Selective UCIe Link Error Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The Universal Chiplet Interconnect Express (UCIe) link training process is inefficient as it requires initializing and training both the sideband and mainband even if only one part has an error, leading to unnecessary power consumption and performance degradation by making error-free parts unavailable.
Innovation Solution
Identify the part of the UCIe link with an error and maintain the other part active during training, allowing only the erroneous part to be initialized and trained, thereby avoiding unnecessary power and time usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the UCIe link training process initializes and trains both the sideband and mainband even when only one part has an error, then the error recovery is comprehensive and reliable, but the power consumption increases and performance degrades due to unnecessary initialization of error-free parts
Solution Approach 1:
The patent segments the UCIe link training process into independent sideband and mainband initialization stages. The training method separately identifies which part (sideband or mainband) contains the error and only initializes that specific part, rather than forcing initialization of both parts. This segmentation allows selective training, reducing power consumption while maintaining comprehensive error recovery for the affected portion.
Solution Approach 2:
The patent applies partial action by performing only the necessary initialization steps required to recover from the detected error, rather than executing the complete initialization sequence for both sideband and mainband. When an error is detected in only one part, the system performs partial initialization limited to that specific part, avoiding the excessive action of initializing error-free components.
2Reliability
If the UCIe link training process initializes and trains both the sideband and mainband even when only one part has an error, then the error recovery is comprehensive, but the training time increases and available bandwidth decreases
Solution Approach 1:
The training process is segmented into separate detectable stages for sideband and mainband initialization. By identifying which specific part contains the error through separate detection mechanisms, the system can apply targeted initialization only to the affected segment, significantly reducing the total training time required while ensuring complete error recovery for the problematic portion.
Solution Approach 2:
The system performs partial initialization action limited to the specific part requiring error recovery. Instead of executing the full initialization sequence for both sideband and mainband, the method performs only the necessary partial initialization for the affected component, reducing training time loss while maintaining comprehensive error handling for the detected issue.
3Reliability
If the UCIe link training process initializes and trains both the sideband and mainband when only one part has an error, then all potential errors are addressed, but the system performance degrades due to making error-free parts unavailable
Solution Approach 1:
The patent segments the link components into independently trainable units (sideband and mainband). By detecting errors in specific segments and applying initialization only to those segments, the system maintains error-free parts in an available state for data transmission. This selective approach ensures comprehensive error coverage for affected parts while preserving productivity by keeping healthy parts operational.
Solution Approach 2:
The system applies partial initialization action only to the specific part requiring error recovery, rather than forcing initialization of the entire link. This allows error-free parts to remain available for immediate data transmission, maintaining system productivity while still providing comprehensive error handling for the detected problematic portion.
Data Source
AI summary
Various embodiments include methods and devices for efficiently recovering from errors that occur in part but not all of a universal chiplet interconnect express (UCIe) link for chiplets of a computing device. Various embodiments may include identifying a first part of a UCIe link in which an error has occurred, and training the first part of the UCIe link in which the error has occurred while maintaining active a second part of the UCIe link in which no error has occurred.