Semiconductor Device Skew Correction via Training Pattern Symbol Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As operating speeds of integrated circuit devices increase, skews between data signals and clock signals, as well as data strobes, become more significant, necessitating effective correction methods to maintain data reliability and support the growing speed of memory devices.
Innovation Solution
A semiconductor device is designed to reduce skew by generating a training pattern, which involves a sequence data generator, a symbol changer that replaces predetermined symbols with alternative symbols, and a driver that outputs this pattern to external devices, allowing for data training and adjustment of signal delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If operating speed of integrated circuit devices is increased, then productivity is improved, but skew between data signal and clock signal increases causing reliability degradation
Solution Approach 1:
The patent applies preliminary action by generating training patterns before actual data transmission to measure and correct signal skew. The system performs skew measurement using training patterns transmitted at high speed, then adjusts timing parameters in advance to compensate for skew, enabling reliable high-speed operation without sacrificing data integrity
Solution Approach 2:
The patent implements feedback by continuously measuring skew between data signal and clock signal using training patterns, then using this measurement information to adjust timing parameters. The feedback loop includes skew measurement unit, training pattern generator, and timing adjustment mechanisms that modify transmission parameters based on measured skew to maintain reliability at high operating speeds
2Reliability
If training pattern is generated to correct skew, then reliability is improved, but device complexity increases due to additional components
Solution Approach 1:
The patent applies universality by designing the training pattern generator to serve multiple functions: it generates patterns for skew measurement, patterns for equalization training, and patterns for calibration. The same hardware components (LFSRs, scramblers, symbol generators) are reused across different training operations, reducing overall device complexity while maintaining comprehensive skew correction capabilities
Solution Approach 2:
The patent merges the skew measurement function with the training pattern transmission function. The training pattern generator and skew measurement unit operate integrally, using the same signal paths and processing logic. This consolidation eliminates redundant components and simplifies the overall device architecture while achieving both skew measurement and correction
3Manufacturing precision
If invalid symbols are removed from training pattern, then manufacturing precision is improved, but loss of information increases
Solution Approach 1:
The patent extracts and removes invalid symbol combinations (such as '111' or '000' sequences that cannot be reliably transmitted) from the training pattern before transmission. This extraction is performed by the symbol generator which filters out problematic symbol sequences, ensuring that only valid, reliably transmittable symbols are sent, thereby improving manufacturing precision without significant information loss
Data Source
AI summary
A semiconductor device includes a sequence data generator, which is configured to generate sequence data on a plurality of data lines, and a symbol changer. The symbol changer is configured to generate a training pattern from the sequence data by replacing, for each of the plurality of data lines, each occurrence of a bitstream within the sequence data that has a predetermined symbol with an alternative symbol. The sequence data generator may include a sequence generator, which is configured to generate a pseudo random binary sequence (PRBS), based on a seed value for each clock cycle.