQUID would be the chosen sensor and used initially to isolate the defective component. Once the R 10 H (Oe) (a) (b) Figure 8: (a) Resistance change of the sensor as a function of angle between the magnetization of the ferromagnetic layers. (b) Resistance change of a linearized sensor as a function of applied magnetic field.

5 defective component is isolated and the location is determined to be in the die, interconnect, or packaging, the part would be partially deprocessed to enable imaging closer to the defect. The SQUID sensor or a magnetoreistive sensor would then be used to obtain a higher resolution image and more precise localization. At this point there may be enough information to move directly to cross-sectioning or parallel deprocessing. Alternatively, other fault isolation techniques may be used to get complementary information before final physical failure analysis. After physical failure analysis, there may be situations that require further proof that the observed defect is actually causing the electrical failure. If the defect is still electrically intact, magnetic current imaging can be used on the crosssectioned or parallel deprocessed part to verify that current is actually present in the defect. This can be very useful in situations where various parties are in dispute over ownership of the defect or when there is some liability involved for the defect. The last example will show this, with a more detailed discussion found elsewhere in this book. One of the primary advantages of magnetic current imaging over other physical fault isolation methods is that the analysis can begin on a fully packaged/assembled device without any deprocessing, which can minimize any risk of losing the defect and simplify the initial analysis. It is applicable to advanced packaging like stacked die. Also, one gets information that was previously not available, that is, a direct image of current in the device. Examples The following examples are a summary of typical applications of magnetic current imaging. There are many other applications for this technique and more being developed. In principle, the technique can be applied any time that an image of the current in a device would be useful. Isolating a Defective Component One of the first challenges in the isolation of a defect is to determine whether it is in the packaging, the die, or the interconnect. Magnetic imaging with a SQUID can be very useful for this due to its high sensitivity and ability to image through packaging materials without deprocessing. There are three primary ways to determine the defective component. The first involves recognizing the current distribution associated with the die, interconnect or package. Figure 9 shows three examples of shorts in a flip-chip device with the short in a) the die, b) the interconnect, and c) the package. For the short in the die, the current can be seen clearly distributed on the die (even die edges can be seen in current distribution) and concentrated at the point of the short. In the interconnect example, the current can be seen concentrated at the short location, but there is no current distributed throughout the die. Also, there is no current seen in the package because the short in the interconnect is much closer to the SQUID sensor and thus dominates the image at (a) (b) (c) Figure 9: Current density images of shorts in a flip-chip device at (a) the die, (b) the interconnect, and (c) the package. the chosen gain setting. In the package short example, current is seen in the package without any significant intensity variation between the short location and the current in the package [12]. This indicates that the short is in the package. This kind of analysis can be taken a step further and compared to CAD for the package or die to get more precise localization. The second approach involves calculating the depth of the current in the device from the shape of the magnetic field. This is a little less accurate and depends on the details of the current distribution and the complexities of the magnetic field distribution; however, it is usually possible to calculate the depth of the current in this way with enough accuracy to determine the defective component. If the second approach is unsuccessful because the current distributions are too complex to accurately calculate the current depth, then a third approach can be used. For this technique a good device is also imaged under the exact same conditions as the bad device. The signal-to-noise ratio can then be calculated for both images. Since the current level is the same in both measurements, the signal-to-noise ratio can be used to determine which current is closer to the SQUID sensor. Depth changes of a few hundred microns will make noticeable changes in the signal-to-noise. This can then be used to estimate if one current distribution is in the package or die. Isolating Low Resistance Shorts While magnetic current