Heterogeneous Computing Patents: Who Leads, Where the Gaps Are 2026
- Filing peaked in 2017 at 23 families and has declined through the midpoint year of 2022 (7 families), suggesting the core claim space around unified memory and task offloading was staked out early rather than being a currently expanding frontier.
- G06F dominates at 126 of 127 records while adjacent AI-model computing (G06N, 10 records) and image recognition (G06T/G06V, 12 combined) remain thin — these are the branches least defended by incumbent filers.
- The United States receives 46 filings versus 24 at the EPO and 14 in China with India's 12 filings signalling a filing office that is easy to overlook when scoping freedom-to-operate.
What this landscape covers
Heterogeneous computing and accelerator integration spans the mechanisms that let a CPU, GPU and other accelerators share work and memory without the programmer managing every data transfer by hand: unified memory addressing, task offloading logic, cache coherence protocols that keep accelerator and host views of memory consistent, and the interconnect bandwidth engineering that makes coherence affordable at scale. The search scope here is anchored in G06F9, G06F13 and G06F15 — the core data-processing and interconnect subclasses — rather than the accelerator hardware itself.
The 127 patent families in this dataset run from 2015 through a partial 2026 year, giving a decade-long view of how the field's core claims were staked out. Publication lags filing by roughly 18 months, so the most recent one or two years in any trend chart will always look thinner than they eventually turn out to be.
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
Two views of the same 127 families: when the claims were filed, and which IPC subclasses they sit in.
A front-loaded filing curve
Filings ran at 23 in the peak year of 2017 and had fallen to 7 by the 2022 midpoint, continuing down toward 2 in the most recent year on record. Read the tail end cautiously — publication lag means 2025 and 2026 filings are still arriving — but the shape from 2017 to 2022 is a real decline, not an artefact of the cut-off.
Concentration in G06F, with thin adjacent coverage
G06F covers 126 of the 127 records, confirming this is fundamentally a data-processing-architecture dataset. G06N (AI-model computing, 10 records), G06T and G06V (image processing and recognition, 7 and 5 records) show accelerator-integration claims touching AI and vision workloads, but at a fraction of the density found in the core subclass — those intersections are comparatively open.
Shares are the percentage of the 127 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Heterogeneous Computing and Accelerator Integration with Eureka
This page is one run against one query. Ask Eureka your own question about heterogeneous computing and accelerator integration and every answer comes back with the patent numbers behind it.
Try EurekaA representative claim in this space
Facilitating efficient communication and data processing across clusters of computing machines in heterogeneous computing environment
A mechanism is described for facilitating efficient communication and data processing across clusters of computing machines in a heterogeneous computing environment. A method includes detecting a request for processing of data using a programming framework and a programming model; facilitating interfacing between the programming framework and the programming model, wherein interfacing includes merging the programming model into the programming framework, wherein interfacing further includes integrating the programming framework with a distribution framework hosting the programming model; and calling on the distribution framework to schedule processing of a plurality of jobs based on the request.Filed by Intel Corporation, published 2018-11-29 as US20180341526A1.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20190220703A1 | Technologies for distributing iterative computations in heterogeneous computing environments | 369 |
| 2 | US20190138934A1 | Technologies for distributing gradient descent computation in a heterogeneous multi-access edge computing (ME… | 333 |
| 3 | US20190347125A1 | Systems, methods, and apparatuses for heterogeneous computing | 245 |
| 4 | US20030046639A1 | Method and systems for facilitating creation, presentation, exchange, and management of documents to facilita… | 222 |
| 5 | WO2018125250A1 | Systems, methods, and apparatuses for heterogeneous computing | 198 |
| 6 | US20200249998A1 | Scheduling computation graph heterogeneous computer system | 95 |
| 7 | US20200401440A1 | Systems, methods, and apparatuses for heterogeneous computing | 80 |
| 8 | CN107239315A | 面向神经网络异构计算平台的编程模型 | 54 |
| 9 | US20130332937A1 | Heterogeneous Parallel Primitives Programming Model | 42 |
| 10 | CN104142845A | 基于OpenCL-To-FPGA的CT图像重建反投影加速方法 | 34 |
Citation counts accumulate over time and favour older filings; treat them as a signal of historical influence within this corpus, not of which claims matter most today.
Patent titles are shown in the language they were filed in, not translated, so that each record stays verifiable against the original filing — a translated title will not match in Eureka or in any national register. 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 a filing decision
Three patterns stand out once the raw counts are put in context.
The core claim space is not currently expanding
A decline from 23 families in 2017 to 7 at the 2022 midpoint, continuing toward 2 in the latest year, indicates the foundational unified-memory and coherence claims were largely filed in the first half of the dataset's window. New entrants now compete against an already-dense prior-art base rather than an open field.
Nearly all activity sits in one IPC subclass
High density in G06F means the core interconnect and data-processing claim space is occupied, not that the underlying problem is solved. Filers targeting this subclass directly should expect to design around existing claims rather than stake new ground.
