Shrinking transistors and components alone is no longer sufficient to advance semiconductor performance. As demand for computing capabilities associated with artificial intelligence grows, the materials used in manufacturing have become a factor determining whether some new structures are feasible at all, rather than merely an input selected after the chip design has been finalized.
This is the main conclusion of an analysis published by Semiconductor Engineering on September 17, 2026, written by Hideaki Okamoto, director of next-generation business development at Mitsubishi Chemical Group. As a piece published in the Sponsor Blog section, the article also reflects an industry perspective connected to the materials sector. Its conclusions should therefore be read as an analysis of industry trends, not as an announcement of a specific product.
From Repeated Experimentation to Computational Research
Traditionally, materials development relied on producing a sample, testing its properties and analyzing the results, then repeating the cycle. Producing a single candidate can take weeks when the material is complex, making the exploration of a large number of alternatives slow and costly.
Increasing computing power is changing the order of this process. Simulation can precede actual manufacturing, helping researchers identify candidates that merit laboratory testing. As models, proprietary databases, and computing resources improve, it becomes possible to examine a broader range of options and identify performance spaces that traditional attempts might not reveal.
The value of this approach is not limited to finding an improved version of an existing material. It may reveal early on that incremental improvement will not achieve the required objective, allowing an unproductive path to be abandoned and resources to be redirected toward a fundamentally different solution. The article describes a gradual transition from using computation as an auxiliary tool to screening before experimentation, then large-scale screening, and ultimately the use of artificial intelligence in materials development.
The Problem Is Not the Material Alone
Advanced manufacturing-node materials must meet multiple, sometimes conflicting, conditions, including electrical performance, thermal behavior, mechanical properties, purity, compatibility with manufacturing processes, and the ability to be produced in large quantities. For this reason, the analysis argues that trial and error alone will not be sufficient as the space of possibilities expands.
However, computation does not solve the problem by itself. The material must work within a specific deposition process, interact correctly with adjacent layers, and contribute to achieving the properties of the final device. Accordingly, the article says semiconductor development is moving from a linear relationship between supplier and customer toward deeper collaboration among device manufacturers, equipment companies, and materials suppliers.
Early collaboration gives the parties access to practical knowledge that does not usually appear in formal specifications, such as process interactions, trade-offs, and constraints discovered by engineers during joint work. However, this model also requires allocating greater resources earlier, rather than gradually increasing investment as technical risks decline.
Where Is the Need for New Materials Emerging?
The analysis presents several examples of fields in which new system requirements may drive materials transitions. In structures with fine pitches, continued device scaling may require the development of new precursor materials for thin-film deposition, combining molecular design, computational screening, and process expertise. The shift from wet to dry processes is highlighted as an example of this overlap.
In artificial-intelligence systems, the movement of data itself is becoming a constraint on performance and energy consumption. As electrical signal speeds increase, the need for optical solutions, including photonic circuits and photonic interconnects, is growing. The article points to opportunities for transparent optical-waveguide materials and organic electro-optic materials capable of converting signals at high speeds, while balancing optical performance, manufacturability, and reliability.
In power electronics, data centers’ energy consumption is increasing the importance of more efficient power conversion. Gallium nitride (GaN) offers potential benefits including lower on-resistance, operation at high voltage, and lower power loss compared with traditional silicon-based solutions. However, realizing these advantages on a large scale requires high-quality, cost-effective GaN substrates, along with scalable processes.
What Is Changing in Practice?
The most important message for the technical reader is that materials are increasingly entering the stage at which the system architecture is defined, rather than serving merely as a later element in the manufacturing chain. This increases the importance of accumulated experimental data, because computational models and artificial intelligence cannot easily compensate for a lack of reliable data. It also makes collaboration among value-chain participants part of technical and strategic capability, not merely a supply arrangement.
Nevertheless, the source does not provide quantitative results or examples of materials that have actually reached commercial production using these methodologies. Nor does it specify timelines, costs, or regulatory obstacles for each path. Therefore, the certain conclusion remains that the industry is moving toward integrating simulation, data, and early collaboration, while the speed at which these ideas move from the laboratory to large-scale manufacturing remains an open question.
certi.news reading: What is actually changing is not the replacement of the laboratory by the computer, but the redistribution of work between them. Simulation may reduce the number of unpromising experiments, but manufacturability and compatibility among the material, equipment, and adjacent layers will still require practical validation. If the article’s thesis is correct, companies that possess high-quality materials data and participate early in chip-development road maps will be better positioned to identify constraints before they become obstacles in design or production.