Share open-source test code
An exclusive interview with a test engineer
Martin Rowe -- Test & Measurement World, 3/1/2007
Mark Marlett is a principal design engineer at LSI Logic where he is the technical lead for SerDes devices. His responsibilities include design, specifications, and standards. Because LSI Logic manufactures custom ICs, the device under test (DUT) that Marlett deals with isn’t the chip itself, but the SerDes intellectual property (IP) on the device. Test chips reside on a test board that contains a power supply, control logic (an FPGA), and connectors for DUT access. SerDes IP includes Fibre Channel, PCI Express, SerialATA, and Serial Attached SCSI. Marlett recently described his test processes to senior technical editor Martin Rowe.
Q: What tests do you perform on SerDes devices?
A: We start by making sure that the IP is functional. It must communicate with another device or with a bit-error-rate tester (BERT). We first must prove to the customer that a device works before we can stress it.
Once the chip is working, we program the SerDes receiver’s adaptive-equalization filters for optimal values. At multigigabit data rates, the signals degrade to the point where there’s no eye opening to measure. A typical IP has over 4000 possible filter settings. During a test, we measure BER, transmitter jitter, receiver jitter tolerance, and signal amplitude while we cycle through the DUT’s states of operation. We measure these parameters over a range of temperatures and power-supply voltages.
Q: Do you automate your measurements?
A: We automate oscilloscopes, BERTs, and the DUT configuration. We use Python (www.python.org), an open-source scripting language.
Q: Why did you choose Python?
A: Python lets us write objects that represent the test equipment and the test chip. We can keep a test script to just a few lines because the control code is embedded in the object. We can, for example, change oscilloscopes and test chips without changing script code. IP designers give us a register map, and the test code reads the map to get the IP functionality. We’re on the fourth generation of SerDes IP, all with the same test code.
We also use Python to perform signal analysis. By adding a math extension to the language, we have as much processing power as we’d get with a commercial data-analysis package—without buying a software license. We developed our own data-analysis objects. We compare our results to those we obtain using the analysis functions on the oscilloscopes. Because we use an open-source language, we can freely share test code with customers.
Q: How do the objects control the test chip?
A: The objects talk to the test board through a standard PC serial port. They control the power supply and send commands to the FPGA to control the test chips. We also have access to error codes on the test chips, and we use those codes to program the receiver equalizers. Thus, we can dynamically program the equalizers to open the receiver’s eye diagram, although not in real time.
Q: How has automation improved productivity?
A: By creating Python objects, we minimize code duplication because engineers can often reuse working code with only small changes. Python code is self-documenting and is thus readable by both people and computers. That makes it easy for one engineer to use another engineer’s code.
Every other month, we will publish an interview with an electronics engineer who has test, measurement, or inspection responsibilities. If you’d like to participate in a future column, contact Martin Rowe at mrowe@tmworld.com.
