On November 9, Mitsubishi Electric announced that it had demonstrated a major reduction in "frequency dispersion," in which current response changes with frequency, in a high-frequency transistor built by forming a GaN crystal layer on a gallium nitride (GaN) substrate. In GaN transistors on silicon carbide (SiC) substrates, long the mainstream approach, crystal defects cause frequency dispersion, which has required correction by digital circuits. The company counts three achievements as world firsts: clarifying the effect of crystal defects, greatly suppressing frequency dispersion, and demonstrating the usefulness of GaN substrates for high-frequency devices. However, this claim rests on the company's survey of 156 papers, and the published materials contain almost no concrete figures showing the size of the suppression. Prior research confirms that GaN substrates improve crystal quality, but how much of an advance this result represents will have to wait for more detailed data.
No figures for the "major suppression"; the "world first" claim rests on a survey of 156 papers
Mitsubishi Electric developed a high-frequency transistor with a GaN epitaxial layer (a single-crystal layer grown on a single-crystal substrate) formed on a GaN substrate. With conventional SiC substrates, the substrate and the epitaxial layer are made of different materials, so defects inevitably arise during crystal growth. These defects impede the movement of electrons during high-frequency operation and cause frequency dispersion, in which current varies with operating frequency. According to Mitsubishi Electric, compensating for this has required complex digital correction circuits, leading to larger systems and greater power loss.
In this study, the company developed a new circuit analysis method that incorporates the electrical response of the material to crystal defects (defect response), and used device characterization to clarify how defects affect frequency dispersion. It then confirmed that transistors using GaN substrates greatly suppress frequency dispersion and yield characteristics that are flat across frequency. The company says this allows the digital correction circuits to be simplified. The results are scheduled to be presented at the international conference IWN2026 (International Workshop on Nitride Semiconductors 2026), to be held November 8–13 at Kumamoto Castle Hall. The research was carried out as part of a project commissioned by Japan's Ministry of the Environment, in which the company has participated since 2024, and it aims for real-world deployment around 2030.
The announcement, however, lacks quantitative data. Neither the press release nor the three-page PDF gives a clear figure for how much frequency dispersion fell relative to conventional devices. What can be found are descriptions such as "major suppression" and "flat frequency characteristics," along with a graph comparing transistors on SiC and GaN substrates. The graph plots the real part of admittance y21 on the vertical axis against period (1 nanosecond to 0.1 second) on the horizontal axis. The SiC-substrate device, shown in blue, rises from about 1.00 to about 1.06, while the GaN-substrate device, shown in red, stays within about 1.00 to 1.02. These are rough values read by eye from the graph, and detailed measurement conditions and device dimensions have not been disclosed.
Mitsubishi Electric also describes three points as world firsts: (1) clarifying how crystal defects affect frequency dispersion, (2) demonstrating major suppression of frequency dispersion, and (3) showing the usefulness of GaN substrates, an area of strength for Japanese companies, for high-frequency devices. The basis for this, however, is a literature survey conducted with Google Scholar as of October 26, 2026. The company says it searched three keywords, "GaN substrate," "GaN wafer," and "GaN on GaN HEMT," and reviewed the 156 papers that matched. This survey alone cannot establish whether prior studies existed outside the search scope. In evaluating the "world first" claim, it is therefore necessary to distinguish its subject from the scope of the survey.
Earlier research had already confirmed the advantages of GaN substrates
The idea that GaN substrates reduce crystal defects and improve device characteristics is not new to this announcement. In a contribution to Compound Semiconductor dated February 1, 2022, a Fujitsu research team reported its development of high-frequency transistors formed on GaN substrates (GaN-on-GaN HEMTs). It explained that the dislocation density in a GaN epitaxial layer on a SiC substrate was 2×10⁸/cm², compared with less than 10⁶/cm² on a GaN substrate. In other words, using a GaN substrate reduced dislocation density to less than 1/200. Because only an upper limit is given for the GaN substrate, the actual improvement may be even larger.
In Fujitsu's report, this difference in crystal quality was also reflected in electrical characteristics. Transistors on GaN substrates showed less "current collapse" than those on SiC substrates. Current collapse is a phenomenon in which current temporarily drops because of electron-trapping levels (traps) in places such as crystal defects, and it is generally evaluated using methods such as pulsed current-voltage measurements. Although the measurement method and metric differ from the frequency dispersion Mitsubishi Electric addressed, both are phenomena in which electrons captured by defects affect the device's current response, so they are physically related.
