AISIN and Toyota develop integrated memory chip for AI efficiency

Japanese automotive component manufacturer AISIN, a major shareholder of Toyota, has unveiled a system-on-chip designed to address the energy and speed bottlenecks of physical AI in vehicles. The chip utilizes built-in STT-MRAM to store both operating systems and AI model weights, reducing the need to access external memory.

Developed in partnership with Tohoku University, the 16nm FinFET process chip features an Arm Cortex A53 dual-core CPU and an Arm MaliG31 GPU. By integrating memory, the system can achieve kernel startup in 65.7 milliseconds, which is reportedly 38 times faster than systems relying on external memory. The researchers also implemented a hybrid multiplication architecture that reduced AI model weight data by 53 percent while maintaining 16-bit precision.

This development follows industry concerns regarding the high power consumption of AI. SK hynix executive Choi Jun-yong noted that the AI bottleneck is shifting from computing performance to memory and electricity usage, as significant energy is wasted shuffling data between processors and external memory.