Historically, combining both high-speed AC and precision DC measurement capability within an automated test system required mechanical relays to keep the two subsystems from interfering with each other. Even with improvements in mechanical relays, it still is desirable to eliminate as many relays as possible from a test system to maximize circuit density and minimize test costs.
To eliminate the relay used to disconnect the AC pin electronics circuitry, the pin-driver IC must have low input leakage current over the full voltage range of the DC measurement system. This has been realized for pin drivers operating as fast as 500 MHz.
To remove the relay that disconnects the DC measurement system, the per-pin DC measurement unit (PPMU) IC and associated circuitry must be designed to preserve the AC waveform passing between the pin driver/comparator and the device under test (DUT). The primary cause of distortion in the rising and falling edges of the AC measurement system has been the parasitic capacitance presented by the DC circuitry. This distortion has been difficult to correct, especially at fast edge rates.
One solution compensates for the lumped capacitance of the PPMU by adding series inductors into the circuit to approximate a transmission line as shown in the upper circuit of Figure 1. For this approach to work at high frequencies, the parasitic capacitance of the DC measurement unit and the trace length to the transmission line must be minimized.
In a practical system, it is difficult to reduce the capacitance of the PPMU to less than 10 pF or the trace length between the PPMU and the transmission line to much less than 1 inch. Together, these factors limit the maximum frequency at which this approach will work to approximately 200 MHz. To go beyond this frequency, either a switch must be inserted to remove the PPMU from the circuit or a completely different approach used.
An alternative to keep the PPMU capacitance from affecting the AC performance uses a high-impedance ferrite in series with the PPMU circuit as depicted in the lower circuit of Figure 1. This approach is not intended to match the transmission-line impedance but to isolate the capacitance from the AC signal. Since a ferrite primarily turns the high-frequency component of an incoming signal into heat rather than storing and releasing the energy like an inductor, this component distorts the waveform very little. In fact, it creates less distortion than inductor-based compensation techniques.
Specifically, ferrite isolation does not slow down the slew rates of the signals going to and from the DUT, even with rise/fall times down to 400 ps. This feature extends the usefulness of relay-free test systems to above 500 MHz.
Figure 2 illustrates the effect a 15-pF parasitic capacitor, combined with different compensation schemes, has on a signal with fast rise and fall times. The smaller effect on fast edge rates of the ferrite circuit not only allows higher frequency operation, but also permits more accurate DUT rise/fall time measurements. This is possible since the rise and fall times of the driver signals and DUT are equally affected by the parasitic capacitance and compensation network.
The primary disadvantage of the ferrite approach is the occurrence of measurable amplitude error with wide pulse-width signals. This effect is similar to RC time constant type behavior if the ferrite is replaced with a resistor.
It is somewhat different, however, since the dissipative characteristic of a ferrite varies with frequency, and ferrite impedance also has an inductive component. Nevertheless, calculating the amplitude error as if the ferrite were an ideal resistor gives a close approximation to the characteristics of the distortion.
For instance, a 600-W ferrite isolating a 15-pF capacitance from a 50-W transmission line should give a worst-case amplitude error of about 8% (50 W/600 W + 50 W). Figure 3 depicts amplitude errors using various ferrite configurations to isolate a 15-pF capacitance.
As shown, the amplitude errors are not significantly improved when placing multiple small ferrites in series. This primarily is due to the rapid drop-off of the absorptive component of the impedance at lower frequencies, making the effect of the inductive component more significant. Increasing ferrite impedance tends to increase the amount of ringing in the signal at low frequencies, which generally is not desirable.
Larger ferrites, such as 1206, have less inductance than the smaller parts, so some advantage is obtained from using larger components; however, this is at the expense of taking up more board space. Using a combination of small (0603 or even 0402) ferrites in series with larger ferrites, such as a 1206, gives a good combination of reduced amplitude error and ringing with a minimum amount of board spaced.
No matter what combination of ferrites is used, the rising and falling edges are not distorted at all, giving the best possible timing accuracy. With ferrites, the amplitude error is consistent and often can be corrected in software. For that reason, the best amplitude accuracy will be obtained in situations where AC measurements are obtained with relatively narrow (and constant) pulse widths, such as when testing memory devices.
Testing microprocessors or random logic can require wider pulse widths. In these situations, signal amplitude errors always will be present to some degree. Since the critical characteristic of AC testing is edge accuracy, amplitude errors usually are less significant than those caused by other approaches or other portions of the test system.
The increased ringing caused by adding ferrites in series underscores the disadvantages of using ordinary inductors in this fashion. Since an ideal inductor does not dissipate energy but instead stores and then releases it, inserting an inductor large enough to present a 600-W impedance to the slowest rise/fall time signals of interest will greatly distort fast waveforms.
For example, to get a 600-W impedance at 200 MHz requires an inductance of about 500 nH. The 500-nH inductance combined with the 15-pF load capacitance gives a resonant frequency of 60 MHz. This is within the range of expected operating frequencies and could easily cause major errors in signal amplitudes.
With ferrite isolation, even a large amount of parasitic capacitance on the PPMU node does not affect the edge rates. While the lower-frequency waveform characteristics are changed by a larger capacitance, the critical waveform behavior is not altered. This allows PPMUs with higher drive capacity and correspondingly higher output capacitance to be used in ATE without relays.
With the rapidly declining impedance characteristic of ferrite and the low frequencies at which PPMUs operate (typically <1 MHz), the ferrite impedance is insignificant, usually less than 1 W. If the maximum PPMU output current is 2 mA, this causes a worst-case amplitude error of less than 2 mV, which is acceptable in most digital test systems.In cases where the maximum PPMU current is higher or where increased accuracy is needed, the sense line can be connected separately from the force output using a second ferrite or a resistor since the current into the sense pin is normally nanoamps or less. If a resistor is used, ensure that the value is high enough to avoid loading the transmission line during transitions and low enough so that the RC time constant of the resistor and the PPMU sense node capacitance are small relative to the PPMU response time. Resistor values from 5 kW to 50 kW typically meet these requirements.
Also, the stub length presented to the transmission line by a 0603 or 0402 ferrite can be less than half that of a coaxial reed relay. The stub length is the total length of the conductor along which a high-speed signal will travel before being reflected back to the transmission line.
The stub from a 0402 ferrite typically is less than 0.08 inch in length. Since the impedance of the first ferrite in a series is more than 10 times the transmission-line impedance at 500 MHz, the stub length to any additional series ferrites has negligible effect on the reflection characteristics.
In comparison, the stub from a reed relay with good high-frequency characteristics typically is 0.2 inch, causing longer time delays in reflected signals that, in turn, prompt more distortion of fast pulse edges. For that reason, a ferrite-based system actually can allow more accurate AC measurements than a relay-based system while permitting smaller guard bands and higher yield, further reducing the effective cost of test.
About the Author
Thomas Bradley is the senior hardware engineer for test and measurement products at Semtech. He has been involved in testing semiconductor devices for more than 20 years, designing test and data acquisition circuits for nearly as long. Mr. Bradley received a B.S. in engineering science from New Jersey Institute of Technology and completed graduate-course work in electrical engineering at California State University, Fullerton. Semtech, Test and Measurement Products, 10021 Willow Creek Rd., San Diego, CA 92131, 858-547-6622, e-mail: [email protected]
Return to EE Home Page
Published by EE-Evaluation Engineering
All contents © 2003 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.0
May 2003