Characterizing a high-density, controlled impedance test interface
impedance test interface.
What are the effects of an interface during a test set up? This article analyzes the performance of an Omniprobe-R when it is used as an interface between a package and a board, or a package and a cable. The device proves to be nearly transparent up to the 20 GHz bandwidth of these measurements.
Ball grid array (BGA) packages using solder balls or a Land Grid Array (LGA) with just pads are the standard packaging technology for all devices with more than about 25 pins. They are advancing on the three axes: ball pitch or density, bandwidth of signals, and total ball count.
It’s the chips assembled in these packages that are the driving force. Signal data rates are advancing into the 15 to 28 Gbps and beyond. At these rates, the attenuation and cross talk in typical circuit board traces often dominate the signal quality. Where the signal is picked up is a prime factor influencing the signal quality. The least distorted signals will be at the package to board interface.
One of the hurdles in testing reference designs with these advanced packages and high performance devices is getting a clean, high bandwidth, controlled impedance interconnect to the package launch pad on the PCB at these high densities.
The penalty paid at any interface is a reflection and loss in signal bandwidth. The engineering challenge is to try to minimize this penalty by keeping discontinuities short and managing the return paths as carefully as the signal paths.
Probing the active signals in a reference design with a BGA or LGA can be a challenge when packages move on just one of these axes. The real challenge is when package features advance on all three axes at the same time.
This article tracks the use of a probing technology from Ardent Concepts , Omniprobe-R, which is designed to enable a high bandwidth, compliant interface between very fine 50 ohm coaxial cables and a pad array. Tiny micro-springs called RC Spring Probes, are embedded in a metal housing proving a short and compliant path for signals and adjacent returns (see Figure 1).
Figure 1. Cross section of the Omniprobe-R showing signal and adjacent return paths.
The ends of the coax cable are embedded in a metal housing to provide mechanical support and the connection to the return path.
Testing the interface
We constructed a 2-channel demonstration vehicle to evaluate the electrical properties of this coax-to-interposer interface. Micro-coax cables, 30 mils in diameter, enable a probe-able signal pitch down to 1mm (the demonstration vehicle had signal pads on 1.27mm pitch). The flexible coax provides the geometry transformation from the fine-pitch of the pads to a much coarser pitch of the tester interface (in this case, 2.29 mm SMKs).
In this evaluation, we used a Teledyne LeCroy 4-port SPARQ signal integrity network analyzer to measure the differential channel composed of the micro-coax on either end of the an interposer (see Figure 2).
Figure 2. (Top) Omniprobe-R interposer connected by micro-coax cables into a network analyzer for 4-port measurements and (Bottom) a schematic illustration of the interface.
The RC Spring Probes directly contact the signal conductor of the micro-coax cables. This is a generic approach to interface to many cable types. In this evaluation, in order to eliminate the variables of PCB design, we also connected the back side of the interface to coaxial cables. In normal practice however, the coaxial cables would only be on one side and the other side would be a device interface, such as a pad array on a PCB as in Figure 3.
Figure 3. Omniprobe-R interposer cross-section in application
The 4-port S-parameters of this differential channel were measured. There are 16 S-parameter elements, each composed of a magnitude and phase which can then be transformed into single-ended or differential in the frequency domain or as the single-ended or differential step response or impulse response in the time domain. All together, this is a total of more than 150 different terms, each one of which can have as many as 2,000 frequency or time points. That’s a lot of information. But, not all of it is equally important. With flexibility in manipulating each term, this massive database can be explored to efficiently data mine useful information.
The first step is to look at the single-ended responses:
• Return loss of each end
• Insertion loss of each channel
• Near end cross talk
• Far end cross talk
Figure 4 shows the measured results for the two signal-ended channels in the SMK connector-cable-interposer assembly.
Figure 4. Measured single-ended response for the probe assembly and just the cables. The frequency scale is 20 GHz full scale in each case.
The return losses show less than -20 dB up to 10 GHz, and rising up to -10 dB toward 20 GHz. This is an indication of reflections from discontinuities. Generally, a return loss below -13 dB has less than -0.5 dB impact on the transmitted signal. It’s difficult looking in the frequency domain to determine where the discontinuities occur. Fortunately, the time domain display of the same data can help answer this question.
The lower left graph shows the cross talk terms on a scale of -160 dB full scale. This shows all the crosstalk less than -75 dB up to 20 GHz. This clearly shows these two channels are uncoupled. The channel-to-channel cross talk in the interposer and the cable assembly is negligible. Since the coupling is so small, the single-ended and differential responses will be identical.
The insertion loss for the two single-ended thru paths is shown in the upper right graph on an expanded scale of 8 dB full scale. While there is a general drop off of about 1.5 dB up to 20 GHz, there is a curious dip at about 10 GHz. This does not appear as a comparable reflected signal in the return loss, nor as coupled energy into the adjacent channel in the cross talk term.
Where does the energy go at 10 GHz that is not transmitted? Some insight is gained by looking at the measured insertion loss of a single section of the micro coax, with the same SMK connectors on its ends, but no interposer. The lower right graph is the measured insertion loss for two different micro-coax cable sections, of slightly longer length than used with the interposer. These cable-only sections also show the same dip in the insertion loss at 10 GHz, suggesting it is intrinsic to the cable or the cable-SMK connection, unrelated to the interposer.
One possible cause of this sort of small dip in the insertion loss may be periodic variations in the impedance of the cable from the extrusion manufacturing process. Periodic discontinuities can cause dips in the transmitted signal at series resonant frequencies in the same way a crystal lattice will cause energy band-gaps in the energy of the electrons. For this reason, these dips are sometimes referred to as Bloch Wave Effects, named for Felix Bloch .
The origin of the reflections in the return loss can be explored by taking the time domain response of the same return loss measurements. Figure 5 shows the frequency domain return loss and the same data, transformed to the time domain as the step response. The step response was further transformed into an impedance profile.
Figure 5. Measured return losses in the frequency domain and transformed into the step response and then impedance profiles. The rise time for the step response was 50 psec.
In this example, the rise time of the step response was 50 psec. At this resolution, the discontinuities at the launches between the SMK connectors and the micro-coax cables are seen on either end as well as the small difference in impedance of the four sections of cable. The compliant Omniprobe-R interposer at the center of the cables is a very slight discontinuity, almost masked by the impedance variations between coax cable segments.
>>Page Two: De-embedding the launches and cables on EDN.com