Special Contribution Invention of High Electron Mobility Transistor (HEMT) and Contributions to Information and Communications Field Takashi Mimura Honorary Fellow Fujitsu Laboratories Ltd. 1. Introduction participated in research on the GaAs metal oxide semi- More than 30 years have passed since Fujitsu’s conductor field effect transistor (MOSFET) for about two announcement of the high electron mobility tran- years. The MOSFET is widely known as an indispens- sistor (HEMT) in 1980.1) Since then, the HEMT has able device for LSIs, and the purpose of my research achieved widespread use as a fundamental technol- was to explore the possibility of achieving LSIs with ogy driving innovation in the field of information GaAs MOSFETs. Specifically, my aim was to eliminate and communications. Applications include satellite the interface state near GaAs and the gate oxide film to broadcasting receivers, mobile phone systems, milli- achieve an electron accumulation layer. Unfortunately, meter-band automobile radar, GPS navigation systems, despite a number of attempts with a variety of tech- and broadband wireless access systems. Furthermore, niques, I could not decrease the interface state density with a goal to achieve even higher speeds in informa- to a level that would cause accumulation to occur. tion and communication technologies in the future, Consequently, in February 1979, concerned about HEMT R&D is becoming extremely active throughout continuing GaAs MOSFET research, I developed an in- the world. Today, in addition to conventional gal- terest in a “modulation-doped superlattice structure,” lium arsenide (GaAs) HEMTs, the development of in which donor-doped aluminum gallium arsenide ultra-high-frequency indium phosphide (InP) HEMTs2),3) (AlGaAs) layers and non-doped GaAs layers are alter- and low-power/high-efficiency gallium nitride (GaN) nately stacked. In this structure, electrons accumulate HEMTs4) is progressing. in the non-doped GaAs layer between n-type AlGaAs In this paper, I would like to present a retrospec- layers on both sides. This phenomenon would make tive on HEMT R&D based on my personal experiences sense for researchers in this field, however for me, this from its invention to commercialization. I will explain was a surprise. This was because it was clear there was some background on its invention and describe factors an accumulation of electrons here, which is something that contributed to its successful implementation. that I could not achieve in GaAs MOSFETs. I was quite impressed, however, since this was a phenomenon in a 2. Idea of HEMT modulation-doped superlattice structure that I was not During the initial development of the HEMT in especially familiar with, no ideas came to mind at that 1979, I belonged to a research group developing GaAs time. metal semiconductor field effect transistor (MESFET) I conceived the idea of the HEMT in July 1979. My devices. The GaAs MESFET invented by C. A. Mead in aim here was to establish a heterojunction structure 1966 is a high-speed device featuring extremely high between n-type AlGaAs and GaAs layers while intro- cost performance.5) For me, who was pursuing high- ducing a Schottky barrier junction at the surface of the speed devices, I wondered whether improving the n-type AlGaAs layer to create a depletion layer. This GaAs MESFET was the only work left. However, I was scheme eliminated electrons within the n-type AlGaAs not interested in any follow-up research themes, so I layer and enabled a field effect to govern a 2D electron FUJITSU SCIENTIFIC & TECHNICAL JOURNAL, Vol. 54, No. 5, pp. 3–8 (October 2018) 3 Cutting-Edge R&D T. Mimura: Invention of High Electron Mobility Transistor (HEMT) and Contributions to Information and Communications Field gas within the GaAs layer. Merging existing device concepts embodied by GaAs Figure 1 shows the energy band diagram ex- MESFET and MOSFET formed a new type of junction, pressing the HEMT operating principle as it appeared resulting in the creation of a new device concept called in the patent specification. If we treat the depleted the HEMT (Figure 2). n-type AlGaAs layer as a gate-insulating film in the manner of silicon dioxide (SiO2), the device concept of 3. Demonstration of HEMT operation the HEMT can be understood to be structurally similar An idea by itself is no more than a fantasy—it is to the MOSFET. At the same time, the Schottky barrier meaningless if it cannot be shown to be feasible by junction used to deplete the n-type AlGaAs layer has experiment. Such experimental verification requires the function of the gate electrode in GaAs MESFETs. a variety of techniques. In setting out to construct an HEMT prototype, an advanced crystal growth technique was indispensable. At that time, however, I was work- ing in a department developing devices, and