Samsung patents method to shrink sparse data for lower power
Samsung has been granted a patent for a way to compress data that contains many empty spaces, known as sparse matrices. The method uses a repeating process to shorten sequences of zeros, which reduces the amount of data the hardware has to move. This reduction in data movement helps lower the power consumption of the device.
In the filing’s own words
A sparse matrix is compressed using recursive application of word replacement based on a bijective process. The bijective process generally shrinks sequences of zeros. The shrinkage generally continues at a geometric rate with additional recursions, up to a limit. The bijective process reaches a global minimum for an output sequence. After the global minimum is reached, the length of the output expands at a linear rate. Due to the bijective process, decompression is able to proceed in a lossless manner to exactly recover the input sparse matrix. Clusters of non-zero bits in the sparse matrix may be in a single line or distributed across neighboring lines. Clusters in a single line are addressed by a design parameter corresponding to a distance between binary-1s in the data. Clusters distributed across neighboring lines are flattened using a pre-processing stage.
What exactly they claim (claim 1)
1. A method of compressing an initial input sequence comprising a plurality of words to obtain a compressed output sequence using a plurality of recursion stages, the method comprising: inputting a first word into a first recursion stage of the plurality of recursion stages, wherein the plurality of words comprises the first word; replacing, in the first recursion stage, the first word to obtain a second word and outputting the second word to a second recursion stage of the plurality of recursion stages; and repeating the inputting, the replacing and the outputting until a final word of the initial input sequence has been replaced by the first recursion stage, wherein the plurality of words comprises the final word, wherein successive recursion stages continue to perform the inputting, the replacing and the outputting until no further second words are provided to the successive recursion stages (?) from an upstream recursion stage, wherein an output sequence at an output of a final recursion stage of the plurality of recursion stages is the compressed output sequence, and wherein hardware power consumption is reduced by reducing bit-toggles.
Facts
- Company
- Samsung (SAMSUNG ELECTRONICS CO., LTD.)
- Official title
- Sparse matrix compression using recursive word replacement maps
- Patent number
- US12801520B2
- Application number
- 19185809
- Filed
- Apr 22, 2025
- Granted
- Nov 17, 2026
- Inventors
- 2
- Claims
- 20
Classified under
G06F 12/0223 Physics & computingMore from Samsung in Chips & hardware
- Samsung patents quantum chip design using tiny electron traps Nov 12, 2026
- Samsung patents chip that combines camera and screen Oct 13, 2026
- Samsung patents chip that hides encrypted data from the main processor Oct 27, 2026
- Samsung patents display with tiny holes to hide cameras behind the screen Nov 3, 2026
- Samsung patents faster chip testing using machine learning Sep 1, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.
