Creating a new programming language for cloud environments

“I am designing the next-generation computers that will be needed in the age of AI. As you know, computers consist of hardware such as CPUs and memory made up of semiconductor elements (transistors), along with software such as operating systems, applications, and programming languages. My aim is to create high-performance, ultra-efficient computers by optimally connecting the interfaces between these two. In other words, I am conducting research that spans both hardware and software,” says Associate Professor Yukinori Sato of Toyohashi University of Technology. True to his words, Associate Professor Sato handles the entire process, from building the cloud computing infrastructure to creating a fully operational prototype. The scale of his research ranges from palm-sized computers to data centers with tens of thousands of servers and even the Fugaku supercomputer.

The key focus of Associate Professor Sato’s research is the design of compilers.

“A compiler is software that translates programs written by humans into a language that computers can understand. Human programmers write code using languages that are easy to use—such as Python or C++—but a wide variety of computing systems are currently in use, ranging from smartphones and large-scale computers to supercomputers, IoT devices, and edge devices; consequently, depending on the specific application, a multitude of programming languages are employed within cloud environments where these systems are seamlessly integrated. For instance, more than ten different languages are used to set up the environments that power familiar travel and e-commerce sites.

Furthermore, alongside advancements in AI, hardware is evolving. The variety of computing hardware is expanding, from standard CPUs to GPUs capable of massive parallel processing, AI accelerators (specialized processors designed for high-speed, energy-efficient AI computation), and field-programmable gate arrays, which enable the virtual customization of circuits—and each type requires different programs to be written. Therefore, we developed a domain-specific language (DSL) that is capable of automatically converting a single program to suit various environments.”

A DSL is a programming language designed for a specific purpose. Unlike general-purpose programming languages, it is designed to facilitate the simple and efficient description of processing in specific fields. In particular, the DSL being developed by Associate Professor Sato focuses on the convenient use of computers operating in the cloud. It is a remarkable tool that allows a programmer to write code in a single language, and with the help of a compiler, automatically convert that code into a program optimized for the specific computer hardware running in the cloud environment.

“In other words, one simply writes the code in a single language, and the compiler automatically translates it into a format that the specific hardware can understand and execute. By using this DSL, programmers can focus on developing the algorithms, for example, how to do the computations. Because of the existence of a multitude of programming languages, such a DSL is necessary. However, the task of creating it is incredibly complex. It requires technological innovation that spans diverse fields, and therefore hardly anyone has attempted it until now. In short, it is an uncharted area of research,” says Associate Professor Sato.

Furthermore, Associate Professor Sato and his team have developed a proprietary, domain-specific communication protocol. They have successfully demonstrated its operation in the cloud, such as at the national-level ETNET workshop in March 2026, and are currently advancing its implementation.

The Turning Point with MIT that accelerated Associate Professor Sato's research

The development of Associate Professor Sato's DSL research is based on collaborative work with a research group led by Professor Saman Amarasinghe of the Massachusetts Institute of Technology (MIT), who has been at the forefront of theoretical and practical DSL research for over 20 years.

“About 10 years ago, I suddenly realized that if I followed the direction my research was going, it consistently led me to the work of the MIT compiler group. I thought that if I could discuss things with them, my own research would take a big leap forward. So, I contacted them personally, and we have been deepening our relationship ever since. Then, in 2024, we were selected for the Japan Science and Technology Agency (JST)-funded ASPIRE (Adopting Sustainable Partnerships for Innovative Research Ecosystem) for Rising Scientists Program in semiconductors, which enabled us to create a system that allows us to conduct collaborative research at the laboratory level, and that brings us to where we are today. Naturally, I have traveled to MIT along with my students, and researchers from MIT have also come to visit us, so we are actively engaging in exchanges. I would like to further accelerate our research for the remainder of this research program, which lasts until 2028.”

International competition is particularly fierce in fields related to AI and cloud technology, yet the joint research between Associate Professor Sato and MIT is said to be proceeding very successfully. Associate Professor Sato explains that this success stems from the MIT researchers' proactive stance on making their compiler infrastructure open source, as well as the ability to divide roles effectively by leveraging each party's respective areas of expertise.
“MIT handles the research at the lower layers—such as developing foundational technologies and establishing theories for DSLs and compilers—while we conduct applied research to implement those theories in AI, cloud, and network systems. Rather than competing, we function as a team working together to create something better.”

MIT researchers involved in this joint project particularly value the high level of implementation capability demonstrated by Toyohashi University of Technology. “They have expressed their delight at seeing that we have further developed and advanced the theories and prototypes they created,” says Associate Professor Sato with pride. It is clear that this collaboration holds great significance, not only for the production of research papers and other tangible outcomes, but also for the development of talent. Looking ahead, Associate Professor Sato expressed his enthusiasm for building on this partnership to publish results that will serve as key developments in the field.

Foundational technologies for future computers

Associate Professor Sato is also focusing on developing computers for the next generation. One example is his research into ultra-efficient computers that can be used in space.

“Eighty years have passed since the invention of the electronic computer, and it has now become a versatile tool capable of doing almost anything, depending on the idea, although the challenge of increasing speed remains. However, while increasing the number of semiconductor transistors used for computation does boost processing speed, the approach of relying on sheer transistor volume to drive performance—as seen in today's data centers—consumes an enormous amount of power. That's why we are trying to improve performance and efficiency through the power of software, such as using compilers to minimize memory access and automating the generation of efficient code by eliminating unnecessary calculations. Even with the current supercomputers, it is frequently observed that only about 10% of peak performance is  achieved. This is because of unavoidable time loss waiting for main memory access and communication synchronization. To solve this, we are working on how to extract maximum performance with the minimum number of necessary processing operations.”

Thus, Associate Professor Sato and his colleagues, in collaboration with JAXA and the University of Tokyo, have succeeded in developing a dedicated computer that can process—in real time and using less than 20 watts—the algorithms necessary for monitoring the conditions required for spacecraft and space vehicles to operate autonomously.

"Further refinement of this technology will be necessary before it can actually be adopted in spacecraft," says Associate Professor Sato, but there is no doubt that such efficiency-enhancing technologies will become critical technologies not only in outer space but also in a carbon-neutral society.

Furthermore, the team are working on developing computers that mimic the mechanisms of the human brain, which can perform advanced information processing with only a little learning process.

“Despite consuming only about 20 watts of power and occupying a volume of roughly one liter, the human brain performs incredibly sophisticated information processing. It possesses processing capabilities that are vastly superior to even the latest AI-oriented computers in a variety of ways. While much about the human brain remains a mystery and the path to a breakthrough is not yet in sight, we have not given up. We are conducting our research with the ultimate goal of one day realizing a computer that matches the human brain in both performance and efficiency,” Associate Professor Sato concluded.

Reporter's Note

Associate Professor Sato's research began with a desire to build his own computer himself. This mindset permeates his current research approach of “building something that actually works first.” However, he also notes that his joint research with MIT has led to a renewed appreciation for the importance of theory. Although Associate Professor Sato spoke in a gentle manner, the interview revealed a real sense of his high aspirations and enthusiasm—culminating in his declaration that he wants to create the “ultimate computer.”