SerDes Link Training with Adaptive Equalization State Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-speed data communication interfaces face challenges in maintaining consistent equalization settings across training runs due to varying conditions and board design/manufacturing uncertainties, leading to inconsistent tap settings and increased bit error rates.
Innovation Solution
An information handling system with a transmitter and receiver components that include an equalization stage and decision feedback equalization (DFE) stage, where the receiver determines tap setting variability and adjusts the equalization state based on predetermined thresholds to ensure consistent settings and reduce bit error rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional link training is used with fixed equalization settings, then the system is simple to implement, but tap settings become inconsistent across training runs due to board design and manufacturing variations
Solution Approach 1:
The patent implements dynamic equalization adjustment by allowing the equalization stage to switch between low and high equalization states based on detected tap setting variability. This dynamic adaptation resolves the contradiction by making the system flexible enough to handle board variations while maintaining simplicity through automated state selection.
Solution Approach 2:
The patent employs feedback mechanisms where the receiver detects variability in tap settings across multiple training runs and provides this information back to the transmitter. Based on this feedback, the equalization state is adjusted to reduce variability, thereby improving tap setting consistency without requiring complex manual calibration.
2Reliability
If high equalization state is always used, then tap setting consistency improves, but power consumption and system complexity increase
Solution Approach 1:
The system dynamically selects between low and high equalization states based on actual tap setting variability detected during training runs. This resolves the contradiction by applying high equalization only when necessary to achieve consistency, rather than continuously, thereby reducing overall power consumption while maintaining reliability.
Solution Approach 2:
The patent changes the equalization parameter (state) based on detected performance variability. By adjusting this parameter dynamically rather than maintaining a fixed high state, the system achieves tap setting consistency only when needed, optimizing the trade-off between reliability and power consumption.
3Reliability
If multiple training runs are performed to detect variability, then tap setting consistency improves, but training time increases
Solution Approach 1:
The patent performs a limited number of training runs (e.g., two runs) to detect variability rather than exhaustive testing. This partial action approach resolves the contradiction by gathering sufficient information to make an informed decision about equalization state without excessive time investment, achieving practical consistency with acceptable training time.
4Reliability
If dynamic equalization adjustment is implemented, then bit error rate reduces, but device complexity increases
Solution Approach 1:
The system performs self-diagnosis by detecting tap setting variability during training runs and automatically adjusts its own equalization state accordingly. This self-service capability reduces bit error rates without requiring external intervention or complex control systems, as the equalizer autonomously optimizes its own operation.
Data Source
AI summary
An information handling system includes a first component including a transmitter for a high-speed serial data interface, and a second component including a receiver for the high-speed serial data interface. The receiver includes an equalization stage and a decision feedback equalization (DFE) stage. The equalization stage has an input to configure the equalization stage in one of a first low equalization state and a first high equalization state. The DFE stage has a plurality of tap inputs. The first component provides a plurality of training runs on the high-speed serial data interface. The second component receives the training runs, provides for each training run a set of tap settings for each tap input, determines whether or not a variation in the tap settings for the training runs is greater than a predetermined variation value, and, when the variation is greater than the predetermined variation value, sets the first input to configure the first equalization stage from the first low equalization state to the first high equalization state.


