Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →Filing growth compares 2021 (4 records) with 2024 (22) — a three-year span. 2024 is the most recent year we treat as complete: publication lags filing by roughly 18 months, so 2025 onwards are still filling in and any growth rate that ends there would understate the field. Top-5 share is the combined record count of the five largest assignees divided by all 160 records in scope (CR5), not by the ranked leaders only.
Vector database system design spans the mechanics that turn a similarity-search index into infrastructure: index sharding, memory-to-disk tiering, replication strategy, consistency models, metadata filtering and multi-tenancy controls layered on top of nearest-neighbour retrieval. This dataset pulls 160 published records filed against exactly that combination of search terms, spanning United States, PCT, Indian, European, Canadian and Australian filings between 2015 and mid-2026.
The scope favours system-level claims over pure similarity-search algorithms: a filing only enters this set if it also touches how the index is operated in production, not just how it computes distance. That framing puts storage vendors, cloud database providers and a scattering of individual inventors in the same table, which is unusual for a retrieval-algorithm search and worth keeping in mind when reading the assignee ranking.
Two views of the same 160 records: how filing volume moved year over year, and which IPC subclasses the claims actually sit in.
Annual filings rose from zero in 2017 to a peak of 47 in 2025, with growth concentrated in the 2021-to-2024 window (4 to 22 records, +450%). 2025 and 2026 figures will keep revising upward as later-filed applications publish, since publication typically lags filing by around 18 months.
G06F (electric digital data processing) appears in 76.3% of the 160 records, confirming this is primarily a systems-and-infrastructure corpus. G06N (AI-model computing) at 13.8% and H04L (digital transmission) at 5.6% mark the smaller but growing overlap with learned indexes and distributed networking; classes outside data processing, such as C12N and A61K, reflect multi-domain assignees filing adjacent biotech work rather than vector-search claims proper.
Shares are the percentage of the 160 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
This page is one run against one query. Ask Eureka your own question about vector database system design and every answer comes back with the patent numbers behind it.
Try EurekaThe filing describes a write path for a vector database that adds index update information to a memory-resident, real-time read-write vector graph index, then flushes that memory table to a distributed file system and converts the in-memory index into a second, disk-resident index structure on flush.Filed by Alipay (Hangzhou) Information Technology Co., Ltd., dated 2026-01-01 — one of the most recent records in scope, illustrating the current state of memory/disk tiering claims for vector indexes.


| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20060059173A1 | Systems and methods for efficient data searching, storage and reduction | 331 |
| 2 | US20060059207A1 | Systems and methods for searching of storage data with reduced bandwidth requirements | 199 |
| 3 | US20130103657A1 | Time-series data management device, system, method, and program | 166 |
| 4 | US7523098B2 | Systems and methods for efficient data searching, storage and reduction | 101 |
| 5 | US20210019063A1 | Utilizing data views to optimize secure data access in a storage system | 36 |
| 6 | US20090228453A1 | Systems and Methods for Efficient Data Searching, Storage and Reduction | 33 |
| 7 | US20090234855A1 | Systems and Methods for Efficient Data Searching, Storage and Reduction | 31 |
| 8 | US9298854B2 | Time-series data management device, system, method, and program | 30 |
| 9 | WO2013063366A2 | Methods for screening viral like particles and identifying neutralizing epitopes and related vaccines, constr… | 29 |
| 10 | US20090234821A1 | Systems and Methods for Efficient Data Searching, Storage and Reduction | 29 |
Citation counts are shaped by age as much as by relevance — older records accumulate more citations simply by being searchable longer, so treat this table as a signal of influence within the corpus, not a ranking of current importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →Three cuts of the same dataset: concentration at the top, where the claim space is thick, and where citation weight actually sits.
The leading assignee alone accounts for 32 records, and the top 5 combined hold 45.0% of all 160 records in scope. But the ranking runs 81 companies deep, with the fifth-place holder at only 6 records and tenth place at 3 — so beyond the leader, the field is a long tail of small filers rather than a closed oligopoly.
Annual filings rose from 4 in 2021 to 22 in 2024, a period that predates the publication lag affecting 2025–2026 figures. That three-year run is the clearest complete signal in the trend line and marks when system-design claims around vector databases moved from niche to active filing category.
