One of the bottlenecks in Ethernet switch production, however, is the need to perform Media-Dependent Interface (MDI) tests. Typically, manufacturers use in-circuit tests (ICTs) to verify the functionality of connectors, transformers, resistors, and other components in a module path. To ensure proper performance, ICT must be performed at several points along the path, and the switch manufacturer must write ICT test programs for each product. This becomes a particularly cumbersome procedure in high-density switch production.

Currently, it is possible to perform MDI testing using the 10GBASE-T PHY itself. Silicon manufacturers are enabling the PHY's transmit and receive functionality to transmit a pattern and predict the path performance based on what it receives in return. This new capability eliminates the need to use ICT or to write ICT test plans, and it reduces the process of component verification to a few seconds per module. In fact, almost all of the available 10GBASE-T products in production have the MDI test hardware and software included in the PHY, including multi-port network interface controllers (NICs) and 16, 24 and 48 port 10GBASE-T switches available from top tier networking companies. This enables equipment manufacturers to increase final yield while decreasing test time and delivering a much higher quality product.

Why have an alternative to ICT?
Every network equipment manufacturer wants to meet specifications with every product and to ensure reliability and performance. This requires testing for component-dependent and component-independent problems. ICT has been a useful testing tool for many years, but with the emergence of such higher-speed protocols as 10GBASE-T it presents several challenges, and most switch and NIC makers avoid using ICT in the MDI path when testing 10GBASE-T products.

To perform ICT, manufacturers must design probe points into the circuit board. This can create a less-than-optimum layout that can have several negative consequences:

• Extra complexity--adding probe points adds to layout complexity and creates a series of sensitive RF areas in the circuit. Since RF is very sensitive, it takes longer to complete the designs and the RF points add capacitance to the circuit. In fact, some manufacturers deliberately stay away from adding probe points because it can have too great an impact on the circuit's performance.
• Reduced bandwidth--because adding probe points increases capacitance, it reduces bandwidth.
• Worse insertion loss and return loss--the antenna stubs used for probes create discontinuities that increase the potential for return loss problems. A circuit trace should have the fewest possible return loss issues, but antenna stubs can cause reflections that work against this goal. While return loss is a much more significant issue, adding probe points can also impact insertion loss under certain conditions because the added probe points can increase the effective capacitance of some components.

Ultimately, these impacts combine to reduce the product's overall performance.

Another drawback to ICT is that it is limited in scope. Traditional ICT probes only test performance between two nodes on a circuit board. The manufacturer ends up performing many discrete, node-to-node tests that produce static values, which increases manufacturing time and costs. Moreover, ICT doesn't always provide a clear view of the circuit's overall performance--it verifies that certain nodes are performing to specifications, but not necessarily that the end-to-end circuit performs as it should.

Due to these issues, many vendors previously skipped ICT and instead performed functional testing at a later stage of the manufacturing process. Unfortunately, this leads to less efficient manufacturing because performance issues that could have been detected at the board stage are not discovered until much later.

Obviously, it is much better to discover performance issues as early as possible in the manufacturing process, rather than investing a lot of time and material in manufacturing only to find out that the product does not meet specifications. What manufacturers need is a way to perform early-stage testing that does not result in ICT issues.

Enabling MDI Testing in PHY
10GBASE-T PHYs incorporate advanced analog and digital signal processing, so that the PHY itself is uniquely capable of performing test and measurement functions during manufacturing. By initiating the use of MDI testing in the PHY new ground in enhancing 10GBASE-T system performance and reducing overall system costs has been broken.

Under software control, the 10GBASE-T PHY can operate as an oscilloscope, a spectrum analyzer, and a network analyzer. This presents opportunities for a different kind of early-stage circuit testing. However, while some basic signal processing capabilities are present in every 10GBASE-T PHY, the PHY manufacturer must implement features and design software that allows manufacturers to perform MDI testing. MDI tests are faster and more accurate than ICT because they test an entire circuit rather than performance between two discrete nodes on a board. The following are sample MDI tests showing various circuit characteristics.

For example, a circuit with a missing capacitor on Lane 2 (Figure 1) would cause a detrimental change in return loss which can easily be detected by the MDI test. In addition, failures such as open/shorted inductors, capacitors, and resistors are common problems in manufacturing assembly that will be detected by the MDI tests.

There are many times where the wrong value of a discrete component is installed. The MDI tests detect an incorrect valued component that has been installed since the IC side return loss of the circuit changes with component value. Other detectable failures by the MDI tests include a broken or mis-installed transformer or RJ-45, as well as poor PCB level connectivity in the entire MDI path.

