Tohoku University and Aisin unveil world's first 16nm AI chip with on-chip STT-MRAM, booting nearly 39 times faster

Tohoku University and automotive parts maker Aisin announced on November 2 that they have jointly developed a 16nm AI system-on-chip with built-in STT-MRAM, which they say is the first of its kind. STT-MRAM is a type of non-volatile memory that keeps data when the power is off; the team stored the operating system and the AI model's weights directly in the memory on the chip, removing the time needed to read them from external DRAM and a solid-state drive.

Boot time and energy figures

According to Tohoku University, the chip is made on a 16nm FinFET process and combines 32MB of STT-MRAM with an AI accelerator, two Arm Cortex-A53 CPU cores and an Arm Mali-G31 GPU. The time from power-on to the operating system kernel starting is 65.7 milliseconds, against 2,535.3 milliseconds for a conventional design, about 38.59 times faster. Running Tiny YOLOv3 object detection, it uses 50.7 millijoules per frame, about one-fiftieth of the energy of a conventional design.

The work comes from a team led by Professor Tetsuo Endoh at the university's Center for Innovative Integrated Electronic Systems, together with Aisin engineers. The results were presented at the IEEE Asian Solid-State Circuits Conference (A-SSCC) 2026, held in Hsinchu, Taiwan, from November 1 to 4. Tohoku University calls the design near-memory computing: the accelerator reads weights straight from the MRAM next to it, without going through external memory.

Applications and a caveat

The partners gave no price, production timeline or customer, saying only that they aim to extend the design to cars, robots, industrial equipment and sensing devices. Semiconductor writer Dan Nystedt noted on social media that this is a real-silicon demonstration but "not revolutionary": Hailo, Ambarella, Renesas and others already use large on-chip SRAM, embedded MRAM or in-package DRAM in similar ways. What matters next is whether Aisin can put the design into production parts and how it performs with larger models.

Photo: file illustration of AI chips, not a photo of the chip developed in this project.