G06F covers more than three-quarters of the 160 records, while G06N (AI-model computing) sits at 13.8%. That gap suggests most current claims still describe sharding, tiering and consistency mechanics rather than the learned or model-driven indexing techniques that are only beginning to show up in the classification mix.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to vector database system design, with the prior art for and against each one.
The assignee ranking mixes established storage and cloud vendors with a small number of individual inventors who co-file repeatedly, alongside a long tail of single-filing entrants.
The top-ranked assignee holds 32 of the 160 records, well ahead of fifth place at 6 records. That gap is the clearest structural feature of the ranking: whoever leads has built a genuinely broad portfolio, while the rest of the field files in single digits.
Among the 10 identified co-assignee pairs, the strongest link — David S. Peabody and Bryce Chackerian — appears together on 8 records, with STC UNM also paired with each of them on 4 records apiece. This points to a university-affiliated inventor duo filing consistently rather than a corporate joint-venture pattern.
Several previously active assignees, including large incumbents, show zero filings in the latest year, and one shows a -100% year-on-year change. Read this cautiously: publication lag means the most recent year is always undercounted, so a drop here may reflect timing rather than a genuine pullback.
| Assignee | Recent year | YoY |
|---|---|---|
| ACURAI INC | 1 | -80% |
| International Business Machines Corporation (IBM) | 0 | — |
| STC.UNM | 0 | — |
| Diligent Technologies | 0 | — |
| Oracle International Corporation | 0 | -100% |
| PEABODY DAVID S | 0 | — |
| CHACKERIAN BRYCE | 0 | — |
| Cisco Technology, Inc. | 0 | — |
The dataset points to a field that is still forming its claim structure around memory/disk tiering and learned indexing. Two natural next steps follow from that.
With one assignee holding 32 of 160 records, any new filing in core sharding or tiering claims should be checked against that portfolio specifically, not just the aggregate corpus.
Run a freedom-to-operate check in EurekaLearned-index tiering and cross-tenant metadata isolation show thin coverage relative to core infrastructure claims, which is where a first-mover claim is more likely to clear prior art.
Explore white space in EurekaThis dataset identifies 160 published records matching vector database system design terms, such as index sharding, similarity search, metadata filtering, consistency models and replication strategy, filed between 2015 and mid-2026. The count reflects a specific search string, so a broader or narrower query on adjacent terms like pure similarity-search algorithms would return a different number. Patent families are the more reliable unit here than raw document counts, since a single invention can generate several continuation or multi-jurisdiction filings that inflate a simple document tally.
The assignee ranking covers 81 companies, with the top-ranked filer holding 32 of the 160 records and the top 5 combined accounting for 45.0% of all records in scope. Beyond the leader, filing volume drops quickly — fifth place holds 6 records and tenth place holds 3 — so most of the ranked list consists of companies with only one or a few filings. That structure means a single leader has built meaningful depth while the rest of the field remains fragmented, which matters if you are assessing competitive risk rather than just market noise.
Yes, based on the complete years available: annual filings rose from 4 in 2021 to 22 in 2024, a roughly 450% increase over that three-year span. The trend continues to rise through 2025, which recorded 47 filings, but 2025 and 2026 figures are still incomplete because publication typically lags filing by about 18 months. Treat 2021–2024 as the reliable growth signal and expect the most recent two years to be revised upward as more applications publish.
The dominant IPC classification is G06F (electric digital data processing), covering 76.3% of the 160 records, confirming that most claims describe system infrastructure — sharding, tiering, replication — rather than pure algorithms. G06N (AI-model computing) appears in 13.8% of records, marking a smaller but present overlap with learned or model-driven indexing techniques. Because a single record can carry multiple IPC classes, these shares add up to more than 100%, and they should be read as class prevalence rather than a strict partition of the field.
The clearest gaps sit where infrastructure claims (G06F at 76.3%) heavily outweigh AI-model-driven claims (G06N at 13.8%), suggesting learned-index tiering and model-informed sharding are still lightly claimed relative to conventional mechanics. Specific under-claimed branches include cross-tenant metadata filter isolation, consistency models for graph-index replicas, and index conversion behaviour on flush-to-disk. These are inferred from thin representation in the current 160-record set rather than an exhaustive freedom-to-operate search, so any filing decision should be paired with a targeted prior-art search in the specific branch.
Go past this page: query the whole vector database system design corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company's registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.
Help us improve
Something wrong, out of date, or missing on this page? Tell us and it goes straight to the team that maintains it.
This page