Physical AI Needs Memory:
The Value of ITTIA DB Platform for Intelligent Devices
Building persistent intelligence with ITTIA DB Lite and STM32N6
A robot detects an unusual movement. An agricultural device identifies possible plant stress. An industrial sensor recognizes a change in motor behavior. Each application produces an AI result. Understanding that result later requires the measurements, operating conditions, and model version surrounding it.
Physical AI needs a useful memory of its interactions with the physical world. ITTIA DB Lite and STM32N6 bring complementary capabilities to this requirement: embedded data management and on-device AI processing.
Connect AI processing with persistent history
STM32N6 provides an Arm Cortex-M55 processor running at up to 800 MHz with Helium vector processing. NPU-equipped STM32N6x7 devices also include ST’s Neural-ART accelerator for supported neural-network workloads. The application still needs to organize observations, retain selected evidence, and retrieve records after inference has finished.
ITTIA DB Lite provides embedded transactional, streaming, and time-series data management for MCU applications. ITTIA describes its use with STM32N6 for local sensor-data management and AI workflows.
These capabilities provide a foundation for connecting what a device observes with what its application decides.
Preserve useful evidence within the device budget
A production device does not need to retain every raw sample indefinitely. It can keep compact operating summaries for longer periods and preserve detailed sensor windows around selected events. Retention rules should coordinate these records so that an inference does not point to evidence that has already been deleted. This creates a deliberate memory strategy: what to preserve, how long to retain it, and what to reclaim when storage reaches its limit.
Coordinate storage with inference
Acquisition, preprocessing, inference, and persistence share system resources. Developers should define bounded queues, task priorities, and storage budgets for their combined workload. STM32N6x7 uses a flash-less configuration, making external-memory selection and storage integration important design decisions.
Validation should measure capture and query latency while inference is active. Recovery tests should also verify that committed events remain consistently associated with their sensor records after interruptions.
Turn device history into engineering value
Persistent history helps engineers investigate intermittent behavior, compare results across model versions, and select representative examples for model improvement.
Instead of reviewing an isolated anomaly score, a service team can examine the conditions surrounding the event. ITTIA Analitica can support this workflow through visualization and exploration of embedded data.
STM32N6 supplies processing resources for capable MCU-based AI. ITTIA DB Lite supplies the embedded data foundation for preserving the evidence surrounding that AI. Explore ITTIA DB Lite for STM32N6 and give your intelligent device a useful, lasting memory.
Experience ITTIA DB Lite and STM32N6 for AgTech
Contact ITTIA to request source code access to our agriculture demonstration and explore how ITTIA DB Lite and STM32N6 work together to support intelligent, data-driven agricultural applications.