Introduction
In the context of the contemporary world economy where knowledge has become king, the advent of an age of great techno-geopolitical uncertainty has begun due to the fundamental shift that has taken place in relation to intellectual property rights in terms of their creation, management, and enforcement. The post-physical capability of semiconductor manufacturing and AI beyond just software has led to patents becoming the most crucial instruments for geostrategic manoeuvring and organizational survival. By the year 2025, the number of international patent applications filed through the Patent Cooperation Treaty of the World Intellectual Property Organization had reached 275,900 thanks to substantial progress in communications and semiconductor technologies.
Macro Level Patent Activity and Global Technology Diffusion Dynamics
The contemporary intellectual property landscape has been marked by the growing rate of technological diffusion, which is the process of dissemination of vital technologies through innovators to other firms within the industry. In its report on the world intellectual property, WIPO mentions the four key factors that determine the diffusion speed of a particular technology: technology-specific attributes, the flow of information, absorptive capacity, and favourable policies/property regime.
While the use of advanced technology such as artificial intelligence enables software inventions to become available in global markets within days of their introduction, the knowledge of the invention remains concentrated in several countries, namely the US, Western Europe, and Eastern Asia. It is in this way that intellectual property innovations have become geographically concentrated, resulting in an arms race witnessed in international IP filings rates.
However, the high rate at which there have been applications for international designs and the comparatively slow decline in the number of international trademarks that have been made through the Madrid system suggests a significant shift in strategy among firms when it comes to intellectual property management. For example, major technology firms such as Huawei Technologies, Beijing Xiaomi Mobile Software, and Apple have seen an increase in the number of design applications that aim to protect the design of consumer electronics and special hardware interfaces. In addition, the marked increase by a factor of 70 percent in dispute resolution at the WIPO Arbitration and Mediation Centre suggests that the existing legal framework for dealing with technology licensing disputes is no longer relevant.
Advanced Transistor Architecture Transitions and Asymmetric Litigation Risks
Moving from 2D to 3D transistors through physical migration in the design of transistors has seen some effects on the legal considerations of patent disputes within the semiconductor industry. The main type of transistor used at the initial stages of the microelectronics industry is the planar MOSFET transistor that acted like a two-dimensional switch with respect to Dennard Scaling economy principle such that reducing the size of the physical structure would improve efficiency while saving power. As a result, no need to worry about the structure of the transistor in past patent disputes. With further shrinking of transistors below 20nm, the planar designs have physical constraints as far as poor electrostatic control and leakage current is concerned.
To address the above problem, there is the advancement of transistor technology through incorporation of 3D structures in the designs starting with FinFET to GAA nanosheets followed by forksheet/CFET stacking.
Such changes make a significant impact on the dynamics of litigation because it creates substantial disparities between the plaintiff and defendant sides as well as between the owners of technical information related to the case. In such cases, patent suits would be brought against the consumer-oriented product manufacturers who sell products with the accused chips. However, the parameters of process used in defending the case will remain with the foundries and material providers who are not a part of the case at all. Given the absence of process parameters from the side of accused companies, litigation of such complicated processes will require extensive discovery of third-party data. Litigation would necessitate access to very confidential information about settings of tools, deposition plans, and metrology of the foundry and tool providers.
Further, as complexity rises, so too does the challenge facing destination patents that seek to protect an entire class of complex architectures, like GAA, but fail to provide an adequate disclosure of an effective way to mass-produce the invention. Defendants have successfully defeated the validity of patents by showing how such patents did not include instructions or processes of making, such as how to get the selectivity window, handle defects, perform the etch process, and other manufacturing processes required for mass-producing the architecture.
Indeed, as these new architectures include terminologies created in the planar era to describe new types of 3D architectures, the issue of claim construction has become central. The issue is even more challenging when it comes to apportionment, where disputes arise as parties seek to prove their share of contribution to the economic success of the inventions.
It becomes even harder to navigate the perilous situation because of the presence of patent backlogs and ongoing process changes within the administrative parts of the US patent system. The patent application backlog at USPTO amounted to more than 800,000 pending applications awaiting examination in early 2025. As a result of such numbers and the lack of examiners, Track One application is becoming ever more popular because of the necessity of protecting the core product features until they are leaked.