Another detectable problem that has plagued manufacturing is cracked or intermittently cracked inductors, which appear as open or partially open in the circuit. The MDI tests coupled with a screen versus temperature expose partially cracked inductors since the cracking level can vary with temperature. This powerful combination can expose early failures that may previously have escaped test.

Return loss/non-linearity tests are performed by using the 10GBASE-T PHY as a network analyzer, effectively computing the return loss that the 10GBASE-T PHY sees by transmitting a specialized pattern using the PHY transmitter and measuring the signal that returns into the 10GBASE-T PHY receiver. Using frequency domain techniques, the return loss is represented in the frequency domain, which is shown in Figure 2 with the x-axis being frequency and the y-axis in dBm/Hz.

The green plot shows the frequency domain return loss for a lane that is manufactured correctly, whereas the red plot shows the frequency domain return loss for a lane with a missing capacitor. Clearly, the difference between the correctly manufactured lane (lane 3) and the missing capacitor lane (lane 2) can be detected. When the return loss is integrated over frequency, the difference between these lanes will be greater than 10dB, which is easily detected at manufacturing.

Similar results are obtained for the other mentioned cases (open/short components, etc.). In addition, the non-linearity of the entire circuit can be measured versus frequency by examining products of the transmitted signal that are received. This can clearly be seen in the differences between the blue and black plots in Figure 2.

Another potential problem during manufacturing is coupling between the pairs of the MDI in the form of short or partial shorts. If undetected, this coupling will limit performance since cross talk cancellers that are built into the 10GBASE-T PHY will become overwhelmed and will not perform well enough to attain maximum cable reach. Since a 10GBASE-T PHY has four independent transmitters and receivers, cross talk testing can be performed to detect this coupling.

Figure 3 shows a single receiver lane (lane 1) with the cross talk from the other three transmit lanes (lanes 0, 2, 3) shown versus frequency. In this scenario, a partial short is inserted between lanes 1 and 3. From the figure, it can be seen that the cross talk from lanes 0 and 2 is at much lower levels than the crosstalk from the transmitter on lane 3, which makes the fault easily detectable. As previously mentioned, these results can be integrated versus frequency to summarize and compare to a pass/fail threshold.

While incorrectly installed components in the MDI and shorts between MDI lanes are detected with these methods, additive effects or shorts between ports may still not be detected. Since 10GBASE-T is a very challenging communication system, these effects cannot be ignored. By nature, the receiver of the 10GBASE-T PHY can be made to be quite sensitive, detecting low-level effects that restrict performance. Thus, the 10GBase-T receiver can be made to act as a spectrum analyzer, detecting additive problems that can come through the power supply network or coupled on the board. In addition, if neighboring PHYs transmit while this test is being performed, alien crosstalk can be detected on the victim PHY, thus detecting potential shorts between 10GBASE-T ports.

For example, Figure 4 shows an actual narrow band interferer that can easily be detected by the PHY. The example interferer turned out to be processor clock on the board that was coupling into the MDI through the power supply, due to missing de-coupling capacitors on the PCB. The MDI test was able to detect this and the problem was easily found and fixed.

Manufacturers certify that 10GBase-T ports can deliver reliable performance over certain cable lengths (i.e., 100 meters), but not every board's performance is identical. By testing for the above criteria with MDI, manufacturers can get an extremely accurate picture of a switch or NIC port's performance in a very short time and thereby perform accurate pass/fail testing early in the production process.

Benefits of MDI testing
MDI testing is particularly useful in the production of 16-, 24-, or 48-port 10GBASE-T switches because every port must work properly and it is necessary to test them all. Switch makers need a way to identify performance issues quickly in order to avoid lengthy debugging. MDI optimizes performance assurance in manufacturing as follows:

• It tests an entire circuit at once, enabling manufacturers to perform testing as much as ten times more quickly than with ICT.
• It works early in the production process, eliminating wasted time and effort that occur when functional testing is used later in the production cycle.
• It is extremely accurate and provides an assessment that the circuit works, not just that the circuit's components work and were properly installed.
• It eliminates overly complex board designs and the potential for reduced performance that occur when ICT probe points are used.

The future of MDI testing
Although Teranetics pioneered the use of MDI testing in 10GBASE-T PHY products, most or all PHY manufacturers will incorporate this capability into their offerings in the 40nm designs that will begin shipping in the latter half of this 2010. By designing PHY to simplify and reduce risk in high-density switch production with MDI, 10GBASE-T PHY manufacturers will make it much simpler, more cost-effective, and less risky for vendors to produce high-density switches and improve data center performance while reducing capital costs.

About the Authors
John Dring is the Senior Director of Hardware and Software Architecture at Teranetics. He can be reached at: jdring@teranetics.com

Jose Tellado is the Vice President of Systems Engineering at Teranetics. He can be reached at: jtellado@teranetics.com
.