Patent Activity in Berlin (2019–2024)

Executive Summary

What the European filing route reveals about patenting in Berlin, 2019–2024

Berlin appears in 6,085 DOCDB patent families whose earliest filing falls between 2019 and 2024 and which carry at least one person record geocoded to the NUTS region DE300 (Berlin) on a European patent application. That places the city seventh among the sixteen German Bundesländer on the same measure — behind Bayern (42,372), Baden-Württemberg (33,116) and Nordrhein-Westfalen (28,281), ahead of Hamburg (3,977) and Sachsen (3,196).

The more distinctive figure is the split between where inventions are made and who owns them. 5,198 families (85.4%) have at least one inventor with a Berlin address, but only 2,898 (47.6%) have an applicant with one. Among the three largest states that applicant share runs at 70–73% (Bayern 71.5%, Baden-Württemberg 70.3%, Nordrhein-Westfalen 73.3%), and Berlin is the lowest of the seven largest. Berlin’s R&D output is to a substantial degree filed by organisations registered elsewhere — the single largest owner of Berlin-invented families in the window is the Fraunhofer-Gesellschaft with 298 families, and not one of its applicant records carries a Berlin address.

Filing volume is flat-to-declining across the four complete years: 1,289 families in 2019 against 1,104 in 2022, a change of −14.4%. The 2023 and 2024 figures (828 and 362) are provisional and are not evidence of a collapse; they reflect the publication lag plus the additional delay before a family’s European member enters the data.

The technology profile is led by pharmaceuticals and medical devices — A61K (medicinal preparations, 507 families), A61B (diagnosis and surgery, 352), A61P (therapeutic activity, 351), A61N (electrotherapy and radiation therapy, 316) — alongside a communications and computing block: H04L (366), G06F (335), H04W (307). The steepest period-on-period movements are in G16H healthcare informatics (31 → 84 families, +171.0%) and A61N electrotherapy (86 → 143, +66.3%), against a scope-wide change of −9.8% between the same two periods.

Berlin filers use the international route heavily: 57.7% of families include a PCT (WO) member, 53.4% reach the USPTO and 39.5% reach China. Only 31.2% include a German national (DPMA) member.

Reading the German national register directly closes part of what the European route cannot show. Twenty-two weekly DPMA register extracts for Berlin yield 654 patent families first filed 2019–2024, of which 478 (73.1%) lie outside the geocoded scope above — among them 62 of the 69 recovered families of IAV, one of the city’s largest automotive engineering employers, which the European route shows with 14. That figure is a floor rather than a total; the method and its limits are set out in Beyond the European Route.

Berlin reads as an invention location more than an ownership location. The gap between 85.4% inventor coverage and 47.6% applicant coverage is the report’s central observation, and it is visible again in the applicant ranking: the largest Berlin-registered filers are BIOTRONIK (390 families) and Bundesdruckerei (282), while the largest owners of Berlin-invented work are Fraunhofer (298), Siemens AG (260) and Volkswagen (253), none of which files from a Berlin address.

One boundary governs every number in this report. Regional geocoding in PATSTAT exists only on European (EP) applications. German national applications filed at the DPMA carry nuts_level = 0 — no region at all. Every figure here is therefore a count of Berlin activity that took the European route, not a census of Berlin patenting. The postcode fallback that would normally close part of that gap does not work on this edition: of all applications filed 2019–2024, not a single DPMA application has a person record from which a German postcode can be extracted (see Methodology). The one route that does reach those filings is the DPMA register itself, and it reaches them for applicants only — every inventor figure in this report stays a European-route figure.

Annual Filing Trend

Patent families by year of earliest filing, split by whether Berlin appears on the applicant side, the inventor side, or both.

* 2023 and 2024 are provisional. PATSTAT Global, Spring 2026 (patstat_2026a) carries the usual ~18-month publication lag; on top of that, a family only becomes visible to this report once its European member has been filed and published, which adds a further delay for the most recent priority years.

Across the four complete years the series moves from 1,289 families in 2019 to 1,104 in 2022 (−14.4%), with the decline concentrated in the inventor-side count (1,127 → 938, −16.8%). The applicant-side count runs the other way over the same years (539 → 608, +12.8%), so the share of Berlin-linked families that are also Berlin-owned rises from 41.8% to 55.1%.

Data table: Annual filing trend 2019–2024

Year	All Berlin-linked families	Year-on-year	With Berlin inventors	With Berlin applicants	Both roles
2019	1,289	—	1,127	539	377
2020	1,285	−0.3%	1,141	540	396
2021	1,217	−5.3%	1,032	616	431
2022	1,104	−9.3%	938	608	442
2023*	828	−25.0%	691	402	265
2024*	362	−56.3%	269	193	100
2019–2024	6,085	—	5,198	2,898	2,011

A family is counted once, in the year of its earliest filing across all family members, not in every year in which it has a filing. The inventor and applicant columns are not exclusive: 2,011 families carry Berlin on both sides, so the two columns sum above the total.

Berlin in the German Context

The same measurement applied to all sixteen Bundesländer: families first filed 2019–2024 with a NUTS-geocoded applicant or inventor on a European application.

Berlin ranks seventh of sixteen by volume, at 14.4% of Bayern’s count. The ranking by volume is unremarkable; the applicant share is not. In Berlin, 47.6% of the families carry a locally registered applicant, against 71.5% in Bayern, 70.3% in Baden-Württemberg and 73.3% in Nordrhein-Westfalen. It is the lowest of the seven largest Bundesländer. Five sit lower overall — Brandenburg (31.5%), Mecklenburg-Vorpommern (38.5%), Sachsen-Anhalt (40.9%), Bremen (44.4%) and Sachsen (47.3%) — and every one of them ranks ninth or lower by volume.

