Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing

This article will focus on the RF switch configuration required for x16 testing. These switch models will support up to 18 channels (PCIe is generally up to x16), and will also support a lower number of channels. It is recommended to use a hard cable to establish a fixed connection between different switch components. The hard cable can be obtained from Mini-Circuits. The CEM test chart was given earlier in this article, but these techniques are also applicable to BASE testing.

Author: Tektronix

The comprehensive characterization of high-speed links requires transmitter (Tx) and receiver (Rx) measurements to be performed through multiple different paths of the link under test, which brings challenges to the fully automated test environment. The lane widths of PCI Express ports are generally x1, x4, x8, and x16, which poses challenges for fully automated Tx or Rx testing. By including the RF switch in the test channel, we can achieve multi-channel testing without excessively changing the DUT and test equipment cables. In order to minimize the electrical impact of the RF switch, ensure that the test is true to the specification requirements or verification test plan. This article describes the use of Mini-Circuits RF switches for Gen5 (32 GT/s) multi-channel testing, and provides some overall guidelines for setup and automatic testing, and provides suggestions for common challenges.

This article will focus on the RF switch configuration required for x16 testing. These switch models will support up to 18 channels (PCIe is generally up to x16), and will also support a lower number of channels. It is recommended to use a hard cable to establish a fixed connection between different switch components. The hard cable can be obtained from Mini-Circuits. The CEM test chart was given earlier in this article, but these techniques are also applicable to BASE testing.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 1: ZTM2-8SP6T-40. Figure 2: ZT-8SP6T-40 4U/5U.

Figure 1 shows the ZTM2-8SP6T-40 modular switch matrix, with 8 SP6T mechanical switches terminated at 40GHz. This configuration will support up to 18 channels. It is recommended to use phase-matched cables to establish a fixed connection between adjacent 40 GHz relays. When the relay is not turned on for the through connection, there will be a 50W termination.

Figure 2 shows the ZT-8SP6T-40 4U/5U switch matrix, with 8 SP6T mechanical switches terminated at 40GHz. This configuration will support up to 18 channels. It is recommended to use a hard cable to establish a fixed connection between adjacent 40 GHz relays. The orientation of the switch components in this matrix maintains a similar electrical path length between all inputs and outputs. This is particularly attractive for multi-channel Rx testing to minimize the path-to-path difference between calibration and testing. When the relay is not turned on for the through connection, there will be a 50W termination.

RF switch matrix C Gen5 Tx test

PCIe Gen5 devices (system host or plug-in) will show different transmitter performance in multiple ports. It is common to verify all paths in order to fully characterize the link and identify issues such as silicon performance, high near-end or far-end crosstalk, and wiring defects. Using the RF switch (Figure 3) in the test setup can achieve multi-channel Tx verification without the need for engineers or technicians to constantly change the connection. The connection for the 32 GT/s Base Tx test (see Figure 10) is similar.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 3: 32 GT/s CEM system transmitter (multi-channel).

The system host configuration requires that a compliance test load circuit board (CLB) be inserted into the CEM connector of the DUT, and a cable is required to connect the RF switch from each path. The plug-in configuration is similar, but the DUT is plugged into the conformance test basic circuit board (CBB). A pair of cables connect the terminated switch matrix back to the 50 GHz oscilloscope. An instrument such as an arbitrary waveform generator (AFG) can automatically generate the required 100MHz burst signal, allowing the DUT to cycle through different transmitters to measure various data rates and code patterns.

Every connection made in the switch setting is very important. Due to insertion loss, it is not recommended to connect more than two relays in series for 32 GT/s Tx test. It is recommended to use a 1 meter 2.92 mm cable between the DUT and the RF switch, and a 0.5 meter 2.92 mm cable between the RF switch and the oscilloscope input. You can use the oscilloscope differential fast edge, with TekExpress software to perform automatic channel-to-channel delay correction. The matching range of all cables, relays and PCB in the channel should fall within +/- 1ps of the positive and negative signal path.

Maintaining the 50W (100W differential) connection input/output of the RF switch will minimize the reflection inside the channel, but will introduce some insertion loss. 32 GT/s signal quality test does not require a physical variable ISI circuit board (required for Gen4 test), so it requires embedding additional channels and packaging losses on the oscilloscope. The test fixture characterization (described in Appendix B of the 5.0 PHY Test Specification) should be performed, including the RF switch. Basically, a filter file with a lower loss will be selected to achieve the worst-case plug-in loss (when testing the system host) or the worst-case system loss (when testing the plug-in). You can use the Tektronix SignalCorrect solution to verify the channel loss, including the RF switch matrix, instead of using an expensive VNA.

The de-embedding technology based on scattering parameters (S-parameters) can be used to remove the influence of the insertion loss of the RF switch. De-embedding leads to increased complexity, but improves accuracy. In addition, the impact of noise amplification must be considered. If a phase-matched cable is used between the relay to the relay connection, there is only a small electrical difference from channel to channel. If you feel that these differences will affect the measurement, you can consider customizing the channel S parameter file. You can use SignalCorrect or Vector Network Analyzer (VNA) to capture the S parameter file, and the nominal S parameter file can also be provided by the Tektronix field project team.

