In-Memory Computing Accelerator Patent Landscape 2026
In-Memory Computing Accelerator Patent Landscape in 2026
The in-memory computing accelerator space is in a growth phase, with a French academic consortium anchored by CNRS holding the leading position and the field expanding sharply on a multi-year basis, though annual volume has eased from its 2023 peak. Activity is highly concentrated among a small number of academic and research institutions, with commercial players still a distinct minority.
French academic institutions lead a concentrated, research-driven field
The Centre National de la Recherche Scientifique (CNRS) holds the top position among filers, followed by the Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) and the Université de Lille. The top five filers account for 41% of the combined total across the hundred largest filers, signaling a notably concentrated competitive structure.
The top-tier gap is substantial: CNRS leads with 11 patent records, CEA follows with 9, and Université de Lille with 7, while the remaining ranked applicants hold 5 or fewer. This stepwise drop-off indicates that a tightly knit French academic cluster commands the field’s core, with no single commercial challenger yet matching that output.
| # | Applicant | Patent records | Share |
|---|---|---|---|
| 1 | Centre National de la Recherche Scientifique (CNRS) | 11 | |
| 2 | Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) | 9 | |
| 3 | University of Lille | 7 | |
| 4 | JUNIA | 5 | |
| 5 | Université Polytechnique Hauts-de-France | 5 | |
| 6 | International Business Machines Corporation (IBM) | 4 | |
| 7 | The Trustees of Columbia University in the City of New York | 4 | |
| 8 | Mentium Technologies Inc | 4 | |
| 9 | Aix-Marseille University | 4 | |
| 10 | Institute of Computing Technology, Chinese Academy of Sciences | 3 |
| # | Applicant | Patent records | Share |
|---|---|---|---|
| 11 | The Arizona Board of Regents on behalf of the University of Arizona | 3 | |
| 12 | Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences | 2 | |
| 13 | Nanjing University | 2 | |
| 14 | KOREA ADVANCED INST OF SCI & TECH | 2 | |
| 15 | Qualcomm Inc | 2 | |
| 16 | HUAZHONG UNIV OF SCI & TECH | 2 | |
| 17 | TECHNION RES & DEV FOUND LTD | 2 | |
| 18 | Beihang University | 2 | |
| 19 | Hangzhou Nano Core Chip Electronic Technology Co Ltd | 2 | |
| 20 | Université Polytechnique Hauts-de-France | 1 |
The leaders’ positions imply that foundational architecture and algorithm-level IP is being shaped primarily in academic settings, which may create licensing or collaboration entry points for commercial actors seeking to build on this base rather than develop from scratch.
The most recent 18–24 months of data are subject to publication lag and likely undercount actual filing activity; trend reads for 2024–2026 should be interpreted with that caveat. Longer-window growth, applicant concentration, and technology-route coverage are therefore more reliable signals than the latest-year bar alone.
Multi-year growth is strong, with AI inference dominating the technology mix
Two charts together reveal a field that has grown sharply since 2019 and is anchored in neural-network and digital-processing IP, with memory-device classes playing a secondary but meaningful role.
Annual filing trend
Filings were negligible before 2019, then accelerated through 2022 and peaked in 2023 at 18 patent records before easing to 12 in 2024 and 8 in 2025. The 2024–2026 figures are understated due to publication lag, so the apparent post-2023 softening should not be read as a genuine decline.
↗ Hover for values · click a bar to ask EurekaTechnology composition
G06N (AI model computing) and G06F (electric digital data processing) together dominate the branch mix, reflecting the field’s focus on neural-network inference acceleration. G11C (static and digital memories) is the next most represented branch, capturing the hardware memory-device layer. Analogue computing (G06G), coding (H03M), image processing (G06T), and novel solid-state devices (H10N) each account for only a handful of records, marking them as under-served adjacent branches.
↗ Hover for values · click a bar to ask EurekaHighly cited patent families surfaced by the query
Citation-heavy patent families returned by the query. Use this section as citation context, not as a curated list of the most topic-specific patents.
