Marvell Technology has been granted a patent for a processing unit supporting inference acceleration for machine learning. The unit includes an inline post-processing unit utilizing lookup tables to perform operations on data from registers, enhancing efficiency by avoiding streaming data from on-chip memory. GlobalData’s report on Marvell Technology gives a 360-degree view of the company including its patenting strategy. Buy the report here.
Smarter leaders trust GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
According to GlobalData’s company profile on Marvell Technology, Network virtualization was a key innovation area identified from patents. Marvell Technology's grant share as of May 2024 was 47%. Grant share is based on the ratio of number of grants to total number of patents.
Processing unit for inference acceleration in machine learning
A recently granted patent (Publication Number: US11995569B2) discloses a novel processing unit designed to enhance efficiency in performing operations on data. The processing unit includes a post-processing unit that maintains lookup tables for operations, accepts data from registers instead of streaming from on-chip memory, and performs operations on each element of the data from a processing block on a per-element basis using the lookup tables. The post-processing unit then streams the results back to the on-chip memory after completing the operation. Additionally, the processing unit features a matrix multiplication block for multiplying matrices A and B to generate matrix C, accumulating results in registers, and providing the output to the post-processing unit.
Furthermore, the patent details the utilization of multiple lookup tables for approximating per-element operations based on numerical analysis. The post-processing unit divides operations into sections represented by curves extrapolated from lookup tables, allowing for efficient determination of values based on input. It also implements operations for integer and floating-point input values using symmetries and piece-wise linear approximations, optimizing performance for operations like tanh and sigmoid. By leveraging dynamic tables and sub-regions within exponent regions, the processing unit minimizes the number of table entries required, enhancing computational efficiency and accuracy in handling complex operations on various types of input data.
To know more about GlobalData’s detailed insights on Marvell Technology, buy the report here.
Data Insights
From
The gold standard of business intelligence.
Blending expert knowledge with cutting-edge technology, GlobalData’s unrivalled proprietary data will enable you to decode what’s happening in your market. You can make better informed decisions and gain a future-proof advantage over your competitors.
GlobalData, the leading provider of industry intelligence, provided the underlying data, research, and analysis used to produce this article.
GlobalData Patent Analytics tracks bibliographic data, legal events data, point in time patent ownerships, and backward and forward citations from global patenting offices. Textual analysis and official patent classifications are used to group patents into key thematic areas and link them to specific companies across the world’s largest industries.