imaging can be applicable to all types of shorts, low resistance shorts can be difficult or impossible to isolate by many techniques due to the low power that they dissipate. For magnetic current imaging, only current is needed. The actual power that is generated by a defect is irrelevant, making this technique the only option in some cases for low resistance shorts. To demonstrate this capability, two results have been included from work published elsewhere in more detail [13]. The first example is a SRAM standby current failure and the second is an ASIC power short. The 4 MB SRAM is 0.4 µm device with 4 levels of metal. It was flip-chip mounted on a ceramic ball grid array and thinned to approximately 70 µm of silicon. The part was biased with 130 mvac pp and drew 100 µa of current (DC average power P eff = 6.4 µw). Figure 10 shows the current image obtained through the backside overlaid on the CAD layout. This matches the defect location and agrees with the

6 Figure 10: Backside current image of a standby current failure on a SRAM. The defect is indicated by the arrow and the whole image is overlaid on a CAD layout. results from other techniques like Thermaly Induced Voltage Alteration (TIVA), liquid crystal, and Photon Emission Microscopy (PEM), but acquired with more than 2 orders of magnitude less power. The ASIC is a 0.1 µm product with 6 levels of metal, which had a power supply short. The device was flip-chip mounted on a CBGA substrate. It was biased with 250 mvac pp and drew 6.5 ma of current (DC average power P eff = 800 µw). Figure 11(a) shows the current image overlaid on a backside optical image with a close-up of this region and current direction shown in Figure 11(b). The four lobes in the current image indicate current coming together on one layer and shorting in the middle to another layer where the current separates into opposite directions. This constitutes a plane-toplane short at the center of the region where the current seems to have disappeared, indicated by the cross in Figure 11(b). The current is missing here because it is moving vertically between the layers, which is a magnetically silent direction for the SQUID. The location agrees with the data obtained with a Schleiren Thermal Mapper, which required about two orders of magnitude more current, also shown at the center in Figure 11(b). The resulting deprocessing at this location shows considerable shorting damage, as shown in Figure 12. Isolating High Resistance Defects (Resistive Opens) Typically, a high resistance defect is the result of a geometrical change in some circuit element such as a delamination process, crack, void, etc. Previously, the main approach for localizing these defects has been time domain reflectometry (TDR). TDR sends a short electrical pulse into the device and monitors the time to receive reflections. These reflections can correspond to shorts, opens, bends in a wire, normal interfaces between devices, or high resistance defects. Ultimately anything that produces an electrical impedance change will produce a TDR response. These signals are compared to a good part and require time consuming layer-bylayer deprocessing and comparison to a standard part. When complete, the localization is typically at best to within 200 microns. Clearly, the current distribution will be affected by such geometric alterations and correspondingly affect the magnetic field distribution as sketched in Figure 13 for a failing flip-chip bump. In this situation, one expects to see a small change in the magnetic field distribution around the defect as compared with Figure 11: Backside current image of ASIC power short overlaid on backside optical image. (b) Close-up of defect location. White arrows show current converging from top and bottom and diverging from left and right with plane-to-plane short at the center of this current cross. Figure 12: SEM cross-sectional image of shorting site identified in Figure 11. that in a good part. The localization of high resistance defects through current imaging is accomplished through a detailed comparison of good and failing parts. The differences in magnetic field explained above are small and therefore require a very careful analysis between the good and failing parts. This requires improvements over conventional technology in two areas. First, the instrumentation for current imaging requires more precise automated control of the sample setup and data acquisition. The scan conditions must be as similar as possible between the good and failing parts, so that an effective comparison can be made. Secondly, even with careful sample setup and data acquisition, there will still be misalignments between the two images, and potential signal differences due to different working distances, or even part Die Substrate Good Die Substrate Fail Delamination Figure 13: Illustration of current distribution in a good and a delaminated (failing) flip-chip bump.

