120 MHz Cortex-M4F with 2.5 MB flash for edge-AI inference
The MAX78002GXE+ is an ARM Cortex-M4F microcontroller running at 120 MHz, with a hardware floating-point unit and DSP extensions that execute single-cycle multiply-accumulate operations — the core math primitive for convolutional neural network inference on sensor data at the edge. On-chip memory comprises 2.5 MB of flash and 384 KB of SRAM, enough to hold a quantised CNN model and its activation buffers without external memory, keeping the BOM compact for battery-powered vision or audio wake-word nodes. Operating from -40°C to 105°C, the part suits outdoor enclosures, engine-bay controllers, or industrial sensor heads that see thermal cycling without active cooling.
Peripheral set and package for vision-node integration
Connectivity includes a CSI camera interface, USB, multiple SPI/I²C/UART ports, and an MMC/SD/SDIO controller — the CSI port directly captures image sensor frames into the inference pipeline without an external FPGA. The 144-CSBGA package measures 12x12 mm with a 0.50 mm ball pitch. The brown-out detect and POR blocks handle the power-up sequencing without an external supervisor.
Analog Devices lists the MAX78002GXE+ as an active product with no announced last-time-buy or discontinuation notice. Current-date-code inventory is available through independent distribution channels. RoHS3 compliance is confirmed per the manufacturer's declaration.