On the other hand, changes introduced in PTAB made the institution pro-patent, thus making the option of taking advantage of IPRs as an inexpensive alternative to lawsuits in federal court unavailable. This forces patent lawyers to adopt the technique of inventive claim drafting that includes using means-plus-function limitation, structure disclosure, and sequence-based structure definition in lieu of vulnerable functional claims. This way, they will be able to develop protective shields both from PTAB invalidations and enablement defence under Section 112.
The Generative AI Intellectual Property Explosion: Algorithmic and Jurisdictional Frontiers
In addition to the hardware race, there is a current transformation in the structure of the generative AI industry. It is followed by the explosion of patent applications and shifts in regulations. In the time frame between 2014 and 2023, the number of GenAI patent families grew from 733 to more than 14,000 with more than 25% of all GenAI patents and 45% of GenAI scientific publications appearing in 2023. This development comes as a consequence of the emergence of the transformer model.
The accelerated implementation of the models in practice results in an extremely favourable environment for the creation of software patents in the United States. Due to the leadership change at USPTO, the strict abstract idea rejection criteria that followed Alice’s decision were abandoned. Instead, new guidance documents such as the SMED framework and Kim Memo emphasize the importance of considering the Technical Delta brought about by the AI system.
In case there is some improvement on the technical working of the computer process and any complex computation that is done by AI which is not possible to carry out manually by humans, it would be easier to get such a patent approval. In addition, according to patenting procedures of the European Patent Office and the Chinese patent office, patenting of AI will be approved in cases where AI software contributes technically to the system and produces technical outputs.
Currently, the patenting rules favour patenting of “vertical AI” rather than general submission of generic neural networks patents. In 2026, valuable AI portfolios will include AI applications in life sciences, financial technology, clean technology, and AI Agentic Loop Processes for automation of supply chains.
Contrastingly, the number of lawsuits concerning over 1,000 AI patents litigated in various jurisdictions around the globe since 2020 has risen to an unprecedented extent, and in more than 70% of these cases, infringement was alleged on deep learning technologies. The major challenge in terms of proving and non-proving infringement is linked to the confidential nature of deep learning technology codes and data.
Nevertheless, in order to surmount this confidentiality problem, firm-affiliated plaintiffs have resorted to utilizing publicly accessible sources including publications and speeches containing mentions of transformer layers and training data in order to meet the pleadings requirements. Yet again, the courts have given plaintiffs leeway to amend their pleadings easily after having accessed new knowledge through the discovery process. Despite all that, proving infringement comes at a cost.
Brain Inspired Paradigms: Neuromorphic Computing and Edge AI Patent Waves
With traditional von Neumann computing architecture failing in handling the heat and power issues associated with large-scale AI workloads, the emergence of neuromorphic computing is becoming a growth opportunity for semiconductors in terms of patents. The term neuromorphic computing refers to computer systems that emulate the structure and operational mechanism of the human brain using event-based Spiking Neural Network (SNN), providing 100 to 1000 times more energy efficiency than GPUs.
A total of 596 patents have already been issued in neuromorphic chip architecture by early 2026, reflecting the remarkable increase in 2025 by 401%, which means the transition from prototype design to real-life production.
Regarding the technological features, the primary purpose of the issued patents lies in addressing the challenge of the memory bus in order to solve the issue of separating memories and processors. In contrast to the usual process of instruction processing at regular intervals of fixed clock cycle, neuromorphic processors use asynchronous spikes only at spike event arrival. Such a spike-based operation uses direct implementation of the Leaky Integrate-and-Fire (LIF) neurons, where a core with 256 LIF neurons fits in 0.12mm.
In order to improve communication within the chip, there has been a shift in the patent application trend toward the interconnect fabric. Newer systems have implemented a hybrid mesh-hierarchical routing scheme with energy efficiency of up to 86% and spike latency reduced by 55%, as compared with a flat network system such as NeuToMa, which is a topology-aware mapping toolchain.
In neuromorphic patents, there is a focus on Spike-Timing-Dependent Plasticity (STDP), wherein the synapses can perform unsupervised learning, reducing computation to 78%, as the unnecessary pathways can be neglected.
The two main groups operating in the neuromorphic sector include those who manufacture at a large scale and research institutions. IBM and Samsung Electronics are the top two manufacturers across the globe and possess rich patent libraries, covering both synaptic memory cells and multi-chip neuromorphic systems.
