Multiple Data Bus Inversion for High-Bandwidth Memory Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data bus inversion (DBI) methods in high-bandwidth memory (HBM) devices face challenges in minimizing data transitions, leading to increased power consumption, signal integrity issues, and heating problems due to the high number of data bits transitioning during data transmission.
Innovation Solution
The implementation of a multiple data bus inversion (DBI) method that selectively performs first, second, and third DBI operations based on the value of data bits, grouping data bits in different directions to minimize data transitions, using an inversion logic unit within the memory device to determine and apply DBI signals accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If traditional DBI method is used in HBM devices, then data transmission is performed, but the number of data transitions increases leading to high power consumption
Solution Approach 1:
The patent segments the data bus into multiple independent inversion groups (first inversion group, second inversion group, third inversion group) instead of treating all data bits as a single unit. Each group can be inverted independently based on its own transition count, allowing finer-grained optimization of power consumption while maintaining data transmission efficiency.
Solution Approach 2:
The patent applies different inversion decisions to different local groups of data bits based on their specific transition characteristics. Each inversion group evaluates its own number of transitions and makes independent inversion decisions, rather than applying a uniform inversion strategy to all data bits, thereby optimizing power consumption locally where needed.
2Use of energy by moving object
If more data bits are inverted to reduce transitions, then power consumption decreases, but signal integrity issues and heating problems increase
Solution Approach 1:
The patent performs partial inversion by dividing data bits into multiple groups and selectively inverting only those groups where the number of transitions exceeds a threshold. This partial action approach reduces overall transitions and power consumption while avoiding excessive inversion that could degrade signal integrity and increase heating.
Solution Approach 2:
The patent changes the inversion parameter from a global binary decision (invert all or none) to multiple group-level decisions based on transition counts. By monitoring and comparing the number of transitions in each group against thresholds, the system dynamically adjusts inversion behavior to optimize both power consumption and signal integrity.
3Device complexity
If data bits are grouped in single direction for DBI, then inversion is simplified, but minimum data transitions cannot be achieved for multiple-bit wide data
Solution Approach 1:
The patent extends the grouping approach from a single dimension to multiple dimensions by creating inversion groups that consider different groupings of the same data bits. The inversion logic unit evaluates multiple grouping perspectives and selects the inversion strategy that minimizes total transitions, effectively adding a dimensional perspective to the inversion decision process.
4Adaptability or versatility
If HBM performs DBI in units of bytes in a channel, then multi-channel interface is supported, but the number of data transitions remains high
Solution Approach 1:
The patent further segments the byte-level DBI approach by dividing data bits within each byte into multiple inversion groups. This finer segmentation allows the system to maintain multi-channel interface support while reducing the number of transitions within each channel by making more granular inversion decisions at the bit group level rather than at the byte level.
Data Source
AI summary
A method of performing multiple data bus inversion (DBI) and a memory device performing the method are provided. The multiple DBI includes first through third DBI operations, wherein the first DBI operation determines whether to perform data inversion on a first data inversion group in which M×N data bits of a M×N data bit structure are grouped and performs the data inversion on the first data inversion group, the second DBI operation determines whether to perform data inversion on second data inversion groups formed by grouping M data bits from among the M×N data bits and performs the data inversion on the second data inversion groups, and the third DBI operation determines whether to perform data inversion on third data inversion groups formed by grouping N data bits from among the M×N data bits and performs the data inversion on the third data inversion groups.