Triax Cables can Improve On-Wafer Parametric Testing by David Rose
in Technology / Biotechnology (submitted 2010-12-29)
In wafer level DC measurements, anticipating the signal magnitudes makes it easier to assess potential sources of error. For simplicity, let's assume all of the signals being measured are neither high voltage (i.e., greater than 100V) or high current (greater than 200mA). We can break down the error sources into four categories: leakage currents, electrostatic interference, mechanical effects, and test lead resistance.
Leakage currents. These are currents flowing through paths other than intended signal paths, such as current flowing through insulator materials that are part of interconnections. This leakage current can be a problem when the impedance of the device under test (DUT) is similar to that of the various insulators in the circuit. The simplest fix is to use a high quality cable with high resistance insulation such as Teflon® or polyethylene. Better quality cables often reduce the effects of dielectric absorption, which is typically a function of their high impedance insulation material.
Although high quality cables can go a long way towards reducing leakage currents, they may not always be sufficient. It's important to understand the nature of the leakage path in a shielded coaxial cable when it's used for the connection between a typical DC instrument, such as a source-measure unit (SMU), and the DUT. Leakage current flows from the center conductor to the shield through the cable's insulation resistance. This causes the SMU to measure the sum of the current flowing through the DUT and the leakage current, rather than just the current flowing through the DUT.
A technique called guarding can eliminate the effects of leakage currents flowing through the insulation. A guard is a low-impedance point in the circuit that's at the same potential as the high-impedance lead in the circuit. In a guarded measurement, the shield is driven to the same potential as the Force Hi output terminal of the SMU using a unity-gain, low-impedance amplifier (guard). Therefore, no leakage current flows through the insulation resistance.
This technique requires a third (GUARD) connection on the instrument in addition to the usual cable shield and signal conductor. Although it's theoretically possible to use a guarded connection on a coaxial cable, it would be unsafe because the shield would be at the same potential as the Force Hi output terminal. The GUARD of an SMU should never be connected to the shield of a coaxial cable. The correct solution is to use a triaxial cable. Its inner shield is connected to the GUARD terminal of the instrument, and the outer shield is connected to the Force LO output terminal.
Electrostatic and magnetic interference (EMI). EMI occurs when electrically magnetically charged objects disturb the circuit under test. Although these effects are negligible at low impedances (because the charges dissipate quickly) that doesn't hold true at high impedances. Typically, the sources of EMI are environmental, such as fluorescent lights, motors, and even people. EMI-induced errors are generally due to some type of capacitive coupling into the circuit, which can create an extra current defined by this equation:
I = C(dV/dt) + V(dC/dt)
Errors due to electrostatic coupling can be minimized in two ways. The first is to use shielded triaxial cables. Another really effective way to reduce electrostatic interference is to connect the prober's head plate, dark box, or metal enclosure to ground so the induced current noise will flow through the shield to ground and not through the DUT.
Although grounding the prober's enclosure or head plate is a good way to reduce electrostatic interference, doing it poorly can make matters worse. Typically, the prober and instrument are connected to two separate power line grounds because they are usually connected to separate power outlets. In that setup, a fluctuating current can flow between the instrument and the probe station, causing the instrument's ground unit or low connection to move, producing errors. These are known as ground loops; to prevent them, the prober enclosures and shield should be connected to a common ground in the test circuit.
Mechanical effects. Triboelectric and piezoelectric effects are also significant sources of DC measurement errors. Triboelectric currents are generated due to charges created by friction between a conductor and insulator. The Friction causes free electrons to be rubbed off the conductor, creating a current charge imbalance that generates current flow. This effect is noticeable when low quality cables are flexed. The insulators of high quality cables are usually coated with a lubricating layer of graphite that minimizes friction. Even high quality triax cables need isolation from vibration, which can cause some current flow due to piezoelectric effects. Placing a prober on an air table and using remote pre-amps to minimize triax cable lengths can help minimize these problems.
Errors due to two-wire test lead connections. No matter how good the test cable, losses through it can still be significant. The conductor resistance of most interconnect cables is only a few ohms for even a very long length. However, when the resistance of the DUT is also relatively low (e.g., a metal structure), the resistances contributed by cables and pad contacts can cause large errors if the instrument is connected to the DUT using two-wire connections. For example, if the resistance of the cables, contacts, and DUT are at single ohm levels, then the meter will measure the sum of these resistances.
