On-Chip Learning Patents: Who Leads, Where the Gaps Are 2026
Filing growth compares 2021 (57 records) with 2024 (51) — 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 423 records in scope (CR5), not by the ranked leaders only.
What the on-chip learning patent record shows
On-chip learning covers circuits and architectures that let a processor adapt its own weights, thresholds or connection strengths locally, without shipping data off-chip for training. The 423 records in scope span 2015 through the 2026 partial year, and they sit almost entirely inside conventional digital-processing and AI-model IPC classes rather than a dedicated neuromorphic-only code, which tells you the field is still being claimed through general compute and accelerator architecture rather than a settled classification of its own. Publication lags filing by roughly 18 months, so the most recent one to two years understate real filing activity.
Filing activity built steadily from 2017, crested at 58 records in 2023, and the 2021-to-2024 window — the most recent span that can be read as complete — shows an 11% decline rather than a plateau. Concentration is real but not extreme: the ranked leader holds 60 records against 15 at fifth place and 11 at tenth, so the top of the field is well ahead of the pack without shutting out newer entrants.
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
Filing trend and technology composition
The chart and class breakdown below draw on the full 423-record set, using patent families as the unit of count and IPC subclass as the unit of technology.
Ten years of filing activity
Volume rose from 29 records in 2017 to a peak of 58 in 2023, then eased to 51 by 2024 — an 11% pullback across the 2021-to-2024 span. 2025 and 2026 figures will continue to fill in as publication catches up with filing.
Where the claims sit
G06F (electric digital data processing) appears in 60.5% of records and G06N (AI-model computing) in 37.8%, with H04L (digital information transmission) at 12.8% and B60W (vehicle control) at 7.3% marking the automotive-adjacent branch. Because a single record can carry several IPC codes, these shares overlap and do not sum to 100%.
Shares are the percentage of the 423 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Neuromorphic Computing: On Chip Learning Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about neuromorphic computing: on chip learning patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaMost-cited records and a representative filing
Computing device, operation method thereof, and system on chip with dynamic arithmetic unit allocation between deep learning accelerator and vector processor
A computing device, an operation method, and a system on chip in which a resource allocation manager sits between an operator's multiple arithmetic units and dynamically allocates them to a deep learning accelerator or a vector processor based on each unit's calculation load. The system accepts operation requests from both consumers and assigns a first arithmetic unit group accordingly, letting a single pool of hardware serve two different compute demands rather than fixing capacity to either in advance.Filed by ITE Tech. Inc., published 2025-11-18 — illustrates how dynamic resource allocation between accelerator and vector processor is being claimed at the system-on-chip level.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20210117242A1 | Infrastructure processing unit | 152 |
| 2 | US9792397B1 | System and method for designing system on chip (SoC) circuits through artificial intelligence and reinforceme… | 149 |
| 3 | US11320588B1 | Super system on chip | 134 |
| 4 | US20180150684A1 | Age and gender estimation using small-scale convolutional neural network (CNN) modules for embedded systems | 128 |
| 5 | US20210133607A1 | Systems and methods for self-learning artificial intelligence of things (AIOT) devices and services | 117 |
| 6 | US20190392297A1 | Deep learning hardware | 117 |
| 7 | US20190266485A1 | Arithmetic unit for deep learning acceleration | 107 |
| 8 | US20210294944A1 | Virtual environment scenarios and observers for autonomous machine applications | 105 |
| 9 | US8667439B1 | Automatically connecting SoCs IP cores to interconnect nodes to minimize global latency and reduce interconne… | 94 |
| 10 | WO2018126073A1 | Deep learning hardware | 86 |
Citation counts reward older filings that have had more time to accumulate references — read them as a signal of influence within this corpus, not as a ranking of current technical importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
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 →MCP server & REST API
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 →What the numbers mean for strategy
Four read-outs from the dataset that matter more for a filing or freedom-to-operate decision than the raw counts alone.
The top of the field is well ahead, not closed
Five companies hold 38.5% of all 423 records, and the leader's 60 records dwarf the 15 held at fifth place. That gap says the leader has built a genuine claim thicket, but the drop-off after the top few names means the field has not consolidated the way a mature, single-standard technology would.
