Multicore Processor Core Performance Degradation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multicore processors, identifying defects or sub-optimal performance in processor cores is challenging due to variations in performance measurements across cores, which can be attributed to design flaws, manufacturing anomalies, or degradation over time, making it difficult to determine if a core is defective or operating sub-optimally.
Innovation Solution
A testing apparatus is configured to concurrently or sequentially measure the performance of each processor core, generating test values that are analyzed against thresholds to identify any core exceeding predetermined limits, allowing for the rejection of defective or underperforming cores and enabling comparison of core performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple cores are expected to perform similarly, then manufacturing consistency is improved, but detecting sub-optimal performance becomes more difficult due to performance variations from design flaws and manufacturing anomalies
Solution Approach 1:
The patent segments the performance measurement process by dividing cores into groups and performing measurements on multiple cores simultaneously. This segmentation allows for statistical analysis of performance variations while maintaining the ability to detect individual defective cores through comparison against group statistics.
Solution Approach 2:
The patent implements feedback by using performance measurements from multiple cores to establish statistical parameters (mean and standard deviation), then using these parameters to identify outliers. The feedback loop compares each core's performance against the collectively derived statistics, enabling detection of manufacturing anomalies while accounting for natural performance variations.
2Reliability
If performance measurements are taken across multiple cores, then defect detection capability is improved, but measurement time and system complexity increase
Solution Approach 1:
The patent merges multiple measurements into a single statistical framework by collecting performance data from multiple cores simultaneously and analyzing them together using mean and standard deviation calculations. This merging approach improves defect detection reliability while reducing total measurement time compared to sequential individual core measurements.
Solution Approach 2:
The patent applies partial action by measuring a subset of cores (e.g., 2-10 cores) rather than requiring exhaustive measurement of all cores in the processor. This partial sampling provides sufficient statistical information to detect defective cores while minimizing measurement time and resource consumption.
3Measurement precision
If performance thresholds are set to detect defective cores, then measurement precision is improved, but false rejection of functional cores increases
Solution Approach 1:
The patent changes the parameter for threshold determination from fixed predetermined values to dynamic statistical parameters (mean ± k×standard deviation) derived from actual measurements. This parameter change allows the threshold to adapt to the specific performance distribution of the measured cores, improving both measurement precision and reducing false rejections of functional cores.
Solution Approach 2:
The patent uses feedback from the measured performance data to dynamically adjust the rejection threshold. By calculating the mean and standard deviation from actual measurements and using these to set the threshold (mean + k×standard deviation), the system adapts to the specific batch of cores being tested, improving both precision and reliability of defect detection.
Data Source
AI summary
An embodiment of a system is disclosed, including an interface configured to communicate to a device under test (DUT). The DUT may include a plurality of processor cores. The system also includes a testing apparatus configured to concurrently measure a performance of a portion of each processor core to generate a first set of test values. Each test value of the first set may correspond to a given processor core of the plurality of processor cores. The testing apparatus may also be configured to analyze the first set of test values, and reject the DUT in response to a determination that at least one test value of the first set of test values exceeds a first threshold.