Read together with the applicant tables that follow, the pattern is consistent with a city whose inventive capacity is distributed across sites and institutes of organisations headquartered elsewhere. This report cannot separate that explanation from a second one — that Berlin-registered applicant addresses are simply recorded less often — so it is offered as a reading of the data, not a conclusion.

Data table: Patent families by Bundesland 2019–2024

Rank	Bundesland	NUTS-1	Families	With local applicant	Applicant share	With local inventor
1	Bayern	DE2	42,372	30,286	71.5%	34,160
2	Baden-Württemberg	DE1	33,116	23,284	70.3%	28,416
3	Nordrhein-Westfalen	DEA	28,281	20,719	73.3%	23,670
4	Hessen	DE7	13,127	6,834	52.1%	11,474
5	Niedersachsen	DE9	11,555	7,080	61.3%	9,995
6	Rheinland-Pfalz	DEB	8,502	5,424	63.8%	7,334
7	Berlin	DE3	6,085	2,898	47.6%	5,198
8	Hamburg	DE6	3,977	2,740	68.9%	3,301
9	Sachsen	DED	3,196	1,513	47.3%	3,048
10	Schleswig-Holstein	DEF	3,012	1,498	49.7%	2,672
11	Brandenburg	DE4	2,126	669	31.5%	1,925
12	Thüringen	DEG	1,965	1,109	56.4%	1,645
13	Saarland	DEC	912	492	53.9%	723
14	Sachsen-Anhalt	DEE	883	361	40.9%	787
15	Mecklenburg-Vorpommern	DE8	684	263	38.5%	580
16	Bremen	DE5	664	295	44.4%	568

The Bundesland column is not exclusive: a family with inventors in Berlin and an applicant in Bayern is counted in both rows, so the column sums to more than the distinct number of German families. The rows are comparable with each other, not additive.

Who Patents in Berlin?

Two different questions, two different answers: which organisations file from Berlin, and which organisations own the work done in Berlin.

Six of the twelve organisations that own the most Berlin-invented families — Fraunhofer, Siemens AG, Volkswagen, BSH Hausgeräte, Siemens Energy and Siemens Mobility — hold zero families under a Berlin applicant address. Fraunhofer alone accounts for 298 families with a Berlin inventor while filing every one of them from its Munich headquarters address. The two rankings overlap in only four names: BIOTRONIK, Bundesdruckerei, Atotech and the Charité.

Data table: Largest Berlin-registered applicants

#	Applicant	Type	Families with Berlin applicant record	Families with a Berlin inventor
1	BIOTRONIK SE & Co. KG	Company	390	244
2	Bundesdruckerei GmbH	Company	282	200
3	Charité – Universitätsmedizin Berlin	University	89	83
4	Atotech Deutschland	Company	87	87
5	Technische Universität Berlin	University	82	75
6	Pfeiffer Vacuum Technology AG	Company	76	0
7	Bombardier Transportation GmbH	Company	66	15
8	CARIAD SE	Company	63	16
9	Brose Fahrzeugteile SE & Co. KG, Bamberg	Company	61	1
10	Freie Universität Berlin	University	51	49
11	Max-Delbrück-Centrum für Molekulare Medizin	Research	44	42
12	Bayer (group)	Company	37	180
13	Wella Germany GmbH	Company	36	1
14	B. Braun New Ventures GmbH	Company	34	0
15	Hach Lange GmbH	Company	32	22
16	Berlin Heart GmbH	Company	24	20
17	Forschungsverbund Berlin e.V.	Research	22	20
18	Häfele Berlin GmbH & Co KG	Company	20	10
19	Helmholtz-Zentrum Berlin für Materialien und Energie	Research	19	15
20	Bellevue Investments GmbH & Co. KGaA	Company	17	17

Data table: Largest owners of Berlin-invented families

#	Applicant	Type	Families with a Berlin inventor	Families with a Berlin applicant record
1	Fraunhofer-Gesellschaft	Research	298	0
2	Siemens AG	Company	260	0
3	Volkswagen AG	Company	253	13
4	BIOTRONIK SE & Co. KG	Company	244	390
5	Bundesdruckerei GmbH	Company	200	282
6	Bayer (group)	Company	180	37
7	BSH Hausgeräte GmbH	Company	172	0
8	Siemens Energy Global GmbH & Co. KG	Company	156	0
9	Siemens Mobility GmbH	Company	131	0
10	Deutsche Telekom AG	Company	127	0
11	Atotech Deutschland	Company	87	87
12	Charité – Universitätsmedizin Berlin	University	83	89
13	Technische Universität Berlin	University	75	82
14	Otis Elevator Company	Company	67	0
15	HERE Global B.V.	Company	51	0
16	Freie Universität Berlin	University	49	51
17	Robert Bosch GmbH	Company	48	0
18	Max-Delbrück-Centrum für Molekulare Medizin	Research	42	44
19	Wobben Properties GmbH	Company	41	0
20	Corning Research & Development Corporation	Company	34	0

Data quality: applicant names and applicant addresses. Three effects are visible in this scope and each of them was measured, not assumed.