Filing offices skew toward the US and Europe
The United States accounts for the largest share of receiving offices, with the EPO second and China a distinct third. India's 12 filings and WIPO's 9 PCT filings are easy to miss in freedom-to-operate work but represent a meaningful share of the total 127 families.
Co-assignee activity is narrow and academic-leaning
The strongest co-filing pairs recur around the same small set of individual inventors (Riera, Ren, Quraishi, Bank Tavakoli), each appearing together up to 4 times. This points to a research-group filing pattern rather than broad corporate joint-venture activity.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to heterogeneous computing and accelerator integration, with the prior art for and against each one.
Who is filing, and who has slowed
Recent-year momentum shows a field where even the more active recent filers are pulling back, not accelerating.
Even large incumbents show zero recent-year filings
Intel Corporation, present among the most-cited records in this dataset, shows 0 filings in the latest year against the prior year — a full pull-back rather than a slowdown. This is consistent with the broader decline from the 2017 peak.
Smaller filers are also decelerating
Entities filing only 1 family in the latest year are still showing year-on-year declines of 67%, meaning even modest, recent activity is trending down rather than building.
A long tail behind a handful of named filers
Corporate assignees such as Partec AG and ZTE sit alongside academic and individual-inventor filings, with the University of Arizona board of regents also present. The mix suggests both corporate defensive filing and academic exploratory filing contribute to the dataset.
| Assignee | Recent year | YoY |
|---|---|---|
| UniFabrix Ltd. | 1 | -67% |
| Intel Corporation | 0 | -100% |
| ParTec Cluster Competence Center GmbH | 0 | — |
| PARTEC AG | 0 | — |
| ZTE Corporation | 0 | — |
| Xilinx, Inc. | 0 | — |
| Arizona Board of Regents on behalf of the University of Arizona | 0 | — |
| RIERA MICHAEL F | 0 | — |
Where to take this analysis
The landscape data points to specific next steps depending on whether the goal is freedom-to-operate, white-space filing, or monitoring.
Check freedom-to-operate against the dense G06F core
With 126 of 127 records classed under G06F, any new filing touching unified memory or cache coherence should be checked against the most-cited records in this dataset before drafting claims.
Explore prior art in EurekaScope the thinner G06N and G06T/G06V intersections
AI-model and vision-workload intersections with accelerator coherence carry a fraction of the core subclass's filing density, which is where new claim scope is more likely to be available.
Map white space in EurekaTrack assignee momentum rather than historical citation counts
Citation leaders in this dataset are largely older filings; recent-year momentum data shows declining activity even among historically influential filers, so monitoring should weight recency over citation volume.
Set up assignee tracking in EurekaCommon questions on this landscape
In this dataset, heterogeneous computing patents are scoped to filings that combine CPU-GPU integration or accelerator coherence with a specific technical mechanism such as unified memory, task offloading, cache coherence protocols, interconnect bandwidth or a defined programming model, classified under G06F9, G06F13 or G06F15. That combination matters: a filing that merely mentions 'heterogeneous computing' in passing without claiming one of those mechanisms typically falls outside this scope. Practitioners running their own searches should expect broader keyword-only searches to return many more, less relevant results.
The filing trend in this dataset shows 23 families in 2017 falling to 7 by the 2022 midpoint and continuing down toward the most recent year. This pattern typically indicates that the foundational architectural claims — how a CPU and GPU share memory addressing, how coherence is maintained across accelerators — were established early, after which filing activity naturally shifts toward narrower refinements rather than new foundational claims. It does not necessarily mean interest in the underlying technology has declined; commercial deployment of heterogeneous systems has continued to grow even as core patent filing has slowed.
The most-cited records in this dataset include filings from Intel Corporation, among others, reflecting significant historical influence within this specific corpus. However, recent-year momentum data shows a marked pull-back: Intel's own filings show 0 in the latest year, a 100% year-on-year decline, and even smaller active filers show double-digit percentage declines. Anyone tracking current activity should look at recent-year filing counts specifically rather than relying on cumulative citation counts, which structurally favour older filings.
Based on IPC composition, the core G06F subclass covering data processing and interconnect architecture is heavily occupied at 126 of 127 records, while intersections with AI-model computing (G06N, 10 records) and image or video recognition (G06T and G06V, 12 records combined) are comparatively thin. This suggests that accelerator coherence and task-offloading claims specifically tailored to AI-model workloads or vision-recognition pipelines carry lower filing density than the general-purpose core, making them a more promising area to scope for new claims. Any white-space assessment should be paired with a full freedom-to-operate check against the dense core subclass.
In this dataset, the United States receives the largest share of filings at 46, followed by the European Patent Office at 24 and China at 14. India, with 12 filings, and WIPO PCT filings at 9 represent a meaningful share of total activity that is sometimes overlooked in freedom-to-operate work focused only on the US, Europe and China. A jurisdiction-by-jurisdiction filing strategy for this technology area should account for all of these offices rather than the top three alone.
Research Heterogeneous Computing and Accelerator Integration in depth with Eureka
Go past this page: query the whole heterogeneous computing and accelerator integration 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.