Seventeen years after launching its first STM32 microcontroller, STMicroelectronics has advanced more than 100 iterations to now include a proprietary neural processing unit to deliver what ST claims is 600 times more machine-learning performance than the last generation.
The new NPU, called the Neural-ART Accelerator, has been in use in the new STM32N6 MCU by some select customers for more than a year and is now ready to ship in high volume, the company said in a briefing with reporters.
Along with ST’s AI software, the N6 “marks the beginning of a long journey of AI hardware-accelerated STM32,” said Remi ElOuazzane, president of MCUs and more at STMicroelectronics.
Putting AI capabilities at the IoT edge will combine energy efficiency from microcontrollers with AI analytics for such applications as computer vision and sensor-driven insights to lower the total cost of ownership of modern equipment, said Tom Hackenberg, principal analyst in memory and computing at Yole Group.
The Neural ART Accelerator has 300 configurable multiply-accumulate units that can reach 600 giga operations per second, which is 600 times the machine learning performance of a typical high-end STM32 MCU today. It runs an 800 MHz Arm Cortex-M55 core. It includes 4.2MB of RAM and an Image Signal Processor for the first time on an MCU. ST will also sell the MCU in a version without the Neural-ART accelerator.
In a short video, ST shared remarks by Zhou Xing, founding partner of Meta-bounds based in China. He showed lightweight augmented reality glasses incorporating the N6 and said the glasses would weigh just 35 grams, partly due to N6 form factor which also allows an enhanced user experience without adding a greater drain on battery life.
ST also is collaborating with Qualcomm for edge AI-powered industrial applications. Self-contained modules will include Wi-Fi6, Bluetooth 5.3 and Thread connections with the STM32 MCUs. Available in first quarter of 2025, the modules will be powered by Qualcomm’s QCC743 SoC.