In June 2006, the IEEE committee ratified the 10GBaseT specification (IEEE 802.3an) establishing the standard for running 10-Gbit/s data throughput over CAT6 or better unshielded twisted-pair copper cabling. This standard gave IC houses and original equipment manufacturers (OEMs) a low-cost alternative to fiber. Designers started to revisit the platforms they were using as they developed 10-Gbit/s solutions.
Clearly, a 10-Gbit (10G) system operates at 10 times the speed of its 1G predecessor and covers a wider frequency spectrum. Yet this transition from 1G to 10G is a challenge in itself. Factors such as heat and size have already been addressed, but there are other issues to consider. Designers cannot simply replace a 1G active output interface (AOI) module with a 10G module and expect everything to work.
A wider transmission bandwidth is required, necessitating the use of new components such as magnetics and silicon that can perform over a wider spectrum. New magnetic components are needed to maximize the performance gains of a 10G port.
For instance, as frequency increases with more complex coding schemes, parasitics such as leakage inductance and distributed capacitance, which affects insertion loss and return loss, may cause the 10G port to no longer meet the required performance standards. Those factors have to be adjusted in a 10G Ethernet solution.
THE INTERFERENCE CHALLENGE
Perhaps one of the biggest challenges in 10G today is the interference present in many data centers. Circuit boards in data centers require 10G because they must handle much higher data rates. Their frequency range stretches up to 500 MHz or 600 MHz.
The lowest band from cell phones operates near the same frequency range. The frequencies used by cell phones and two-way radios coincide with some of the frequencies used by 10G chips, creating unwanted interference.
For example, if a person is using a cell phone in a lab or in close proximity to the switch rack, the signal-to-noise ratio (SNR) goes so low that it causes erroneous data transmission. Two-way radios can cause such a spike that they create havoc far worse than that of cell phones.
The interference problem brought by these devices triggered such attention that 10G system and physical-layer (PHY) vendors began to look for ways to rectify the problem. Upon ratification of the standard, no one really knew of the system’s vulnerability to cell-phone interference.
One proposed remedy is to implement the use of low-pass filters. Nevertheless, it is easy to deploy and has proven to be useful thus far. Note that the main signal spectrum for 10G stops just slightly above 400 MHz. However, extra bandwidth is required for 10G to improve SNR. This is the reason for using a low-pass filter: to filter out cell-phone noise at greater than 800 MHz.
The low-pass filter sets the bandwidth desired for operation and limits or blocks interfering frequencies from other devices in the area. Although low-pass filters remedy the interference problem, they also create new challenges for circuit board design and magnetic components.
CONQUERING THE SOLUTION
Magnetic modules (Fig. 1) that integrate a transformer and a low-pass filter within a single device offer significant advantages to designers of 10G systems. Once the transformer and filter combine with the PHY, it becomes a turnkey system for OEMs to create their switching systems and servers.
One major advantage of integrating the transformer and low-pass filter into one component is the superior performance, especially in relation to meeting key parameters such as return loss (S11) and insertion loss (S21). With less return loss, the signal can travel farther with little to no problem.
While it may seem that a perfect overall performance solution is achievable by putting a transformer next to a well-matched low-pass filter, in reality this is not the case. Putting them in close proximity helps with interference, but it does not ensure that the system can meet the return loss requirements caused by the interaction between them.
When the transformer and low-pass filter are in series, a signal has to pass through more components. Thus, there will be an increase in the number of traces potentially resulting in mismatched impedance and more parasitics that could further result in reduced performance. Parasitics differ on each board, and a small change in parasitics makes a big difference in a system’s performance. Traces must be short and the impedance requires perfect matching from beginning to end. Essentially, everything must balance well.