Harmonised names merge different organisations. The applicant record Technische Universität Berlin (82 families here) carries the PATSTAT harmonised name TECHNISCHE UNIVERSITAT MUNCHEN — the same value the Munich university carries. Every figure in this report is built on person_name, with harmonised names used only as a cross-check, precisely because of this.
The same organisation appears under several spellings. Volkswagen files as both Volkswagen Aktiengesellschaft and Volkswagen AG, the Fraunhofer-Gesellschaft under two records, Atotech as Atotech Deutschland GmbH & Co. KG and ATOTECH Deutschland GmbH. Variants were consolidated before counting, and families were counted distinctly rather than summed — the two Fraunhofer records show 250 and 73 families but only 303 distinct ones.
An applicant address is a filing address, not a headquarters. The clearest case is Brose Fahrzeugteile SE & Co. Kommanditgesellschaft, Bamberg (61 families), which carries a Berlin geocode while naming another city in the applicant name itself. Pfeiffer Vacuum Technology AG (76 families) shows the same symptom without the naming evidence: a Berlin applicant geocode and zero families with a Berlin inventor. Read the applicant-side count as “filed from a Berlin address”, not as “headquartered in Berlin”.

Beyond the European Route

What the German national register adds to a picture the NUTS geography cannot draw

Every figure in the preceding sections rests on European applications, because that is the only filing route for which PATSTAT carries a region. The consequence is not a rounding error: applicants who file only at the German Patent and Trade Mark Office have nuts_level = 0 and are therefore not merely undercounted, they are absent. To measure what is missing rather than assert it, the DPMA register itself was read for Berlin — 22 weekly register extracts between 15 November 2024 and 14 August 2026, yielding 2,244 distinct Berlin register records after removing 358 cross-week duplicates.

Those records carry their application numbers, so they can be matched back into PATSTAT exactly, without relying on applicant-name harmonisation. Of the 2,244, one set is genuinely national — 824 patent applications, 297 Gebrauchsmuster (utility models) and 35 PCT entries into the German phase. The remaining 914 twelve-digit records are German validations of European patents, which PATSTAT holds as EP applications and which therefore say nothing new about the national route. The national set resolves to 654 patent families first filed 2019–2024.

478 of the 654 families (73.1%) lie outside the NUTS-anchored scope the rest of this report is built on. They are not a correction to it — they are a population it cannot reach. Only 186 of the 654 have a European family member at all, and 177 of the 654 are Gebrauchsmuster, an instrument that leaves no European trace by construction. Note also what the year profile does not do: where the European-route series falls away in 2023 and 2024 under the publication lag, the recovered series peaks in 2023 (139 families). German national applications publish 18 months after filing without waiting for a European member, so this route sees recent years the other one cannot yet.

Who the European route does not see

The gap is not spread evenly. It is concentrated in engineering service providers, mechanical and construction firms, and in the utility-model portfolios of companies that are otherwise well represented.

Berlin applicants in the register-recovered set, 2019–2024. “Outside scope” counts families absent from the NUTS-anchored set.

Applicant	Recovered families	Outside scope	of which Gebrauchsmuster
IAV GmbH Ingenieurgesellschaft Auto und Verkehr	69	62	1
Bundesdruckerei GmbH	66	8	1
BIOTRONIK SE & Co. KG	23	22	21
Bombardier Transportation GmbH	10	7	1
HPS Home Power Solutions AG	9	3	0
Helmholtz-Zentrum Berlin für Materialien und Energie	9	6	1
Veridos GmbH	9	5	0
Brose Antriebstechnik GmbH & Co. KG, Berlin	8	8	0
Deutsche Bahn AG	7	6	0
bett1.de GmbH	7	6	7
Berlin Heart GmbH	6	1	1
Betterguards Technology GmbH	6	1	0
ecoworks GmbH	5	5	1
Häfele Berlin GmbH & Co KG	5	1	5

IAV, one of Berlin’s largest automotive engineering employers, holds 14 families in the European-route view and 69 here — 62 of them outside that view, so for this one firm the register more than quintuples what is visible. The Berlin arm of Brose Antriebstechnik shows 9 inside the scope and 8 more of which every one is outside it. The pattern reverses for firms that file internationally as a matter of course: Bundesdruckerei holds 282 families inside the scope and gains 8. BIOTRONIK is the third case — 394 families inside the scope, and the 22 recovered here are almost entirely (21 of 22) Gebrauchsmuster, a parallel protection layer the European route never shows.

A different technology profile

The subclasses that dominate the recovered set are not the ones that dominate the European-route set. Leading are G06F electric digital data processing (37 families), B42D printed matter and security documents (32), G01N investigating or analysing materials (27), A61N electrotherapy and radiation therapy (23) and A61B diagnosis and surgery (23), followed by H04L digital transmission (22), A61F prostheses (18) and G06Q ICT for administrative purposes (17). Below those sit subclasses that do not appear in the European-route ranking at all: E04B general building constructions (13), F24F air-conditioning and ventilation (13), B65D containers (13), A47C chairs, sofas and beds (12) and A61G patient transport (12).

The absence is as informative as the presence: neither A61K medicinal preparations nor A61P therapeutic activity — the two largest subclasses of the European-route set, with 507 and 351 families — appears among the fourteen most frequent subclasses here. Pharmaceutical filings take the international route as a matter of commercial necessity; construction, furniture and ventilation filings frequently do not. The European route is not a random sample of Berlin patenting, and this is the shape of its bias.

Data table: Register-recovered families by first filing year

First filing year	Outside the NUTS-anchored scope	Already inside it	Total recovered
2019	79	31	110
2020	60	40	100
2021	74	34	108
2022	80	34	114
2023	116	23	139
2024	69	14	83
Total	478	176	654

These 478 families are a floor, not a total. Three limits set it. First, the register extract records movement — publications, grants, transfers, lapses — not filings, so an applicant whose files happened to be quiet across the harvest window does not appear. Second, the harvest covers 22 weeks of an available range that was measured at at least 130 — extracts still resolve for 16 February 2024, while 15 November 2023 returns none — and the count of newly seen applicants per extract had not flattened by the last one (43 new names in the final extract), which means the stream was not yet drained. Third, the register resolves regions for applicants only — the inventor figures in this report remain European-route figures and are not repaired by this method. A larger harvest would raise the number; nothing here would lower it.