Digital-analog hybrid system architecture for neur…
A hybrid accelerator architecture consisting of digital accelerators and in-memory computing accelerators. A processor managing the data movement may determine whether input data is more efficiently processed by the digital accelerators or the in-memory computing accelerators. Based on the determined efficiencies, input data may be distributed for… (excerpt from the patent abstract)
| # | Patent | Citations |
|---|---|---|
| 1 | Hybrid processor | 92 |
| 2 | Hybrid processor | 18 |
| 3 | Memory cell for dot product operation in compute-i… | 11 |
| 4 | Microcontroller unit integrating an SRAM-based in-… | 9 |
| 5 | Programmable in-memory computing accelerator for l… | 9 |
| 6 | 一种存内计算加速器及其优化方法 | 8 |
| 7 | 一种针对存算一体芯片的快速编程方法 | 6 |
| 8 | 存算一体芯片、指令调度方法及相关装置 | 4 |
Ranked by total forward citations. Citation counts favour older and broadly cited patent families, and broad or adjacent patents may appear when they match the search scope. Treat this section as citation context, not as a curated list of the most topic-specific patents. Some patent titles may be shown in their original, non-English language where an accurate translation could not be guaranteed.
What the competitive structure means for R&D investment decisions
The field’s growth trajectory, academic concentration, and tight collaboration network all shape where new entrants can realistically compete or partner.
Growth stage, easing from a 2023 peak
The lifecycle is classified as Growth, with recent three-year filings sitting well above the prior three-year window — a 375% increase. Annual volume has eased from its 2023 peak, but publication lag means the true current pace is likely higher than recorded. Entrants moving now still face a relatively open field compared with mature semiconductor domains.
Growth stageTop five filers hold 41% share among the top hundred
The top five filers account for 41% of combined output across the hundred largest filers, and all five are French academic or engineering institutions. Commercial players such as IBM and Qualcomm appear further down the ranking with 4 and 2 patent records respectively. This leaves a notable gap between academic IP depth and commercial filing activity that a focused industrial program could begin to close.
High concentrationA dense French academic consortium anchors co-filing activity
The most active co-filing pairs are CNRS with Université de Lille (7 joint records), CNRS with JUNIA (5), and Université de Lille with JUNIA (5), forming an interlocking northern-French cluster. CEA co-files with both CNRS and Université d’Aix-Marseille (4 records each). Outside France, Columbia University and the University of Arizona co-filed on 3 records, representing the most active transatlantic pairing.
Consortium-drivenChina leads by filing office; European and US coverage also significant
China is the leading filing jurisdiction with 21 patent records, followed by the United States with 11, EPO with 8, and WIPO with 5. France and Germany add 4 and 3 records respectively. The geographic spread suggests that both European foundational work and Chinese applied development are being protected, while PCT filings indicate some applicants are pursuing broad international coverage.
Multi-jurisdictionGo beyond the landscape: Eureka’s TRIZ Solution agent breaks down an R&D problem and returns patented concept solutions, each with a technical approach and cited patent & literature evidence.
| Applicant | Collaborator | Co-filings |
|---|---|---|
| Centre National de la Recherche Scientifique (CNRS) | University of Lille | 7 |
| Centre National de la Recherche Scientifique (CNRS) | JUNIA (Hautes Études d’Ingénieur) | 5 |
| University of Lille | JUNIA (Hautes Études d’Ingénieur) | 5 |
| JUNIA | University of Lille | 5 |
| JUNIA | JUNIA (Hautes Études d’Ingénieur) | 5 |
| JUNIA | Centre National de la Recherche Scientifique (CNRS) | 5 |
| Centre National de la Recherche Scientifique (CNRS) | Aix-Marseille University | 4 |
| Centre National de la Recherche Scientifique (CNRS) | Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) | 4 |
| Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) | Aix-Marseille University | 4 |
| The Trustees of Columbia University in the City of New York | The Arizona Board of Regents on behalf of the University of Arizona | 3 |
Co-filing pairs, ranked by the number of jointly-filed patent families.
CNRS and CEA lead on AI-inference architecture; IBM brings commercial depth
The leading filers entered the space recently as new entrants and have built their positions rapidly. Technology emphasis differs meaningfully between the top academic institutions and the lone major commercial player in the ranking.
Centre National de la Recherche Scientifique (CNRS)
CNRS holds the top position with 11 patent records, classified as a new entrant — meaning its entire recorded portfolio falls within the recent filing window. Its technical focus is squarely on G06N 3 (AI model computing), with secondary coverage in G06F 17 and G06F 7 (digital data processing). CNRS is the hub of the dominant French academic co-filing cluster.
patent records: 11International Business Machines Corporation (IBM)
IBM is the highest-ranked commercial applicant, with 4 patent records and a new-entrant trajectory. Its focus spans G06N 3 (AI model computing) and G06F 15 (digital data processing), with a notable presence in G06G 7 (analogue computers) — a combination that reflects IBM’s interest in hybrid analogue-digital in-memory inference engines. IBM’s commercial orientation distinguishes it from the predominantly academic field.