Fortunately, four-wire or Kelvin connections offer a way to solve the problem of losses due to voltage drops across interconnect cable conductors. In a Kelvin connection arrangement, a voltmeter is connected right at the DUT while a current source stimulates the DUT. An SMU offers both source and measurement capabilities, and can calculate resistance from I-V measurements using two different sets of leads for its current source and voltmeter circuits. With this arrangement, given that the SMU's voltmeter circuit has high input impedance, very little test current will be diverted to this part of the instrument. Therefore, the source current is essentially the same at all points in the test circuit and the voltmeter measures only the voltage drop across the DUT and contact resistance, but not the conductor resistance of the interconnect cabling.
Cabling for C-V measurements. Semiconductor C-V measurements are typically performed using a multi-frequency capacitance meter with four terminals: high current (Hcur), high potential (Hpot), low current (Lcur), and low potential (Lpot). These terminals are coaxial, so they normally require a coaxial interconnect cable, but can also use the appropriate triaxial cable. The instrument terminals also have fixed characteristic impedance, typically 50 or 100 ohms. A multi-frequency capacitance meter measures impedance by sourcing a small AC voltage across the DUT and measuring the resulting AC current and AC voltage.
The signals used to measure capacitance are AC, so take care that the signal paths minimize impedance changes, which will generate signal reflections that can impact the AC source and measured signals, causing erroneous results. Using cables with the same characteristic impedance as the instrument can help minimize impedance changes. Also, most capacitance meters perform some form of cable compensation, which involves entering the length of the cables being used into the meter setup. In addition, all four cables must be the same length to minimize impedance changes in the signal path and ensure the meter's cable compensation calculations are correct. Improperly shielded coaxial cables can also cause errors in capacitance measurements, so check that the cables' shields are not open.
Stray cabling and interconnect inductances can cause resonances in the test circuit, which may also produce erroneous capacitance measurements. Reduce the inductance by connecting the shields of the coaxial cables as closely as possible to the DUT to isolate the inductance from the measurement.
Cabling challenges for ultra-fast DC and pulse measurements. These measurements involve signals in the AC or high frequency domain, which requires different considerations about the interconnect compared with DC or C-V measurements. Fourier series analysis shows us that no matter the speed of a pulse, it can be modeled as a series of sinusoids of varying amplitudes and frequencies. If the signal path can't pass higher frequencies accurately, some fundamental frequencies are affected, distorting the shape of the pulse and affecting the power transmitted to the DUT. Therefore, pulse testing requires a signal path with a higher bandwidth (approximately 150MHz) than either DC or C-V testing.
Pulse generators don't typically include measurement capabilities and oscilloscopes don't include a stimulus source. The test connections on these instruments are single-ended; i.e., the center conductor carries the signal and is connected to the DUT. The shield of the coaxial cable is the return path for the signal and ideally should be connected to the low terminal of the DUT. Today, however, a new class of pulse instruments does offer measurement capabilities, but even these instruments use a single coaxial cable to the probes for each channel. To prevent reflections, the interconnect's impedance should match that of the source. Most pulsing instruments have a 50-ohm input impedance so one normally uses 50-ohm coaxial cables when connecting to them. To minimize inductances, it's a good practice to connect the shields of multiple coax cables together.
Ideally, to ensure the entire signal path is free of impedance changes or insertion losses, the shield of the coaxial cable should be connected to the low side of the test circuit. And while connecting the low side of the DUT to DC ground will work, the return signal may then follow two paths back to the instrument: one through earth and the second through capacitive coupling to the coaxial shield. Connecting the low side of the test circuit to a DC return path will ultimately limit the fidelity of the pulse.
Different measurements - different cabling requirements. As discussed above, each measurement type requires different considerations in the selection of cabling. Up to now, performing DC I-V, C-V, and pulse testing on a DUT required cable changes when moving from one type of measurement to another. This was especially true when a switch matrix was not being used, but switch matrixes present their own set of complications.
Changing cables can be time-consuming and often requires raising and lowering the probes between tests, which risks damaging wafer pads. Fortunately, a new test kit of multi-measurement prober cables (Keithley's MMPC kit) optimized for use with Suss and Cascade probers offers a simple solution to this problem. This cable kit supports I-V, C-V, and pulsed measurements using a single set of triax interconnects. The MMPC kit has special triaxial cables and adaptors that allow both high precision DC and high bandwidth AC connections to the probes and DUT. The cables can be arranged in both Kelvin and non-Kelvin configurations. More importantly, all of this can be done without disturbing the cabling to the probes; only the cables at the prober connector bulkhead need be moved.
Although the MMPC kit was initially designed for use with Keithley's Model 4200-SCS Semiconductor Characterization System, they can be used with many other test instruments designed for wafer level device testing with Suss and Cascade probers. In addition, the Model 4225-RPM Remote Amplifier/Switch, a companion to Keithley's 4225-PMU Ultra-Fast I-V Module, automates reconfiguring MMPC cables when they are used with the Model 4200-SCS.
