Toyota shows off SoC that stores model weights and its OS on integrated memory

What happened

Toyota, a major shareholder in automotive components maker AISIN, and AISIN have shown off an AI processor SoC aimed at automotive and 'physical AI' applications. AISIN contends existing chips rely on external memory to store AI models and generate heavy traffic between processor and model, making image recognition and other on-device AI less effective, and that DRAM is currently vastly expensive. The SoC is built on a 16nm FinFET process with an Arm Cortex A53 dual-core CPU and an Arm MaliG31 GPU, and includes built-in STT-MRAM that stores both model weights and an operating system. AISIN and its research partners at Tohoku University say the chip goes from power-on to kernel startup in 65.7 milliseconds – 38 times faster than possible with external memory – and can run the Tiny YOLOv3 object detection algorithm on-chip. Using a hybrid multiplication architecture combining 8-bit and 16-bit multiplication, the AI model weight data was reduced by 53 percent, from 17.7MB to 9.36MB, while maintaining 16-bit precision. AISIN is an established parts supplier already selling to Toyota and other automakers, but has not said if or when this chip can enter production.

Impact and signal

Affected components and suppliers

Source and time

Source: The Register · 11/2/2026 · View original source