The first families in the i.MX9 series will be built in 16/12nm FinFET process technology with an ARM Cortex-M33 microcontroller core for the real time performance. NXP would not comment on which applications processors would be used, but the current i.MX8 family ranges from two A72 and four A53 processors, down to for A35 cores.
The family will have i.MX9 scalable AI across the range, and NXP was the lead developer with ARM for the U65 micro neural processing unit (microNPU) for edge AI.
NXP says specific i.MX 9 applications processors will support multi-object recognition and multi-face recognition in millisecond inference time, voice-based systems that can recognize natural language and accents, sequence analysis for gesture recognition, anomaly detection for industrial predictive maintenance and synthetic sensors in smart home as well as industrial and IoT applications.
NXP says the blocks will be managed by the eIQ development software that takes the AI code and automatically allocates it to the approriate block or mix of blocks.
This is aimed at smart home, smart city and public safety systems, fleet management, precision farming and agriculture, consumer audio, healthcare and energy applications, building on the i.MX8 family. The edge AI capability opens up Industry 4.0 applications in industrial automation and the real-time domain is also well-suited to industrial applications such as the CAN network requiring fast boot under 100ms.
The i.MX9 architecture will use the new EnergyFlex architecture that allows different blocks in the chip to be powered up and down independently to save power.
The family will also use the EdgeLock secure enclave introduced with the i.MX8ULP that provides self-managed, autonomous on-die security subsystem. For security that spans from the edge to the cloud, specific members of the i.MX 9 series will be certified for use on Microsoft’s Azure Sphere cloud service. This provides security management for the chip, which is separate from the data and application cloud service.