Key Players Over Time

Annual family counts for the five organisations with the largest Berlin-linked portfolios in the window.

* 2023 and 2024 are provisional (publication lag plus the delay before the European family member enters the data). No trend statement below rests on those two years.

Over the four complete years the five portfolios move in different directions. Volkswagen falls from 97 families in 2019 to 20 in 2022 and Fraunhofer from 76 to 31, while BIOTRONIK ends the complete-year series at its highest value, 126 in 2022, after a 2020 low of 32. Siemens AG and Bundesdruckerei are comparatively steady (59 → 41 and 58 → 43). The 2022 figures for Volkswagen and Fraunhofer together are 51 families against 173 in 2019 — a change of that size in two of the five largest portfolios is enough to account for most of the −14.4% movement in the scope-wide series over the same years.

Data table: Key players by year

Organisation	2019	2020	2021	2022	2023*	2024*	Total
BIOTRONIK SE & Co. KG	67	32	64	126	76	29	394
Fraunhofer-Gesellschaft	76	90	63	31	29	14	303
Bundesdruckerei GmbH	58	74	50	43	24	33	282
Volkswagen AG	97	63	68	20	20	3	271
Siemens AG	59	53	46	41	49	12	260

Technology Landscape

IPC subclasses assigned to the Berlin-anchored European applications. Families carry several subclasses each, so the categories are not exclusive.

Health-related subject matter occupies four of the seven largest subclasses: A61K (medicinal preparations, 507 families), A61B (diagnosis and surgery, 352), A61P (therapeutic activity of compounds, 351) and A61N (electrotherapy, magnetotherapy and radiation therapy, 316). The communications and computing block follows with H04L (366), G06F (335) and H04W (307).

Between the two complete-year periods the scope as a whole moves 2,574 → 2,321 families (−9.8%). Against that baseline the clearest movements are G16H healthcare informatics, 31 → 84 (+171.0%) and A61N electrotherapy, 86 → 143 (+66.3%), with G06N (computing models based on biological or knowledge models, 54 → 73, +35.2%) and A61B (110 → 142, +29.1%) behind them. Moving the other way: H04N pictorial communication, 99 → 43 (−56.6%), B33Y additive manufacturing, 67 → 32 (−52.2%) and B29C shaping of plastics, 54 → 27 (−50.0%).

Data table: Top 14 IPC subclasses — period comparison

IPC	Subclass	Families 2019–2024	2019–2020	2021–2022	Change
A61K	Preparations for medical, dental or toiletry purposes	507	199	210	+5.5%
H04L	Transmission of digital information	366	158	135	−14.6%
A61B	Diagnosis; surgery; identification	352	110	142	+29.1%
A61P	Specific therapeutic activity of chemical compounds	351	133	150	+12.8%
G06F	Electric digital data processing	335	123	128	+4.1%
A61N	Electrotherapy; magnetotherapy; radiation therapy	316	86	143	+66.3%
H04W	Wireless communication networks	307	125	121	−3.2%
G01N	Investigating or analysing materials	297	142	112	−21.1%
G06N	Computing arrangements based on specific computational models	192	54	73	+35.2%
H04N	Pictorial communication, e.g. television	167	99	43	−56.6%
G06Q	ICT specially adapted for administrative or commercial purposes	163	71	58	−18.3%
C07K	Peptides	150	52	62	+19.2%
D06F	Laundering, drying, ironing, pressing or folding textiles	150	63	51	−19.0%
G16H	Healthcare informatics	150	31	84	+171.0%

Data table: IPC subclasses ranked 15 to 24

IPC	Subclass	Families 2019–2024	2019–2020	2021–2022	Change
C07D	Heterocyclic compounds	139	68	54	−20.6%
C12N	Microorganisms or enzymes; compositions thereof	134	49	62	+26.5%
B33Y	Additive manufacturing	123	67	32	−52.2%
G02B	Optical elements, systems or apparatus	109	45	46	+2.2%
B29C	Shaping or joining of plastics	105	54	27	−50.0%
H01L	Semiconductor devices — subclass retired on 2025-12-31, no title in the current IPC edition (see note below)	104	49	39	−20.4%
G06V	Image or video recognition or understanding	102	33	41	+24.2%
B61L	Guiding railway traffic; ensuring railway safety	100	38	43	+13.2%
B60L	Propulsion of electrically-propelled vehicles	98	49	40	−18.4%
G06T	Image data processing or generation, in general	98	43	39	−9.3%

A subclass without a title. H01L carries 104 families in this scope, and the current IPC edition (20260101) returns no title for it: the subclass was deprecated on 2025-12-31 after 57 years in force and replaced by H10D (inorganic electric semiconductor devices) and H10P. The label above was supplied by hand. Every one of those 104 families was filed before the deprecation, and a label lookup that discards untitled rows would have dropped them from this table without any error appearing. The counts here keep them.

Subclasses are not exclusive. A family usually carries several IPC symbols, so a family appears in as many rows as it has subclasses. The 24 rows above are also only the largest of a much longer list. Their sum, 4,915, is therefore neither a subset count nor a total and must not be compared with the 6,085 distinct families in the scope. Read each row against the total, never against the other rows.

Research Landscape

Universities, university medicine and non-university research institutes filing from a Berlin address — and who they file with.

