Machine Learning Margin Tester Calibration for PCIe Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The Tektronix Margin Tester (TMT) and traditional methods face instrument-to-instrument mismatches in measurement results, leading to variability and inefficiency in testing PCIE devices, particularly when using multiple channels simultaneously.
Innovation Solution
A machine learning-based approach using neural networks is employed to account for the 'signature' of each margin tester, allowing the network to adjust measurements and provide consistent results across different TMT instruments by training on reference data and using tensors to normalize and denormalize inputs and outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional real-time oscilloscope and BERT methods are used to test DUT transmitter and receiver, then measurement accuracy is maintained, but testing throughput is limited due to only four channels available
Solution Approach 1:
A machine learning model is introduced as an intermediary between the margin tester measurements and the reference oscilloscope/BERT measurements. The model learns the transformation relationship between different instrument measurement spaces, enabling consistent measurements across multiple channels while maintaining accuracy. This mediator resolves the contradiction by allowing parallel multi-channel testing without sacrificing measurement consistency.
Solution Approach 2:
The patent transforms the measurement parameters by applying machine learning-based parameter transformations. The margin tester's raw measurements are converted through learned parameter mappings to match the reference instrument's measurement space. This parameter transformation enables consistent results across multiple channels simultaneously tested, resolving the throughput-precision tradeoff.
2Productivity
If multiple margin tester instruments are used to increase testing capacity, then throughput improves, but instrument-to-instrument measurement variability increases
Solution Approach 1:
The machine learning model serves as a universal translator that works across multiple margin tester instruments. By learning the specific characteristics of each instrument and mapping them to a common reference space, the system enables universal measurement consistency across all instruments in the fleet, allowing increased testing capacity without sacrificing precision.
Solution Approach 2:
The system uses feedback from reference measurements (obtained periodically using the oscilloscope/BERT) to continuously refine and update the machine learning model. This feedback mechanism ensures that instrument-to-instrument variability is continuously corrected, maintaining measurement consistency as more instruments are added to increase testing capacity.
3Reliability
If traditional four-channel oscilloscope testing is used, then measurement reliability is maintained, but testing time increases due to sequential channel testing
Solution Approach 1:
The machine learning model is pre-trained using reference measurements collected in advance. This preliminary training phase captures the relationship between margin tester and reference instrument measurements, enabling rapid inference during actual testing. The pre-computed model allows parallel multi-channel testing with reliable results, eliminating the time-consuming sequential testing while maintaining measurement reliability.
Data Source
AI summary
A margin tester includes one or more ports to allow the margin tester to connect to a device under test (DUT), a memory, the memory containing a margin tester signature, a transmitter, a receiver to receive signals from the DUT, one or more processors configured to execute code that causes the one or more processors to: receive multiple signals from the receiver through the one or more ports, generate a performance indicator from the multiple signals, send the performance indicator and the margin tester signature to one or more machine learning networks, and receiving a result from the one or more machine learning networks containing a performance measurement prediction for the DUT.