Toyota shows off SoC that stores model weights and its OS on integrated memory
AISIN developed a system-on-chip using a 16nm FinFET process with dual-core Arm Cortex A53, Mali G31 GPU, and integrated STT-MRAM for storing AI model weights and OS. The design addresses limitations of traditional chips that rely on external memory, reducing latency and cost. The chip boots from kernel startup in 65.7 milliseconds—38 times faster than external-memory designs—and runs the Tiny YOLOv3 object detection algorithm on-chip.
Key points
- AISIN AI SoC stores model weights and OS in integrated STT-MRAM
- 16nm design boots to kernel in 65.7ms, 38x faster than external-memory chips
- Targets image recognition and on-device AI for automotive and robotics
- Hybrid 8-bit/16-bit multiplication architecture optimizes limited memory
Summary and key points written by AI from the headline and a short excerpt, so they can be incomplete or wrong. The Register did the reporting; read the original for the full story.
Read the full story at The Register