However, the academic institutions of China, such as Tsinghua University and Peking University, are not left behind with regards to their domestic neuromorphic ecosystems. In particular, Tsinghua University leads among all other companies when it comes to research technology impact, thanks to its impressive average of 4.9 forward citations per patent. This shows that there is much effort being put towards the future post-GPU hardware paradigm by the state-sponsored research in China.
Specialised Compute Accelerators and NPE Aggression in Federal courts
Financial motivation has ensured that tech companies strive to participate in the creation of specialized compute accelerators, DPUs, and HBM stacking. The outcome of the patent infringement case brought by Singular Computing against Google clearly shows the value of algorithms and mathematical innovations. The technology developed by Singular Computing included the Low Precision, High Dynamic Range (LPHDR) approach, which trades accuracy for faster computations.
Given the high resistance of the process of neural network training to any slight deviations from correct computation results, the architecture allows hundreds of millions of computations to be performed during each cycle without hindering the work of the artificial intelligence models. In their lawsuit against Google, Singular Computing alleged that the company stole the patented LPHDR architecture to develop TPU 2 and 3 chips, which saved it billions of dollars on increasing its physical resources. Google explained its stance by pointing out the independent development of the chips and a difference between “approximate mathematics” used by the firm.
In contrast to the above, the rapid uptake of these technological advances in the field of AI hardware technologies by businesses is increasingly being viewed critically by Non-Practicing Entities (NPEs). In 2024, it was estimated that roughly 80% of all patent litigations filed against South Korean tech firms in the U.S. were brought about by NPEs having broad patents or ambiguous patents to extract large sums from their targets.
Samsung Electronics was involved in 86 patent litigation suits in the U.S. in 2024, which is much higher than 2023 when it faced just 51 lawsuits. Moreover, these numbers easily trumped the number of suits facing companies like Apple (43 suits), Amazon (46 suits), Google (39 suits), and Meta (11 suits). It is noteworthy that these suits have been primarily focused on certain special jurisdictions in the U.S., which are known as hotbeds of high-tech dispute resolution. The increasing trend towards suits has compelled business lawyers to work with heavy datasets during discovery procedures, which include revisions of source codes, algorithms, and customized chipsets.
Apart from these, judges have also started to punish the use of unverified AI-based technologies in preparing for litigation proceedings. For instance, a patent infringement case in Kansas resulted in the imposition of a penalty of $12,000 on the lawyers working for the patent holding company for including false citations to legal cases and fictitious quotes in the prepared legal documents using unverified AI model. This has made a statement to the IP attorneys of the current age that even when an AI model assists them in writing documents, they alone will be answerable for their content.
Strategic Intellectual Property Risk Management in the AI and Semiconductor Era
The rapid growth of AI, semiconductor innovations, and political polarization calls for strategic thinking by companies to ensure a successful IP strategy in the future. Being in the world of the innovation-driven economy, which implies that laws help alert individuals rather than the sleepers, meaning people who ignore their legal rights.
Companies should consider making dynamic IP risk maps in order to identify highly dense areas of patents, specialty hardware vendors, and architecture development. Companies will find it easier to avoid patents that can lead to infringement lawsuits by creating risk maps that indicate possible overlaps with their proprietary innovations. Since the semiconductor industry is very fragmented, companies should ensure that there are clear agreements regarding indemnity, litigation costs, disclosures, and third-party claims made by foundries and manufacturing organizations.
Further, the patent filing application should prove the existence of a tangible “technical delta” rather than algorithm-based patents. With stringent regulations for patent filing processes across multiple jurisdictions, such as the USPTO, EPO, and China, the focus must be on proving that there have been tangible improvements in terms of performance achieved through faster computation, improved memory usage, or functionality.
Human invention as set out in Thaler v. Vidal necessitates thorough documentation of all the work done by the human mind when inventing with the help of artificial intelligence. Prompting and modifications performed during the inventive process can help in ensuring the validity of the patent. Finally, becoming members of defensive patent pools and open-source alliances will protect companies from threats emanating from NPEs and their competitors.
Author: Anshika Kumari. In case of any queries please contact/write back to us via email to [email protected] or at IIPRD.