That said, fewer crystal defects do not immediately raise high-frequency output. In Fujitsu's research, too, the high-frequency output of GaN-on-GaN initially failed to exceed that of conventional GaN-on-SiC. Investigation found that a conductive layer caused by silicon contamination had formed at the interface between the GaN substrate and the epitaxial layer. Fujitsu then introduced a chemical treatment of the substrate surface and improved characteristics by keeping the silicon concentration below the iron concentration. An amplifier with further improvements, including interface refinements and through-via formation, achieved a maximum output of 64.3 W at 2.6 GHz and a peak power-added efficiency of 71.1% (in pulsed operation).
Fujitsu, too, had been advancing its research with Ministry of the Environment support, beginning with the development of amplifiers for microwave heating. Whether this is the same project as the one under which Mitsubishi Electric announced its results cannot be confirmed from public information.
Comparing the two companies' results clarifies how much of what Mitsubishi Electric calls a world first had already been confirmed by earlier work. Fujitsu reported in 2022, using measurements of dislocation density and current collapse, that GaN substrates reduce crystal defects and that device electrical characteristics improve accordingly. By contrast, there is no evaluation item in Fujitsu's report that directly corresponds to linking the effect of defects to frequency dispersion through a new circuit analysis method, or to how much that dispersion could be suppressed. The technical novelty and the size of the improvement in the new work therefore cannot be judged by comparison with prior research alone.
Judging its true value requires quantitative data, and GaN substrate supply remains a challenge
It is important to note that current collapse and frequency dispersion are different metrics. What Fujitsu showed was the drop in current in pulsed measurements, while what Mitsubishi Electric showed was the effect of reducing changes in device characteristics across frequency. Even if the two phenomena are related, demonstrating improvement in one does not negate the novelty of results on the other. Differences in what is being measured may be behind Mitsubishi Electric's claim that major suppression of frequency dispersion is a world first. Still, judging its novelty and practical value requires concrete measurement results.
Reading the published comparison graph by eye, the SiC-substrate device's measured value rises by about 0.06, while the GaN-substrate device varies within about 0.02. However, because the measurement conditions and the performance required in circuit design are unknown, it cannot be determined whether this difference is an improvement large enough to eliminate digital correction circuits or only enough to reduce the amount of correction. At this stage, all that can be said is that the published comparison suggests GaN substrates reduce frequency dispersion.
The research presentation at IWN2026 could significantly change this assessment. If it presents quantitative data comparing against SiC-substrate devices under matched operating frequencies and measurement conditions, and makes clear the basis for simplifying correction circuits, the results could be evaluated more concretely. In that case, the key question would be not only the new circuit analysis method but how much GaN substrates themselves contribute to high-frequency circuit design. Conversely, if no specific figures are released, the practical meaning of "major suppression" will remain unclear.
There are also challenges beyond frequency dispersion suppression to resolve before GaN substrates can be put to practical use. One is heat dissipation. Mitsubishi Electric describes SiC as a substrate with high thermal conductivity that excels at heat management, and Fujitsu's contribution also points out that GaN's thermal conductivity is lower than SiC's. Fujitsu estimated through simulation that channel temperature can differ by up to 100°C between continuous-wave and pulsed operation. However, because continuous-wave operation was actually achieved, it judged the heat dissipation of its device to be sufficient.
Another challenge is stable supply and larger diameters of GaN substrates. According to the Nikkan Kogyo Shimbun on September 24, Mitsubishi Chemical is considering mass production of GaN substrates at a pilot line inside the Muroran Plant of Japan Steel Works, and customer evaluation of 4-inch products is under way. It also plans to begin shipping 6-inch samples as early as the fiscal year ending March 2027, and is working on developing 8-inch products. However, that report covers substrates for power semiconductors, and there is no information that they will be used in the high-frequency transistor Mitsubishi Electric has now developed. Mitsubishi Electric itself targets real-world deployment around 2030, but has not disclosed its GaN substrate suppliers or mass-production setup.
The significance of this research lies in linking the high crystal quality of GaN substrates to improvement in frequency dispersion suppression, a characteristic important to high-frequency circuits. However, how much circuit simplification or power-consumption reduction this would actually bring has not yet been sufficiently explained. If quantitative results, including measurement conditions, are released at IWN2026, it will become possible to judge more objectively the technical advantage Mitsubishi Electric claims and the prospects for practical use.