RF switch matrix C Gen5 Rx test

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 4: 32 GT/s CEM Rx test points.

The PCIe Gen5 device (system or plug-in) receiver uses a finely calibrated stressed eye signal for testing. This “worst case” signal is established on the reference plane (no channel) through multiple calibration steps, and the “worst case” channel used must be 34 dB ~ 37 dB @ 16 GHz. This section will discuss how to use the terminated RF switch in this signal calibration during Rx testing, and then test the DUT through multiple links.

Calibrating amplitude, Tx equalization, random jitter and sinusoidal jitter at the TP3 test point requires direct connection of Anritsu MP1900A BERT PPG and Tektronix 50 GHz oscilloscope. It is recommended to use 1 meter 2.92mm cable (such as Tektronix product number: PMCABLE1M) to complete this connection. Figure 5 shows the TP3 calibration connection. The RF switch is not included in this step. Because the RF switch introduces some electrical channel differences, it is recommended not to include this effect in front of the TP3 reference plane.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 5: 32 GT/s TP3 pressure eye diagram (basic and CEM).

In TP2P, use differential mode interference (DMI), common mode interference (CMI) and the final pressure eye diagram to calibrate the crosstalk term. This test point comes after TP2 (the physical channel between the BERT and the oscilloscope), but TP2P includes the effects of package embedding and Rx equalization and clock recovery. Figure 6 adds an RF switch to the TP2 calibration, where the switch is introduced after the test fixture (basic or CEM). At this point, the engineer must determine whether a single TP2 calibration is required (recommended for ZT-8SP6T-40 4U/5U) or more than two TP2 calibrations (it is better to consider the different electrical path lengths of the ZTM2-8SP6T-40 ). It is not recommended to cascade more than two relays for 32 GT/s pressure eye calibration.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Table 6: 32 GT/s TP2 pressure eye diagram.

It is recommended to use a 1 meter 2.92mm cable between the BERT and the RF switch, and a shorter 0.5 meter 2.92mm cable between the RF switch and the oscilloscope. You can use the oscilloscope differential fast edge, with TekExpress software to perform automatic channel-to-channel delay correction. The matching range of all cables, relays and PCB in the channel should fall within +/- 1ps of the positive and negative signal path.

PCI Express Gen5: automatic multipath test

Maintaining the 50W (100W differential) connection input/output of the RF switch will minimize the reflection inside the channel, but will introduce some insertion loss. 32 GT/s signal quality test does not require a physical variable ISI circuit board (required for Gen4 test), so it requires embedding additional channels and packaging losses on the oscilloscope. The test fixture characterization should be performed, including the RF switch. Basically, a filter file with a lower loss will be selected to achieve the worst-case plug-in loss (when testing the system host) or the worst-case system loss (when testing the plug-in). You can use the Tektronix SignalCorrect solution to verify the channel loss, including the RF switch matrix, instead of using an expensive VNA.

The de-embedding technology based on scattering parameters (S-parameters) can be used to remove the influence of the insertion loss of the RF switch. De-embedding leads to increased complexity, but improves accuracy. In addition, the impact of noise amplification must be considered. If a phase-matched cable is used between the relay to the relay connection, there is only a small electrical difference from channel to channel. If you feel that these differences will affect the measurement, you can consider customizing the channel S parameter file. You can use SignalCorrect or Vector Network Analyzer (VNA) to capture the S parameter file, and the nominal S parameter file can also be provided by the Tektronix field project team.

Using calibrated 32 GT/s pressure eye diagram signals in multiple channels for receiver testing requires two RF switch matrices, as shown in Figure 7. When the link is x8 or lower, a single RF switch matrix can be considered. The signal from the Anritsu MP1900A PPG must be distributed to all PCIe channels. The device will be in loopback mode, so the digitized signal will be passed back through the Tx pin, switching back to the single input of the BERT error detector. Many systems that support 32 GT/s will exhibit a high loss return path to the error detector, requiring an external redriver to equalize the signal so that it can be detected by the test equipment.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 7: 32 GT/s system Rx LEQ test (multi-channel).

It is recommended to use a 1 meter 2.92mm cable between the BERT and the RF switch, and a shorter 0.5 meter 2.92mm cable between the RF switch and the oscilloscope. The shortest 2.92mm cable should be considered between the DUT Tx and the error detector.

How to establish communication with Mini-Circuits

TekExpress TX automation software provides a built-in control function to control the Mini-Circuits switch matrix during the automatic TX test process. Users develop their own automation software to control the RF switch during TX or RX testing. How to connect: There are two ways to communicate with Mini-Circuits switch: USB uses dll (dynamic link library); Ethernet HTTP request.

Basic calibration and test chart

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 8: 32 GT/s basic root or non-root Tx (multi-channel).

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 9: 32 GT/s basic Rx test points.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 10 32 GT/s TP2 pressure eye diagram.

Facing the challenges of high-speed link testing, it is easy to realize PCIe 5 automatic multi-channel testing
Figure 11: 32 GT/s system Rx LEQ test (multi-channel).

The Links:   SKM200GAL123DKLD LP150X12-TL01

Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *