NTT Research and MIT Explore AI Hardware Efficiency

By Kazuhiro Gomi, President and CEO, NTT Research, Inc. 

The Massachusetts Institute of Technology (MIT), one of the world’s leading academic institutions, has launched a timely initiative on artificial intelligence (AI) – the MIT AI Hardware Program. NTT Research joined the program in March 2022 as a member of its inaugural class of industry partners. We’re very excited about this opportunity to collaborate.

There is well-deserved interest in AI, which is revolutionizing how businesses and organizations are run. At the R&D end of the spectrum, scientists and researchers are seeking higher computational performance for AI. Yet they are also concerned about energy consumption and are looking for greater efficiency. The MIT AI Hardware Program exists at the crossroads of these two goals, with an emphasis on energy efficiency.

As such, the program aligns with basic research being conducted by some members of our Physics & Informatics (PHI) Lab. For instance, groundbreaking work by PHI Lab Senior Scientists Logan Wright and Tatsuhiro Onodera and colleagues at Cornell University, recently discussed in Nature, involves a new computational model for the machine learning (ML) subcategory of AI. Not constrained by existing energy requirements, this new method, called Physical Neural Network (PNN), configures ML without relying on digital computations. Another PHI Lab Senior Scientist, Ryan Hamerly, is currently engaged in joint research at MIT on interferometers, passive devices (i.e., those requiring no energy) which can perform certain types of mathematics called matrix vector multiplications, commonly used in today’s AI engines and demanding high digital computational power, thus energy.

MIT has an outstanding reputation in AI research. Its Computer Science and Artificial Intelligence Laboratory (CSAIL), for instance, counts more than 900 researchers, including nine who have won the A.M. Turing Award. The new MIT AI Hardware Program links MIT’s School of Engineering and its Schwarzman College of Computing with interested parties in government and industry. The program is taking a holistic approach, covering AI from materials to architectures to algorithms and beyond. Some of the technologies it plans to prioritize are new complementary metal-oxide-semiconductor (CMOS) designs, monolithic 3D systems and analog non-volatile memory devices. It will also include areas of particular interest today, such as “edge” applications and integrated intelligent sensors.

Others interested in those areas, in particular power consumption and smart sensors, include participants in the Innovative Optical and Wireless Network (IOWN) initiative, which was launched by NTT, Intel and Sony and now includes more than 20 members. All members of the MIT AI Hardware Program, including us here at NTT Research, bring to bear internal momentum and unique perspectives. We look forward to seeing how this powerful collaboration plays out over the next several years.