The Charité (89 families) and the Technische Universität Berlin (82) lead the Berlin-registered public research applicants, followed by the Freie Universität Berlin (51) and the Max-Delbrück-Centrum für Molekulare Medizin (44). In each of these four portfolios between 91% and 96% of the families also carry a Berlin-based inventor (Charité 93.3%, TU Berlin 91.5%, Freie Universität 96.1%, Max-Delbrück-Centrum 95.5%) — the address on the application and the addresses of the people named on it agree, which is not the case for most of the corporate filers in this scope.

Two large research organisations are absent from this chart and present in the data: the Fraunhofer-Gesellschaft owns 298 Berlin-invented families and the Max-Planck-Gesellschaft 26, both filing under their central addresses outside Berlin. On an applicant-address reading, Berlin’s largest research portfolio is invisible; on an inventor reading, it is the largest in the city.

Data table: Berlin-registered research applicants

Institution	Type	Families filed from Berlin	Families with a Berlin inventor
Charité – Universitätsmedizin Berlin	University	89	83
Technische Universität Berlin	University	82	75
Freie Universität Berlin	University	51	49
Max-Delbrück-Centrum für Molekulare Medizin	Research	44	42
Forschungsverbund Berlin e.V.	Research	22	20
Helmholtz-Zentrum Berlin für Materialien und Energie	Research	19	15
Ferdinand-Braun-Institut, Leibniz-Institut für Höchstfrequenztechnik	Research	12	12
Deutsches Rheuma-Forschungszentrum Berlin	Research	11	10
Humboldt-Universität zu Berlin	University	11	9
Fraunhofer-Gesellschaft (filed from Munich)	Research	0	298
Max-Planck-Gesellschaft (filed from Munich)	Research	0	26
IHP GmbH – Innovations for High Performance Microelectronics	Research	0	8

Co-Applicant Pairs

Joint filings were derived per application, not per family, so that two spellings of one organisation cannot pair with each other. Pairs with at least three joint families are shown; name variants were consolidated before pairing.

Applicant	Co-applicant	Joint families
Seat, S.A.	Volkswagen AG	21
Charité – Universitätsmedizin Berlin	Max-Delbrück-Centrum	15
Bayer (group)	The Broad Institute, Inc.	15
Bayer (group)	Deutsches Krebsforschungszentrum	12
CARIAD SE	Volkswagen AG	12
Dana-Farber Cancer Institute, Inc.	The Broad Institute, Inc.	12
Bayer (group)	Dana-Farber Cancer Institute, Inc.	12
Deutsche Telekom AG	Technische Universität München	8
Charité – Universitätsmedizin Berlin	Freie Universität Berlin	6
Fraunhofer-Gesellschaft	Universität des Saarlandes	6
Bode – Die Tür GmbH	Brose Fahrzeugteile SE & Co. KG, Bamberg	5
Bayer (group)	Regeneron Pharmaceuticals, Inc.	4
F. Hoffmann-La Roche AG	Roche Diabetes Care GmbH	4
Fraunhofer-Gesellschaft	Volkswagen AG	4
BASF SE	Technische Universität Berlin	3
Bundesdruckerei GmbH	Freie Universität Berlin	3
Charité – Universitätsmedizin Berlin	Sapreme Technologies B.V.	3
Charité – Universitätsmedizin Berlin	Universidade de Santiago de Compostela	3
Freie Universität Berlin	Sapreme Technologies B.V.	3
Freie Universität Berlin	Universidade de Santiago de Compostela	3
Forschungsverbund Berlin e.V.	Ludwig-Maximilians-Universität München	3
Fraunhofer-Gesellschaft	Huawei Technologies Co., Ltd.	3
CARIAD SE	Robert Bosch GmbH	3

The strongest research pairing in the city is Charité with the Max-Delbrück-Centrum, 15 joint families, followed by Charité with the Freie Universität Berlin (6). Two three-way clusters are visible rather than isolated pairs: Bayer, the Broad Institute and the Dana-Farber Cancer Institute appear in three pairs among themselves (15, 12 and 12), and Charité, Freie Universität Berlin, Sapreme Technologies and the Universidade de Santiago de Compostela appear in four pairs of three joint families each — the signature of a small number of consortium filings rather than a broad partnership.

Residual name variants in the pair list. Consolidation cannot be complete. The Broad Institute appears both as The Broad Institute, Inc. and The Broad Institute Inc., and pairs with Bayer under both spellings (15 and 3 joint families). The two counts must not be added — the underlying families may overlap — so the pair is shown at its larger measured value only, and the true figure lies between 15 and 18.

Top Inventors

The most frequently named inventors with a Berlin address, and the applicant that appears most often alongside them.

The scope contains 5,109 distinct inventor name strings with a Berlin address. The top of that list is narrow: thirteen of the top eighteen names file most often with the Fraunhofer-Gesellschaft — the same 298-family portfolio that carries no Berlin applicant address anywhere in the data. Two further names file with Volkswagen (Pfadler 100, Jornod 76) and two with BIOTRONIK. A concentration this narrow means the inventor ranking describes a handful of research groups rather than the city as a whole.