Volume peaked in 2023, then eased back
Filings climbed from 29 in 2017 to a peak of 58 in 2023, then the 2021-to-2024 window shows an 11% decline to 51. Treat 2025 and 2026 counts as still filling in given the roughly 18-month publication lag rather than reading them as a further drop.
Claims are framed as compute architecture first
G06F (digital data processing) covers 60.5% of records and G06N (AI-model computing) 37.8%, meaning most on-chip learning inventions are claimed through general processor and system architecture rather than as a self-contained AI-model method. The 7.3% B60W share marks a distinct automotive-control branch worth tracking separately.
Filing activity is heavily US-weighted
The United States receives the largest share of filings at 273, with WIPO/PCT (45), India (37) and the EPO (29) trailing well behind. That pattern suggests the commercial contest is still centred on the US market first, with PCT filings used to keep international options open.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to neuromorphic computing: on chip learning patent landscape, with the prior art for and against each one.
Where to take this analysis
The figures above establish the shape of the field; the next step is usually to test a specific claim or design against it.
Check freedom-to-operate against the leader's claim set
With one assignee holding 60 records against 15 at fifth place, any new filing in this space should be checked against that leader's claim language before committing engineering resources to a similar architecture.
Explore freedom-to-operate in EurekaTrack the automotive-control branch separately
B60W appears in 7.3% of records, a distinct pocket from the core G06F/G06N compute claims, and it is worth its own watch list if vehicle-integrated learning is part of the roadmap.
Set up a branch watch in EurekaModel the white space before drafting
The gap between the top few assignees and the long tail below tenth place suggests specific sub-claims remain open; mapping them against existing claim language before drafting reduces the risk of filing into dense prior art.
Run a white-space check in EurekaCommon questions about on-chip learning patents
The dataset ranks 100 assignees across 423 records, with the leading company holding 60 records — well ahead of the fifth-ranked holder at 15 and the tenth-ranked holder at 11. The top five combined account for 38.5% of all 423 records, and the top ten for 53.0%. That means the field has a clear leader but is not closed off: more than 45% of filings sit outside the top ten, spread across a long tail of single- and few-filing entrants including chipmakers, automotive groups and individual inventors.
Filing volume rose from 29 records in 2017 to a peak of 58 in 2023, then the most recent complete span — 2021 to 2024 — shows an 11% decline to 51 records. Because patent publication typically lags filing by around 18 months, the lower counts visible in 2025 and 2026 are an artefact of that lag rather than firm evidence of a slowdown; those years will keep filling in as more records publish. The honest read is that the field peaked around 2023 and has since eased back modestly, not that it is in decline.
Most records are classified under G06F (electric digital data processing, 60.5% of 423 records) and G06N (computing based on AI models, 37.8%), meaning the majority of inventions are framed as general processor or system architecture rather than narrow AI-method claims. Smaller but distinct pockets exist in H04L (digital information transmission, 12.8%), B60W (vehicle control, 7.3%) and A61B (diagnosis and surgery, 5.9%), the latter two pointing to automotive and medical-device integration of on-chip learning hardware. Because a single record can carry multiple IPC codes, these shares overlap rather than summing to 100%.
The concentration figures point to it indirectly: the top ten assignees hold 53.0% of all 423 records, leaving nearly half the field spread thinly across many smaller filers, which is where under-claimed sub-branches tend to survive. The automotive-control branch (B60W, 7.3% of records) and the medical-device branch (A61B, 5.9%) are both far smaller than the core compute classes, suggesting those application-specific integrations of on-chip learning are less densely claimed than the underlying processor architecture itself. A rigorous white-space check should still be run against current claim language before drafting.
A high citation count inside this corpus, such as the 152 citations on US20210117242A1 ('Infrastructure processing unit'), signals that later filers repeatedly built on or referenced that document's claims, which is a mark of influence within the searched set. It is not a measure of current commercial importance, because citation counts mechanically favour older records that have simply had more time to accumulate references. A newer, less-cited filing can still be more relevant to a live design-around question than an older, heavily cited one.
Research Neuromorphic Computing: On Chip Learning Patent Landscape in depth with Eureka
Go past this page: query the whole neuromorphic computing: on chip learning patent landscape 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.