7 deformations (e.g. warping). These differences need to be sorted out from those due to the high resistance defect. For this, advances have been made in image difference analysis (IDA) to assist in the identification of failing defects. The following example is a flip-chip bump failure, which can be found elsewhere along with other examples [14]. TDR was used to determine that the failure was likely in the interconnect since the signal in the failing part did not get far beyond the substrate, as shown in Figure 14. This failure was between power and ground for which there were several bumps. TDR did not definitively indicate that the problem was in the bumps nor did it identify which bump had the problem. Magnetic current imaging was used to further isolate the problem. Figure 15 shows the region of interest from the IDA image overlaid on the CAD for the particular structure involved. associated with the defect are clearly visible. The defect was isolated with an accuracy of 30 µm and verified by FA as shown in Figure 16. IDA on magnetic field images has successfully localized high resistance defects to this same level of accuracy on a variety of defects, including cracked traces and delaminated vias. This 30 µm localization by IDA is entirely non-destructuve and represents about an order of magnitude improvement over TDR. Bare Substrate Failing Passing Figure 16: Optical image of cross-sectioned failing bump. Non-wet part of bump is circled. Figure 14: TDR results indicate a resistive open that is close to, but beyond the package substrate. The metal trace connecting to the failing bump is marked with the dashed yellow line. The flip-chip bump is connected at the end of it before the green grid. The centroid for the magnetic anomaly is the black dot, aligning very well with the position of the bump. The IDA plot shown in Figure 15(a) does not present adequately the relative intensity of the magnetic anomaly. A better way of doing that is by using a 3D representation as in Figure 15(b), where the z axis is the magnetic field intensity corresponding to the IDA results in Figure 15(a). We plot the absolute value of the magnetic field zoomed-in around the anomaly location for the sake of clarity where the two peaks High Resolution Imaging of Current Paths As was discussed in the applications section, the resolution of magnetic current imaging is dependent on sensor size and working distance. In many imaging situations, the working distance is the limiting factor to resolution, but in situations where the sensor can be brought close to the current source, submicron imaging of current is possible and can provide critical localization of die level defects. Two examples of the resolution attainable are given here using a magnetoresistive sensor. In Figure 17 (a), a current image is shown for a front-side accessible chip. A few milliamps of current were applied for this image, which was made at a working distance of about 3 µm. The detailed features in the current image can be used as references to do a two-point alignment for the overlay shown in Figure 17 (b). Figure 18 Figure 15: (a) IDA results corresponding to the HR flip-chip bump damage overlaid on the CAD. (b) Zoomed-in 3D version of the IDA results presented in (a). (a) (b) Figure 17: (a) High resolution current density image of frontside wire bonded device. (b) Overlay of current density image on optical image.

8 (a) (b) Figure 18: Optical image and high resolution current density images of a serpentine process monitor. High resolution current images are overlaid on SEM images. Line widths are 300 nm with 300 nm spacing. shows a current image of a serpentine process monitor. The metal lines in this image are 300 nm wide with 300 nm spacing. The current used was 1 ma. The zoomed-in image easily resolves these features and could clearly resolve even finer structures. Defect Validation Once a defect has been localized and physical failure analysis has identified it, the failure analyst s work is usually done. In some cases, it is necessary to go one step further and verify that the defect identified is actually the electrical fault under investigation. This final validation can be necessary when ownership of the corrective action is in question or when there is legal liability involved. Magnetic current imaging can be used to do this final validation (assuming that the electrical defect is still intact) by allowing one to see the electron flow through the physical defect. This can be irrefutable proof that the physical defect is also the cause of the electrical fault. For example, Figure 19 (a), shows an optical image of a PCB with a current image overlay. This was originally a large board that had been sectioned to an area-of-interest as seen in the optical image. The bright spot in the current image corresponds to the location of a plane-to-plane short. After cross-sectioning at this point, a physical defect can be seen between the two shorted layers in Figure 19 (b). However, additional proof was needed to show that the physical defect was the source of the electrical short. Figures 19 (c) and (d) show a current image of this defect taken in cross-section registered with the optical image. The cross-hair positioned on the peak in the current image lines up with the location of the defect in the optical image. This data proves that the observed physical defect is in-fact the electrical fault in this PCB. The full case history of this part is discussed elsewhere in this book. (c) (d) Figure 19: (a) Optical image