SQUIDs locate IC defects

Failure analysis benefits from small sensors that locate high-current paths.

Jon Titus, Editorial Director -- Test & Measurement World, 5/1/2001

The capability to measure small magnetic fields particularly interests people who must locate the causes of semiconductor failures. That’s because the strength of a magnetic field relates directly to the amount of current flowing in a circuit, which is an indication of the circuit’s “health.” Tiny magnetic sensors called SQUIDs (Superconducting QUantum Interference Devices) have come out of the lab and now find use in scanning microscopes that can plot current flow and locate circuit defects.

Although a few companies supply the SQUIDs, electronics, and mechanical components for scanning microscopes, most failure analysts will buy a complete microscope that’s ready to use. A scanning SQUID microscope (SSM), as shown in Figure 1, moves a sample across a SQUID in a raster-scan pattern and at a uniform height to gather magnetic-flux information. Prices for SSMs suitable for semiconductor failure analysis start at a few hundred thousand dollars.

In failed ICs, short circuits usually appear as a small area of intense magnetic flux. When failure analysts overlay SSM images on a CAD map of an IC’s features (Figure 2), they can locate where faults occur in a device and can then determine why the fault occurred. Under ideal conditions, a SQUID can detect as little as 10 nA flowing in a conductor that is 100 µm from the sensor.

The SSM’s electronics convert signals from the SQUID into a gray-scale or false-color image that indicates the intensity of the magnetic flux at each scanned point. (Generally, scan steps of 0.5 µm provide good results, although microscopes offer many other step sizes.)

How do SQUIDs work?

SQUIDs can detect magnetic flux down to about 10^–20 Wb, or a field of a about 5 fT for a coil 1 cm in diameter. By comparison, the flux from the earth’s magnetic field is about 10 orders of magnitude greater. In general, a SQUID measures the z-axis component of a magnetic field, so the sensor will put out a zero value right over a conductor carrying current, a positive value on one side of the conductor, and a negative value on the other side. (If necessary, you can get SQUIDs that measure magnetic flux in all three orthogonal axes.)

SQUIDs rely on three quantum-related effects: superconductivity, the Josephson effect, and flux quantization. Although quantum physics goes beyond the scope of this article, quick descriptions of these effects may help you better understand how a SQUID works.

Superconductivity—the absence of electrical resistance—occurs in metals and some ceramics at low temperatures. The temperature at which superconductivity occurs is called a material’s critical temperature, T _c. At that temperature, current flows unimpeded through a conductor.

SQUIDs make use of two types of superconductors: low-temperature superconductors (LTS) and high-temperature superconductors (HTS). The LTS SQUIDs use niobium (Nb) metal and operate at about 4 K (–269° C), the boiling point of liquid helium. The HTS devices—high-temperature being relative—operate at 77 K (–196° C), the boiling point of liquid nitrogen. The HTS materials make use of thin-film ceramics composed of yttrium, barium, and copper oxides, often abbreviated YBCO.

Figure 1. In addition to the SQUID and its associated electronics, a scanning SQUID microscope provides both a stage to position the sample and cryogenic cooling equipment for the SQUID (and perhaps for the sample).

LTS SQUIDs provide less noise and better sensitivity than HTS devices, but they may require more magnetic and thermal shielding. The low-temperature devices also can require more expensive cryogenic systems to maintain their temperature at 4 K.

In 1962, physicist Brian Josephson described the measurement of current flow due to quantum tunneling through a thin insulating layer between two conductors at the same potential. That condition exists in a loop or ring of superconducting material. Magnetic flux passing through the thin layer—the Josephson junction—determines the maximum current flow. Josephson’s theoretical work described the relationship mathematically. (In 1973, Josephson received a Nobel prize in physics for his quantum-tunneling discoveries.) 

Figure 2. A false-color gradient pattern of current overlaid on a device-pad layout shows areas of high current flow. The pattern shows a short circuit from a via to ground metallization. Courtesy of Neocera.

The flux-quantization effect describes the observation that magnetic flux within a superconductor can exist only in quantized multiples of 2.068 x 10^–15 Wb, usually noted as F₀. The unit F₀ goes by the name magnetic flux quantum. Even though flux can exist only in discrete quanta, a SQUID can measure magnetic flux many orders of magnitude smaller. The flux quantization does, however, affect how these measurements take place.

SQUIDs use all three quantum effects to convert magnetic flux into a measurable signal. You can obtain two types of SQUIDs, DC and RF, but DC devices predominate. A DC SQUID includes a superconducting loop or ring and two Josephson junctions (Figure 3). Electronic circuits bias the ring with a small current and measure the voltage developed across the ring. Because the ring is superconducting, the slight current flow depends on Josephson tunneling, and that in turn depends on the magnetic flux through the SQUID.

The DC SQUIDs can be small, typically 100 µm on a side, so they can get very close to a source of magnetic flux. Usually, though, the Josephson junctions don’t directly sense the magnetic field of a sample. Instead, the scanning microscope uses a small pickup coil in parallel with a larger coil near the Josephson junctions. The parallel coils simply transfer energy from the sample to the SQUID. People use the term SQUID loosely to include the Josephson junctions and superconductive loop as well as the coils. You can think of the entire assembly as the SQUID.