Data table: Most frequently named Berlin inventors

#	Inventor (as recorded)	Families	Most frequent applicant
1	SCHIERL, Thomas	124	Fraunhofer-Gesellschaft
1	HELLGE, Cornelius	124	Fraunhofer-Gesellschaft
3	WIEGAND, Thomas	112	Fraunhofer-Gesellschaft
4	Pfadler, Andreas	100	Volkswagen AG
5	FEHRENBACH, Thomas	78	Fraunhofer-Gesellschaft
6	Jornod, Guillaume	76	Volkswagen AG
7	MARPE, Detlev	75	Fraunhofer-Gesellschaft
8	GÖKTEPE, Baris	73	Fraunhofer-Gesellschaft
9	SCHWARZ, Heiko	63	Fraunhofer-Gesellschaft
10	Wirth, Thomas	52	Fraunhofer-Gesellschaft
11	SKUPIN, Robert	45	Fraunhofer-Gesellschaft
11	HAUSTEIN, Thomas	45	Fraunhofer-Gesellschaft
13	SÁNCHEZ DE LA FUENTE, Yago	42	Fraunhofer-Gesellschaft
14	WIRTH, Thomas	41	Fraunhofer-Gesellschaft
15	DÖRR, Thomas	40	BIOTRONIK SE & Co. KG
16	DOERR, Thomas	37	BIOTRONIK SE & Co. KG
17	Selvanesan, Sarun	36	Fraunhofer-Gesellschaft
18	Brombach, Johannes	28	Wobben Properties GmbH

Inventor names are not harmonised in PATSTAT. This table shows the effect directly: rows 10 and 14 are Wirth, Thomas (52) and WIRTH, Thomas (41), rows 15 and 16 are DÖRR, Thomas (40) and DOERR, Thomas (37) — case and umlaut transliteration produce separate entries. The counts are therefore lower bounds per person, the ranking positions are approximate, and the figure of 5,109 counts name strings, not people. The two spellings are not added together here because the underlying families may overlap.

International Filing Strategy

Where the 6,085 Berlin-linked families are filed, and how far each cohort has travelled so far.

Every family in this scope has a European member by construction, so EP is 100% and carries no information. Of the offices that do: 57.7% of families include a PCT (WO) member, 53.4% reach the USPTO and 39.5% reach China, while only 31.2% include a German national (DPMA) member. For a set of patents anchored in a German city, the national office is reached by less than a third of the portfolio, and it is reached less often than the Chinese one.

Read the right-hand chart as elapsed time, not as strategy. Both cohorts end in a complete filing year, but a family’s national-phase members are filed 30 months or more after its priority and published later still, so the 2021–2022 cohort has simply had less time to arrive at the national offices. That is why every national-office share falls while the PCT share, which is decided early, is the only one that rises (60.3% → 62.3%). The falls of 8.1 points at the USPTO and 9.1 points in China are consistent with an incomplete cohort and are not evidence of a change in filing behaviour.

Data table: Offices reached by the Berlin family set

Office	Families	Share of 6,085	Utility-model members
EP — European Patent Office	6,085	100.0%	0
WO — PCT	3,508	57.7%	0
US — United States	3,252	53.4%	0
CN — China	2,402	39.5%	16
DE — Germany (DPMA)	1,901	31.2%	123
JP — Japan	1,222	20.1%	2
KR — Korea	800	13.1%	1
CA — Canada	655	10.8%	0
AU — Australia	486	8.0%	0
ES — Spain	345	5.7%	2
MX — Mexico	289	4.7%	0
IL — Israel	258	4.2%	0
TW — Taiwan	251	4.1%	0
BR — Brazil	244	4.0%	0
PL — Poland	194	3.2%	0

Data table: Cohort shares and grant rates by office

Office	2019–2020 cohort	Share of 2,574	2021–2022 cohort	Share of 2,321	Granted, 2019–2020 cohort
EP	2,574	100.0%	2,321	100.0%	1,253 — 48.7%
WO (PCT)	1,551	60.3%	1,445	62.3%	—
US	1,635	63.5%	1,285	55.4%	980 — 59.9%
CN	1,203	46.7%	872	37.6%	631 — 52.5%
DE	952	37.0%	706	30.4%	195 — 20.5%
JP	582	22.6%	479	20.6%	383 — 65.8%
KR	407	15.8%	296	12.8%	189 — 46.4%

Grant rates are cohort-limited and still incomplete. The grant column covers only the 2019–2020 cohort, because families first filed in 2021 or later are largely still in examination and would produce artificially low rates. Even for 2019–2020 the figures are floors, not final outcomes, because examination is still pending on part of every cohort. One possible explanation for the German national rate (20.5%) sitting below the Japanese one (65.8%) is a difference in pendency between the two offices; this report holds no data on pendency and cannot test it. The rate is measured as granted families over families present at that office, not over applications.

Offices are not exclusive: one family usually has members at several offices, so the columns sum far above 6,085. The utility-model column counts families with a UM member at that office; those members are excluded from every other figure in this report, which counts patents of invention (ipr_type = 'PI') only. The 123 German utility-model families are 2.0% of the scope.

Methodology

How the Berlin patent set was defined, what it counts, and where it stops.

Data basis, recomputed 15 August 2026. Every figure in this report was produced in one session against EPO PATSTAT Global, Spring 2026 (patstat-mtc.patstat_2026a) on Google BigQuery. Classification labels come from the current editions in force at that date: IPC 20260101 (in force 1 January 2026) and CPC 202608 (in force 1 August 2026), read from patstat-mtc.classification.dpt_edition. Two reports computed on different editions are not comparable, and the difference looks exactly like real change — which is why the edition is named here rather than assumed. One section draws on a second source alongside PATSTAT — the DPMA register, read for Berlin between 15 November 2024 and 14 August 2026; its method and limits are set out under The German National Register below.

Search strategy, family counting and data source

Search strategy — how Berlin was identified

This is a regional report. Its patent set comes from geography, not from a classification code list and not from an applicant name list.