patent records: 4| Applicant | Recent (3 yrs) | Trend |
|---|---|---|
| Centre National de la Recherche Scientifique (CNRS) | 11 | ▲ new entrant |
| Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) | 9 | ▲ new entrant |
| University of Lille | 7 | ▲ new entrant |
| JUNIA (Hautes Études d’Ingénieur) | 5 | ▲ new entrant |
| JUNIA | 5 | ▲ new entrant |
| The Trustees of Columbia University in the City of New York | 3 | ▲ new entrant |
| Aix-Marseille University | 4 | ▲ new entrant |
| International Business Machines Corporation (IBM) | 4 | ▲ new entrant |
Under-served branches adjacent to the dominant AI-computing core
Several IPC classes appear at the margin of the current corpus, each representing a plausible extension of in-memory computing accelerator work that has not yet attracted concentrated filing activity.
H10N · Other electric solid-state devices
With only 1 patent record, H10N covers emerging solid-state device types — including memristors and phase-change elements — that are frequently cited as next-generation memory cells for compute-in-memory. The sparsity here is notable given that device-level innovation is a prerequisite for higher-density analog in-memory inference. Teams with materials or device fabrication capability could find relatively open ground by linking novel H10N device work to G06N inference architectures.
Search this in Eureka →G06T · Image data processing & generation
Only 1 patent record sits under G06T, despite image inference being one of the most commercially relevant workloads for in-memory accelerators. The dominant filers concentrate on model-level and architecture-level abstractions in G06N rather than application-layer image pipeline integration. An applicant targeting vision-AI edge deployment — autonomous systems or smart cameras — could differentiate by bridging the G06T image-processing layer with the in-memory compute substrate, a combination currently underrepresented.
Search this in Eureka →How leading filers differ by technology branch emphasis
Strength of each leader across the main technology routes.
| Player | G06N 3 · Computing based on AI models | G06F 15 · Electric digital data processing | G11C 11 · Static & digital memories | G06F 9 · Electric digital data processing | G06F 7 · Electric digital data processing |
|---|---|---|---|---|---|
| Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) | Strong · 9 | Absent | Strong · 6 | Absent | Moderate · 4 |
| Centre National de la Recherche Scientifique (CNRS) | Strong · 11 | Absent | Moderate · 3 | Absent | Moderate · 4 |
| Aix-Marseille University | Strong · 4 | Absent | Strong · 3 | Absent | Strong · 4 |
| The Trustees of Columbia University in the City of New York | Strong · 4 | Moderate · 2 | Absent | Strong · 3 | Moderate · 2 |
| The Arizona Board of Regents on behalf of the University of Arizona | Strong · 3 | Strong · 2 | Absent | Strong · 3 | Strong · 2 |
| International Business Machines Corporation (IBM) | Strong · 4 | Strong · 3 | Moderate · 1 | Absent | Absent |
| University of Lille | Strong · 7 | Absent | Absent | Absent | Absent |
Frequently asked questions
The corpus contains 54 patent families in scope. This is a relatively small but fast-growing field, with filings negligible before 2019 and a 375% increase in recent three-year activity versus the prior three-year window.
The Centre National de la Recherche Scientifique (CNRS) leads with 11 patent records, followed by the Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA) with 9. Both are French public research institutions and entered their entire recorded portfolios within the recent filing window.
Academic and public research institutions dominate. The top five filers — CNRS, CEA, Université de Lille, JUNIA, and Université Polytechnique Hauts-de-France — are all French academic or engineering bodies. IBM is the highest-ranked commercial applicant, with 4 patent records, followed by Mentium Technologies and Qualcomm.
China leads by filing office with 21 patent records, followed by the United States with 11 and the EPO with 8. WIPO PCT filings account for 5 records, indicating some applicants are pursuing multi-jurisdiction protection. France and Germany add 4 and 3 records respectively.
G06N (computing based on AI models) and G06F (electric digital data processing) together account for the large majority of patent records, reflecting focus on neural-network inference acceleration at the architecture and algorithm level. G11C (static and digital memories) is the next most represented class, capturing the underlying memory hardware layer.
H10N (other electric solid-state devices, such as memristors and phase-change elements) and G06T (image data processing) each have only 1 patent record, despite their direct relevance to next-generation device substrates and vision-AI workloads respectively. G06G (analogue computers) and H03M (coding and code conversion) each show only 2 records, also representing sparse coverage relative to the dominant branches.
Ready to map your own in-memory computing accelerator landscape?
Join 18,000+ innovators using PatSnap Eureka to map any technology landscape: search 2B+ patents and papers, surface key assignees, and generate a report like this in minutes.
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.