of PCB with current overlay. (b) Optical image of defect cross-section. (c) Current image obtained from cross-section. (d) Optical image registered with current image where the crosshair marks and validates the current through the defect. Summary and Future Directions The examples presented here demonstrate that magnetic current imaging can be a very effective means of fault isolation in die, packages and assemblies. One of the factors making this technology unique is its ability to image current, regardless of voltage or power dissipation, in an integrated circuit or package in a noncontact, nondestructive way. Since low frequency magnetic fields are not affected by any of the materials in an integrated circuit assembly, this technique is especially beneficial in situations where the defect is buried under layers of metal, silicon, or encapsulation materials where other techniques are very limited. Many defects can be imaged for coarse localization without any deprocessing of the sample at all using the high sensitivity of a SQUID sensor. High resolution scans with a magnetoresistive sensor may require deprocessing, but submicron resolution can be achieved when the sensor is close to the source. The capability of SQUIDs to work at large distances due to their sensitivity makes them ideally suited for isolating defects in packaged devices. They can be used to isolate defects between package, die and interconnect. This high sensitivity allows them to detect currents in the 100 s of nanoamps even in packaged assemblies, and enables the detection of subtle differences in current paths between good and failing devices. The latter capability enables isolation of high resistance defects like cracked flip-chip bumps and delaminated vias in packages. After physical analysis reveals the defect, magnetic currentimaging by SQUIDs or magnetoresistive sensors can be used

9 to uniquely verify the exact current path and validate the electrical fault. There is ongoing work to expand the capabilities of magnetic current imaging. High-speed data acquisition for waveform extraction could provide a means of probing switching currents rather than voltages. Quantitative current measurements on test structures could provide valuable data to designers for improving models that would ultimately speed new circuit design. With the combination of SQUID and magnetoresistive sensors, magnetic current imaging has the capability to meet the fault isolation needs of today s complex structures, including stacked die packaging, and densely populated die with numerous levels of metal. Coupling this technology with advancements in other techniques will provide the failure analyst with a complementary set of tools to isolate the defects of today and tomorrow. Acknowledgements The authors would like to thank the suppliers of the samples and data presented here. The authors also thank J.O. Gaudestad, A. Gilbertson, C. Hillman, D. Vallett, and Z. Wang for their assistance and insightful discussions. References 1. David P. Vallett, From Microns to Molecules Can FA Remain Viable Through the Next Decade?, Proceedings of the 28th International Symposium on Testing and Failure Analysis (ISTFA), pp (2002). 2. J. P. Wikswo, Jr. The Magnetic Inverse Problem for NDE, in H. Weinstock (ed.), SQUID Sensors: Fundamentals, Fabrication, and Applications. Kluwer Academic Publishers, pp , (1996). 3. S. Chatraphorn, E.F. Fleet, F.C. Wellstood, L.A. Knauss and T.M. Eiles, Scanning SQUID Microscopy of Integrated Circuits, Applied Physics Letters, vol. 76, no. 16, pp (2000). 4. Michael J. Caruso, Tamara Bratland, C. H. Smith, and Robert Schneider, A New Perspective on Magnetic Field Sensing, Sensors Magazine, vol. 15, no. 12, pp (1998). 5. H. Weinstock (ed.), SQUID Sensors: Fundamentals, Fabrication, and Applications, (Kluwer Academic Publishers, The Netherlands), (1996). 6. E.F. Fleet, S. Chatraphorn, F.C. Wellstood, S.M. Greene, and L.A. Knauss, HTS Scanning SQUID Microscope Cooled by a Closed-Cycle Refrigerator, IEEE Transactions on Applied Superconductivity, vol. 9, no. 2, pp (1999). 7. J. Kirtley, IEEE Spectrum p. 40, Dec. (1996). 8. F.C. Wellstood, et al., IEEE Transactions on Applied Superconductivity, vol. 7, no. 2, pp (1997). 9. S.Y. Yamamoto and S. Schultz, Scanning magnetoresistance microscopy (SMRM): Imaging with a MR Head, J. Appl. Phys., vol. 81, no. 8, pp (1997). 10. B.D. Schrag, X.Y. Liu, M.J. Carter and Gang Xiao, Scanning magnetoresistive microscopy for die-level submicron current density mapping, Proceedings of the 29th International Symposium on Testing and Failure Analysis (ISTFA), pp. 2-5 (2003). 11. Xiaoyong Liu and Gang Xiao, Thermal annealing effects on low-frequency noise and transfer behavior in magnetic tunnel junction sensors, Journal of Applied Physics vol. 94, no. 9, pp (2003). 12. K.S. Wills, O. Diaz de Leon, K. Ramanujachar and C. Todd, Superconducting Quantum Interference Device Technique: 3-D Localization of a Short within a Flip Chip Assembly, Proceedings of the 27th International Symposium on Testing and Failure Analysis (ISTFA), pp (2001). 13. David P. Vallett, Scanning SQUID Microscopy for Fault Isolation, Proceedings of the 28th International Symposium on Testing and Failure Analysis (ISTFA), pp (2002). 14. A. Orozco, E. Talanova, A. Gilbertson, L.A. Knauss, Z. Wang, L. Skoglund, D. Smith, and R. Dias, Fault Isolation of High Resistance Defects using Comparative Magnetic Field Imaging, Proceedings of the 29th International Symposium on Testing and Failure Analysis (ISTFA), pp (2003).