Anchor. A person record in tls206_person with nuts = 'DE300' (Berlin) and nuts_level IN (3, 4), linked through tls207_pers_appln to a European patent application (appln_auth = 'EP', ipr_type = 'PI', docdb_family_id > 0), in either role.
Why both NUTS levels. nuts_level is a geocoding-source flag, not a hierarchy depth: 3 means EPO/ECOOM geocoding, 4 means OECD REGPAT. Both are genuine district-level codes. In this scope level 4 carries 5,885 families and level 3 carries 2,068; restricting to level 3 would discard the larger part of the evidence, and mixing in level 0 would mix ungeocoded records into a regional count. No NUTS code other than DE300 exists under the DE3 prefix, so Berlin as a Bundesland and Berlin as a district are the same set.
Role separation. A person’s role is a property of the (person, application) pair, never of the person. Applicants are applt_seq_nr > 0, inventors invt_seq_nr > 0; the two are reported separately throughout because they answer different questions — who owns the invention against where it was made.
Expansion. Once a family is anchored, all of its members at all offices are used for the office, grant and PCT figures. The appln_auth = 'EP' restriction applies only to the anchoring step.
Entity names. Applicant aggregation is built on person_name. The PATSTAT harmonised name han_name was used only as a cross-check, and it failed the check: the applicant record Technische Universität Berlin carries han_name = 'TECHNISCHE UNIVERSITAT MUNCHEN', which would have merged 82 Berlin families into a Munich entity. Variants were consolidated by hand and families counted distinctly, never summed across variants.

Family counting

Every count is COUNT(DISTINCT docdb_family_id). One invention filed in EP, WO and US is one family and three applications.
A family is assigned to one year: the minimum appln_filing_year across all of its members, computed before the 2019–2024 window is applied. Taking the minimum over the windowed members instead would pin families first filed before 2019 onto 2019 and inflate the first year; the annual columns would still sum to the total under either reading, so that error is not detectable by a sum check.
Design filings and utility models are excluded by ipr_type = 'PI', applied both when anchoring and when expanding. This is not boilerplate: 123 families in the scope (2.0%) do have a German utility-model member, and 16 have a Chinese one. Those members are counted nowhere except the utility-model column in the International Filing Strategy section.

Multi-assignment note

A family carries several IPC subclasses, so it appears in several rows of the technology table. Those 24 rows are also only the largest of a longer list, so their sum of 4,915 is neither a subset count nor a total and is not comparable with the 6,085 distinct families in the scope. Rows are comparable with the total, never with each other, and never additive.
A family reaches several offices, so the office table sums far above 6,085.
The applicant and inventor perspectives overlap: 2,011 families carry Berlin on both sides. For the same reason the Bundesland table is not additive — a family invented in Berlin and owned in Bayern appears in both rows.
No pie or doughnut chart is used anywhere in this report, because none of the breakdowns above is a parts-of-a-whole split.

Classification scope

The three code-set completions that a technology landscape requires — expanding each code to its subtree, unioning reclassification predecessors, and keeping deprecated symbols — do not apply here, because no classification code selects the patent set. IPC subclasses are used for description only, after the set is fixed by geography.

One consequence of the classification lifecycle still reaches this report, on the labelling side. Subclass H01L holds 104 families in the scope and was deprecated on 2025-12-31; in IPC 20260101 its row carries no title at all. Labels were joined with a LEFT JOIN so the row survives, and the title was supplied by hand. An inner join would have removed those 104 families from the technology table silently.

Data source and stack

EPO PATSTAT Global, Spring 2026 (patstat-mtc.patstat_2026a) on Google BigQuery.
Classification views: patstat-mtc.classification.dpt_ipc, joined on symbol_patstat.
Regional reference: tls904_nuts for Bundesland labels.
Fifteen SQL queries, all reproduced in queries.sql; every number in this report comes from one of them.
Analysis and visualisation: mtc.berlin patent intelligence pipeline with the patstat-mcp server and Claude.

Scope Limitations

The European route only. NUTS geocoding exists in PATSTAT only for European applications. Person records on German national (DPMA) filings carry nuts_level = 0 — no region. A Berlin applicant that never files at the EPO is invisible to this method, and no figure here should be read as a total for Berlin patenting.
The postcode fallback does not close that gap on this edition. The standard remedy is to extract postcodes from person_address. Measured: across all applications filed 2019–2024, the offices reached by German person records carrying an extractable Berlin postcode are EP (5,346 families at nuts_level 4, 2,077 at level 3, 241 at level 0 — rows that overlap and must not be added) plus a handful of small ones (IL, CL, HR, DK) — and zero DPMA applications. Of all German person records at nuts_level = 0, only 4.4% contain a five-digit postcode at all. The postcode route therefore re-covers the same European pool that NUTS already covers and adds nothing for the national route. The size of the DPMA gap cannot be measured from PATSTAT.
A residual geocoding gap inside the European route. 241 families with an EP application filed in 2019–2024 carry a person record with a Berlin postcode and no NUTS code at all. Some of those are already in scope through a second, geocoded person, so the figure is an upper bound on what the geocoding filter misses within EP — roughly 4% when held against the 6,085 families of the scope, which is counted on a different basis (year of earliest family filing).
Applicant address is not headquarters. The applicant-side count reads “filed from a Berlin address”. Organisations with a central filing address elsewhere — Fraunhofer, Siemens, Volkswagen, the Max-Planck-Gesellschaft — show zero Berlin applicant families while owning substantial Berlin-invented portfolios; conversely two applicants named after other cities carry a Berlin geocode.
2023 and 2024 are provisional. The ~18-month publication lag is compounded here: a family enters this report only once its European member has been filed and published. The complete-year series is 2019–2022, and every trend statement in this report is limited to it.
Grant rates are floors. They are reported for the 2019–2020 cohort only and still reflect pending examination.
Person names are not harmonised. Inventor counts are per name string, not per person; two spellings of the same name appear separately in the top-18 list. Applicant variants were consolidated by hand, and residual variants remain (the Broad Institute appears under two spellings).
No external comparison was made. Company and university reports are checked against Lens.org or Google Patents. No public counterpart exists for “patent families with a NUTS-DE300 person on an EP application, by earliest family filing year”, so the plausibility check here is internal: the same query run for all sixteen Bundesländer reproduces the expected national ordering and returns Berlin at exactly the 6,085 families counted everywhere else in this report.

