By Dr. Eric Bogatin
www.BeTheSignal.com
January 2007
What do you get when you cross precision time domain reflectometry (TDR), high bandwidth microprobes, and a CAD database-driven, multi-head flying prober? Answer: You get the Introbotics Automated Controlled Impedance Tester, shown in Figure 1. This system enables automated, accurate measurement of the important high-speed electrical properties of interconnects, across an entire board, at speeds of 800 to 2000 measurements an hour.
Figure 1. The Introbotics CI1000 Automated Controlled Impedance Tester
With the precision of a commercially available TDR front end and a 25-psec rise time, this system is capable of characteristic impedance measurements repeatable to +/- 0.05 Ohms and time delay measurements repeatable to +/- 4 psecs. What can you do with this kind of information?
“We can provide an OEM with the complete, high-speed performance process capability of a fab supplier," Brian Butler, president of Introbotics, said. “Our goal is to offer a system with metrology capability for characteristic impedance and time delay with high throughput measurements.”
In addition to single-ended impedance, a dual channel TDR also enables differential impedance measurement and line-to-line skew in a differential pair and checks for imbalance between lines in a differential pair. With the measured time delay and the length of the line values extracted from the database, you can calculate the propagation speed and from this, the effective dielectric constant for each layer.
Introbotics sells the automated measurement system, with five fully automated and five manual systems currently deployed in the field, and performs contract measurement services. “We typically are asked to evaluate the variation in performance across a board, from layer to layer and from board to board,” Butler said. “And yes, we see a difference between the measurements on a test coupon on the periphery of a board compared to lines embedded in the board.” This is why the largest OEMs, such as IBM, Cisco, and Intel are customers.
With as many as 40 tested lines per layer, a 26 layer board can have over 400 measurements. Across 10 boards, this is over 4,000 measurements. Figure 2 shows an example of the distribution of measured impedances for 3,380 similar line width traces across a series of test boards.
Figure 2. Measured Characteristic Impedance Distribution
According to Butler, the chief advantages of automated testing driven by a CAD database are reproducible measurements that are not operator dependent or subject to operator error and that there is virtually no limit to how many measurements can be taken, in any location, on the board.
Butler says the data they collect on customer products is used for vendor capability studies, design and simulation validation, materials analysis, and interconnect bandwidth verification of specific lines.
Working with the largest OEMs and measuring most of their leading edge, high-speed boards, Butler sees first hand the demands on interconnect performance. Until recently, he saw downward pressure on reducing the impedance tolerances from +/- 10 to +/-8 percent and even to +/- 5 percent. However, he says the trend has not accelerated, and he now sees less concern on tightened impedance tolerance and a growing focus on controlling and tracking losses in the lines.
Introbotics will present a paper at the upcoming IPC Expo conference on a technique they’ve developed to use the rise time of the transmitted signal as a measure of the interconnect bandwidth of selected traces. They translate the 20 to 80 rise time of the transmitted signal, compensating for the system rise time, into an interconnect bandwidth. Butler says this is a good relative measure of the line losses, a growing concern to all backplane designers.
The second area of growing concern Butler sees in high-performance interconnect systems is the intra-line skew in differential pairs. As Intel, Teraspeed, and Teradyne (now Amphenol) pointed out, the glass weave in traditional laminates can cause a local variation in the effective dielectric constant each line sees, resulting in a different time delay for each line in a pair. This skew causes mode conversion and anomalous loss of the differential signal. The ability to measure intra-line skew to within 4 psecs for all the lines in a backplane has become an important product performance test.
As high-speed serial links proliferate in cost-driven systems, design margins tighten. The ability to quantify the impedance, time delay, skew, and loss distributions across manufacturing variations will become increasingly important in the delicate balancing act of acceptable performance for the lowest cost.
This and other signal integrity topics are covered in Eric’s public classes and online lectures, available from his website, www.BeTheSignal.com. Send your signal integrity technical questions to DoctorIsIn@BeTheSignal.com
Reader Q & A
If you have a question for the SI Doctor, please send it to DoctorIsIn@bethesignal.com.
Question from Sainath in Austin, TX: Dr. Bogatin, if I want to "be the bit (pulse)," what do I experience as I propagate down the metal trace?
Answer: Good for you for thinking about what the bit would see by "being the signal!" Keep in mind that the “signal” is the voltage between the two signal lines that make up the pair. The differential impedance is really the instantaneous impedance this difference voltage sees. It is the series combination of the impedance between each line and the return path. As long as the geometry is well designed so the instantaneous differential impedance is constant, the differential signal just sees the losses in the line and no reflections.
Question from Robert in Germany: Where does one get the S-parameters from?
Answer: The S-parameters can be obtained by simulation of the differential channel using a tool like Agilent's ADS, Mentor's Hyperlynx, or Synopsis HSPICE, which each have an integrated 2D field solver, or by measurement from the ends of the channel using a four-port VNA or TDR. Whether the S-parameters come from a time domain measurement or a frequency domain measurement, they can be converted between the two. Likewise, whether the measurement is made as single ended or differential, they can easily be converted between the two forms.
Bio: Eric Bogatin is president of Bogatin Enterprises. Many of his papers are available on his website, www.BeTheSignal.com. He is the author of Signal Integrity - Simplified, published by Prentice Hall.