The output is relative

The raw output from a DC SQUID looks like the triangle wave shown in Figure 4. The wave goes through a complete cycle for each multiple of the magnetic flux quantum, F₀. Each quantum step in flux produces a periodic signal, thus you can’t equate that signal directly to magnetic flux. So, how do instruments use a SQUID to measure flux?

Figure 3. A DC-type SQUID uses two Josephson junctions in a small superconducting loop. The junctions produce a small current flow due to tunneling caused by the presence of a magnetic field.

Figure 4. The raw output of a SQUID shows the nonlinear signal generated as magnetic flux increases and passes through multiples of the magnetic-flux quantum, F0.

The electronic portion of an SSM uses the SQUID as a sensitive null detector in a negative feedback loop, usually called a flux locked loop (FLL). The loop circuit measures the voltage across the SQUID, amplifies it, and feeds it back to a magnetic coil near the SQUID to null out the magnetic flux the SQUID detects. The linear null voltage applied to the coil represents the change in magnetic flux density as seen by the SQUID. To provide actual current measurements, a computer—usually part of an SSM—uses Maxwell’s equations and Fourier analysis to convert the magnetic-flux data into current values.

SSMs also may modulate the feedback-loop signal and use signal-processing circuits to increase the accuracy of measurements. Keep in mind that it takes time—in the order of milliseconds—for the FLL to acquire a signal and settle, so sensor response isn’t instantaneous. An SSM can perform a coarse scan of an IC in about 10 min. A fine scan can take 30 to 40 min.

Check out resolution

The geometry of a SQUID plays a role in how well it can spatially resolve magnetic flux. Resolution specifies the minimum separation that must exist between two magnetic features so the SSM can still resolve both features. If an SSM can place its sensor very close to the sample, the inside dimension of the superconducting loop limits resolution.

As a rule of thumb, for a coil diameter of D, an SSM should place the SQUID within 2D of the active device you want to scan. At that distance, the SQUID’s spatial resolution approaches 0.1D. Thus, for a 100-µm coil diameter, you need to place the SQUID as close as 200 µm to get a resolution of about 10 µm. You cannot resolve closely spaced conductors that carry current in the same direction because the magnetic fields will “add” and look like one conductor. (In the example above, the diameter refers to the diameter of the small sensing coil that scans the sample. It does not refer to measurement of the superconducting loop.)

In practice, an SSM may not reach a resolution of 10 µm. Actual resolution depends on the SSM model, the sensor, and other operational variables, although it’s not unusual to get a resolution of about 16 µm. An instrument’s ability to resolve two nearby conductors differs from its ability to locate a single current peak, which would represent a short circuit. An SSM using an HTS SQUID can locate single current peaks with an accuracy greater than 5 µm, and an SSM using an LTS device can achieve an accuracy of 2 µm. That accuracy may not be sufficient to locate a specific submicron transistor, but it can place you in the vicinity so you know where to perform other tests.

Often, you can trade off coil dimensions for resolution and sensitivity. A smaller coil will decrease sensitivity because it samples a smaller area. But it will increase spatial resolution. As another rule of thumb, the sensitivity times the square of the spatial resolution equals a constant. That rule isn’t perfect, but it gives you a good sense of the relationship between sensitivity and resolution.

When you use an SSM to scan a semiconductor device, you can power the device with either DC or modulated power. If you modulate the power at a frequency between 2 kHz and 20 kHz, you can use a lock-in amplifier to reject noise. Some semiconductor packages may contain small amounts of magnetic materials, so by using a modulated power signal and a lock-in amplifier, you effectively filter out background flux from those materials.

Noise matters, too

As always, you want to ensure you don’t try to measure a signal below an instrument’s noise floor. A typical noise floor for a HTS SQUID is about 20 µF₀/EDHz. Thus, if you move a SQUID to a new position and sample its signal for 1 s, you can expect a noise level of about 20 µF₀. If you sample at a faster rate you acquire less noise, but you also have less time to make a measurement. Thus, if you want to speed sampling, you must ensure the SQUID can effectively sample a magnetic field at the increased rate. And you’ll have to compensate for the settling time of the FLL and the lock-in amplifier.

SQUIDs now seem limited to failure-analysis equipment that provides the needed precision mechanical positioning equipment, cooling apparatus, and electronics that make them work. But scientists may find superconductors that operate at higher temperatures, thus making SQUIDs available for more routine measuring tasks. T&MW

For more information

Kirtley, J.R., et al., “High-resolution scanning SQUID microscope,” Applied Physics Letters, Vol. 66, February 27, 1995. pp. 1138–1140. ojps.aip.org/aplo.

Knauss, L. A., et al., “Detecting Power Shorts from Front and Backside of IC Packages using Scanning SQUID Microscopy,” International Society for Testing and Failure Analysis, November 1999, San Jose, CA. ASM International, Materials Park, OH. pp. 11–16. www.asminternational.org/istfa/past.htm. Editor's Note 10/24/03: This paper is no longer available.

Wellstood, F.C., et al., “Magnetic Microscopy Using SQUIDs,” IEEE Transactions on Applied Superconductivity, Vol. 7, No. 2, June 1997. pp. 3134–3138.

 Jon Titus has written real-time software and designed embedded systems and computer/instrument interfaces. He worked in electronics for 10 years and spent nine years at EDN magazine prior to joining T&MW in 1993. He has a BS from WPI, an MS from RPI, and a PhD from VPI.