The German National Register (DPMA)

The NUTS-anchored set above cannot reach applicants who never file at the European Patent Office, because PATSTAT geocodes person records on the EP route only. That population was measured directly rather than estimated, through the DPMA register.

Harvest. Twenty-two weekly register extracts for NUTS region DE300, spaced across the available extract range from 15 November 2024 to 14 August 2026. All twenty-two returned data; the extracts held between 18,986 and 73,214 records each. Of 2,602 matching records, 358 were the same file moving in more than one week, leaving 2,244 distinct records. Only applicants whose own region resolved to DE300 were kept — a matched record also carries its co-applicants from elsewhere, and those do not make the record a Berlin record.
Which records are national. The extract mixes five German number series without labelling them. Series 10 (824 records) are patent applications, 20 (297) Gebrauchsmuster, 11 (35) PCT entries into the German phase. Series 50 (532) and 60 (382) are German validations of European patents, which PATSTAT holds as EP applications; none of a 50-record sample resolved as a German application, and they are excluded. A further 174 records carried European or PCT numbers rather than German ones.
Bridge into PATSTAT. The 1,156 national records were matched on appln_auth = 'DE' and appln_nr — the application number, not the applicant name, so no name harmonisation enters the step. On a 40-record control sample the match resolved 34 records (85%); all six failures were 2024–2025 patent publications, and every record filed 2022 or earlier resolved. The register runs roughly a year ahead of the PATSTAT edition, which is what those failures are.
Counting. Family year is MIN(appln_filing_year) over all members. One detail differs from the rest of the report and is stated rather than hidden: elsewhere the minimum is taken over ipr_type = 'PI' members only, which cannot be done here — 177 of these families have no patent member at all, so that restriction would delete them instead of dating them. The minimum here therefore runs over patent and utility-model members alike. Gebrauchsmuster are counted and reported separately and are never added into a patent-family total: the ipr_type = 'PI' filter used elsewhere would have discarded 177 of the 654 recovered families in silence.

What the 478 figure is, and is not. A register extract records file movement — publications, grants, transfers, lapses — not new filings, so the recovered set is the portfolio of applicants who were active in the register during the harvest window, not a census of Berlin applicants. The harvest covers 22 weeks of an available range measured at at least 130 (an extract still resolves for 16 February 2024; 15 November 2023 returns none), and the number of newly seen applicants per extract was still 43 in the final extract, so the stream was not drained: the curve measures how much of the extract stream was read, never how much of the region was found. And the register resolves regions for applicants only — no inventor figure in this report is affected by it. The number is a floor.

Region assignment in this section uses the postcode-to-NUTS crosswalk © EuroGeographics / Eurostat (pc2025_DE_NUTS-2024_v1.0), CC-BY-SA 4.0. The register service returns regions only; no postal addresses were retrieved or stored.

Glossary — patent terms explained

Patent family (DOCDB): A group of patent applications protecting the same invention in different countries. Counting families instead of applications avoids counting one invention many times.
Applicant vs. inventor: The applicant owns the application; the inventor created the invention. In PATSTAT the role is recorded per application, not per person, so the same person can be both.
NUTS: The EU’s regional coding system. DE3 and DE300 are both Berlin, which is simultaneously a Bundesland and a district.
nuts_level: Not a hierarchy depth but a flag for how the code was derived: 0 = not geocoded, 3 = EPO/ECOOM geocoding, 4 = OECD REGPAT. Levels 3 and 4 are both district-level codes.
IPC subclass: A four-character technology code such as A61B. One family usually carries several.
PCT / WO application: An international application under the Patent Cooperation Treaty, filed once and later continued in individual countries.
ipr_type = PI: Patent of invention — the filter that keeps designs and utility models out of the counts.
Gebrauchsmuster (utility model): A German registered right granted without substantive examination, held in PATSTAT as ipr_type = UM. It has no European equivalent, so it never appears on the EP route.
Register extract: A periodic DPMA file listing every register record that moved in a period — published, granted, transferred or lapsed. It records movement, not new filings.
Provisional year: A recent filing year whose applications are not yet fully published, so its count will rise in later PATSTAT editions.

Report Files

This report consists of meta.json, cover.html, kpi-strip.html, twelve section files under sections/, and queries.sql. The SQL file carries the PATSTAT and classification editions in its header and contains all eighteen queries in the order in which they are numbered, so every figure shown here can be reproduced against the same dataset.

Like what you see?

This report was built with a fully reproducible pipeline: EPO PATSTAT Global on BigQuery, a custom MCP server, and Claude AI for analysis and visualization.

Regional Patent Intelligence Need the same picture for your region, cluster or Bundesland — including the parts NUTS geocoding does not reach? We build tailored regional patent reports.

The Toolchain Want to run your own AI-powered patent analyses? We help you set up PATSTAT on BigQuery with MCP server integration.

Arne Krueger — LinkedIn Book a Call depa.tech mtc.berlin