"At 28 Gbits/sec everything matters," says Dave Dunham, Director of signal integrity at Molex. Even above 5 Gbits/sec, interconnects are no longer transparent. The beautiful, clean, pristine signals from the pads of the die get distorted, attenuated and turned to mush by IC packages, circuit board traces, connectors and cables.
Limitations from the fundamental physics of interconnect losses can't be eliminated. The only practical way around this limitation in the interconnects is to implement significant signal processing on-die at the transmitter in the form of FFE (feed forward equalisation) and on-die at the receiver in the form of CTLE (continuous time linear equalisation) and DFE (decision feedback equalisation).
Interconnects can distort a signal so much that at the receiver's pads, an eye diagram can be completely closed. On-die signal processing can open the eye. Figure 1 shows an example of the measured real time signal and resulting eye as measured at the pads of the receiver package, before the on-die equalisation has a chance to clean it up.
Figure 1. A eye diagram at the pads of a receiver package before any equalisation shows how interconnects can distort the signal and close the eye.
Upon looking at the measured real-time signals, you can't determine if on-die equalisation circuitry can clean the eye, recover the embedded clock and interpret the signal at an acceptable bit error ratio. Both the test and measurement industry and the semiconductor transceiver providers have responded to this challenge with innovative solutions, borrowing a little from each other.
As a result, oscilloscopes from the top test and measurement instrument providers—Agilent Technologies, Rhode Schwartz, Tektronix, and Teledyne LeCroy—have implemented the same equalisation techniques and CDR (clock data recovery) algorithms as found in typical receivers, to emulate what the receiver might see.
"The actual signal on the pads of the receiver chip can be so distorted above 5 Gbits/sec that we have to clean it up before we can even say if it is good or bad," said Alan Blankman, product manager for serial data analyser products at Teledyne LeCroy. "We're careful to use the same algorithms in processing the signals measured by the scope as used in the receiver circuitry itself. This gives the engineer a realistic view of what the receiver would actually see."
An on-die CTLE filter is defined by a few simple pole-zero parameters. These can be selected based on the actual values used by the receiver, or optimised with software resident in the scope to maximise the eye opening. Figure 2 shows an example of the impact from the CTLE filter in the scope on the measured eye before and after the filter is applied.
Figure 2. Measured eye at the input to the CTLE filter implemented in the scope and the resulting eye.
After the CDR algorithm, the same DFE filter on-die is implemented in the scope to replicated the final eye opening that would appear at the receiver. This filter will take a fraction of a few previous bits and add them to a later bit. The fraction of each prior bit is the "tap coefficient." The coefficient values used in the scope can be selected as the same ones used on-die, or optimised based on the maximum eye opening. Figure 3 shows an example of the impact on the measured eye after the CDR and DFE filter are applied in the scope.
Figure 3. Final signal emulated by the scope based on the CTLE, CDR and DFE features that would be implemented on-die at the RX.
next; scope functions on-the-die