there were no techniques available for fabricating crystal at the level of precision required for HEMTs. The 2D electron gas serving as the HEMT current channel lies near the GaAs and AlGaAs heterojunction. It is a small region having the breadth of an electron wave function, or in terms of atomic layers, an ex- tremely narrow region of a dozen of atomic layers. Consequently, for crystal growth, there was a need for a technique having precise control to create a hetero- junction on the order of atomic layers. At that time, however, molecular beam epitaxy (MBE) was the only Sorce Jorl o the Ititte o Ie Iortio method that could satisfy such a demanding level of Teleiio ieer, Vol. 52, No. 4 (18) precision. Figure 1 MBE has two key features: a crystal growth rate HEMT energy band diagram. that, at approximately one atomic layer per second, is SFT Sorce te ri Ioie oors lectro eletio reio T Sorce te ri elete te OSFT te lyer reio te ere oie Sorce ri eteroctio 2 electro ihrit lyer ihrit lyer Sorce II Trctio Vol. J100, No. 10 (201) Figure 2 HEMT creation from merging existing device concepts. 4 FUJITSU Sci. Tech. J., Vol. 54, No. 5 (October 2018) Cutting-Edge R&D T. Mimura: Invention of High Electron Mobility Transistor (HEMT) and Contributions to Information and Communications Field more than ten times slower than that of conventional at an unbelievably high level of precision. We therefore crystal growth methods; and the use of mechanical prepared a variety of etching solutions and tried them shutters that enables successive growth of different out with a series of simple experiments, but reproduc- types of semiconductors. ibility of etching precision was still a problem. I therefore sought out the cooperation of an We decided not to miss the most honorable mo- MBE group outside my department and established ment of announcing the first HEMT device by taking too an HEMT prototype “circle,” consisting of only of two much time to improve the device fabrication process. researchers from the MBE group and myself, the former We therefore decided to put aside this problem of re- of whom participated out of personal interest. We were producibility for the time being and start working on a able to initiate research activities through this format prototype. We were intent on realizing a functioning of an informal prototype circle, which I believe was one HEMT before our competitor, and to this end, even a important factor in our successful development of the single working prototype would suffice. Towards the HEMT. If this work had been taken up as an official re- end of December 1979, after two or three failures, we search theme at the idea stage, I think it would have succeeded in finding a chip operating as an HEMT in been exposed to various types of interference while a low-yield wafer. This took place about four months losing flexibility in development activities, all of which after forming our informal HEMT prototype circle. could have increased the risk of failure. The results of this HEMT research were presented Immediately after starting up this prototype circle, at the 38th Device Research Conference (DRC) held in I received a letter. To my surprise, it was from Dr. R. June 1980. After giving my presentation, I was handed Dingle of Bell Laboratories, the author of papers on a document by a researcher unknown to me. To my modulation-doped superlattices, who expressed a de- surprise, it was a copy of a paper on a device similar to sire to talk with me about the GaAs MOSFETs that I had our HEMT (called an “inverted HEMT” in which the HEMT been researching. He visited Fujitsu Laboratories on AlGaAs and GaAs layers were upside down). Listening August 30, 1979. We discussed his modulation-doped to what he had to say later, I found out that he was superlattice and our MOSFETs, but our work on HEMT a researcher at the Thompson-CSF company in France development that we had just started was naturally and that, much to his disappointment, they had been confidential. To my relief, I learned through these beaten by our presentation. At any rate, we were the discussions that they had not yet reached the point of first to be recognized for the invention of the HEMT, having a clear device concept of the HEMT. but only by a hair. This type of occurrence happens However, in early September soon after his frequently in competitive technology development. It visit, I learned from the program of that year’s GaAs is essential that the existence of a strong competitor be IC Symposium that his group was announcing ex- considered and never ignored. perimental results on the control of electrons in a modulation-doped superlattice. Clearly, they were mov- 4. Achieving practical HEMTs ing in the direction of research pointing to some type The HEMT began to show signs as a potential of device.