Database Array and Tensor Storage Query Processing Patents

The assignee ranking behind this dataset covers 100 companies, with the leader holding 471 records and the fifth-placed filer holding 130. That gap shows a small group of large, diversified technology companies sitting well ahead of the rest of the ranked field. Even so, the top 5 combined account for only 9.9% of the 11,639 records in scope, so no single filer or small group controls the field outright; a long tail of smaller filers holds the majority of the activity.

Filings rose from 304 in 2007 to a peak of 580 in 2022, then fell to 358 by 2024 — a 31% decline from the 2021 figure of 517. That 2024 figure is the most recent year that can be treated as reasonably complete, because publication typically lags filing by around 18 months. Years from 2025 onward will keep rising as those filings publish, so the current numbers understate real recent activity rather than confirming a genuine slowdown.

Electric digital data processing (IPC class G06F) is the dominant category, present in 54.7% of the 11,639 records in scope, reflecting the centrality of storage-engine and query-execution mechanics. Digital transmission (H04L, 11.2%) and business-data processing (G06Q, 10.2%) are the next largest, followed by material analysis, data recognition, image processing, pictorial communication and AI-model computing, each in the 5-8% range. Because a single record can carry multiple IPC classes, these shares add up to more than 100% of the record total.

The thinner IPC branches — AI-model computing (G06N, 5.4%), image data processing (G06T, 6.3%) and pictorial communication (H04N, 6.2%) — carry meaningful volume without the density seen in the G06F core (54.7%). Claims drafted at the boundary between array/tensor storage and one of these adjacent fields, rather than purely inside the core storage-engine space, are likelier to avoid the heaviest prior art. That said, any freedom-to-operate conclusion should be checked against the specific claim language of the leading assignees' filings before committing resources.

US9501550B2, filed by Renmin University of China and published 2016-11-22, claims an OLAP query processing method for a database and Hadoop hybrid platform: queries run first on a main working copy, and on node failure the system locates a fault-tolerant copy via namenode lookup and invokes MapReduce to finish the task. The blocking scope is specific to that fault-triggered fallback sequence, not to OLAP-on-Hadoop generally. Systems using a different replication or failover mechanism, or that do not invoke MapReduce on the fault-tolerant copy, are likely to sit outside this claim's coverage.