CAS — Connect and SenseEdge AI

    Your AI.
    On the device.

    CAS Edge AI Module brings the Nordic nRF54LM20B with its integrated Axon NPU to your product. Process sensor data locally with hardware for AI inference, plus cellular connectivity and battery management in 36 × 19 mm.

    Discuss your AI applicationExplore the specifications

    Drag to rotate · PCB design visualization

    3D based on the PCB design; some component bodies are simplified. Assembly may vary by configuration.

    Axon NPU
    integrated AI accelerator
    nRF54LM20B
    128 MHz Cortex-M33
    2 MB / 512 KB
    NVM / RAM
    36 × 19 mm
    module footprint
    Top side of the actual CAS Module with nRF9151, components and u.FL connectors, held between two fingers
    Photos of the actual hardware, revision 0.0.0.

    From sensor data to local decisions.

    The nRF54LM20B is at the heart of CAS Edge AI Module. Its integrated Axon NPU accelerates suitable AI models on the device itself. A hardware foundation for pattern recognition, classification and anomaly detection in sensor data — built around your application.

    The AI hardware is on board. Models, sensors and firmware are developed and validated for each application; no generic AI model comes preinstalled.

    01

    Measure at the source

    Connect the sensors your application needs through a custom carrier. Collect the data we use to train and test the model.

    02

    Process locally

    Run a suitable AI model on the LM20B. Local inference does not have to wait for a cloud connection.

    03

    Send what matters

    Send a status, classification or event to your platform. You decide which data leaves the device.

    Edge AI at the core. Connectivity around it.

    The LM20B handles local processing. The nRF9151 connects results to the outside world, while the nPM1300 manages energy. Together they provide a foundation for a connected edge-AI product.

    nRF54LM20B

    The edge-AI core

    128 MHz Axon NPU for suitable AI inference, alongside a 128 MHz Cortex-M33 and RISC-V coprocessor. 2 MB NVM and 512 KB RAM for your embedded application.

    Component information

    nRF9151

    Connects to the outside

    LTE-M, NB-IoT and GPS/QZSS. IoT NTN and DECT NR+ are additional chip capabilities, with separate firmware and network requirements.

    Component information

    nPM1300

    Manages the energy

    Charging, power path, 1.8 V buck and battery measurements. Firmware adapts charging settings and energy modes to the selected cell.

    Component information
    LIS2DW12 accelerometerRV-8803-C7 RTC4 MB NOR flashLTE / GNSS u.FL2.4 GHz antenna

    Every interface. Every building block.

    Specifications from the hardware design and component documentation. Revision 0.0.0 is a development platform; chip capabilities are not automatically validated product features.

    Preliminary specifications · hardware rev. 0.0.0

    Pinout: 63 edge contacts

    Processing & memory
    Application processor
    Nordic nRF54LM20B · 128 MHz Arm Cortex-M33 · FPU/DSP and TrustZone.
    Coprocessor & AI
    128 MHz RISC-V coprocessor and Axon NPU. Application software and models are developed for each product.
    Internal memory
    nRF54: 2 MB NVM and 512 KB RAM. nRF9151: 1 MB flash and 256 KB RAM.
    External memory
    Winbond W25Q32JW · 32 Mbit / 4 MB NOR flash at 1.8 V. SPI or software QSPI through the RISC-V coprocessor.
    Firmware
    nRF Connect SDK / Zephyr. Custom firmware on both SoCs; nRF9151 can also run Nordic Serial Modem.
    Radio & positioning
    Cellular
    Nordic nRF9151 · LTE-M and NB-IoT. IoT NTN and DECT NR+ are chip capabilities with separate modem firmware, network and integration requirements.
    LTE-M bands (hardware)
    B1, 2, 3, 4, 5, 8, 12, 13, 18, 19, 20, 25, 26, 28, 66, 85, 106.
    NB-IoT bands (hardware)
    B1, 2, 3, 4, 5, 8, 12, 13, 17, 19, 20, 25, 26, 28, 65, 66, 85, 106. NTN: B23, 249, 252, 255, 256.
    Transmit power (chip)
    LTE: class 3, 23 dBm / class 5, 20 dBm. 2.4 GHz: up to +8 dBm. Not measured module values.
    Short range
    Bluetooth LE, including Bluetooth 6.0 Channel Sounding, IEEE 802.15.4 and 2.4 GHz proprietary. Thread, Zigbee and Matter require suitable firmware and product validation.
    GNSS
    GPS L1 C/A and QZSS L1 C/A through nRF9151. SKY65943-11 GNSS front end with LNA and filters.
    Antennas
    Two u.FL connectors for LTE and GNSS. On-board TDK 2.4 GHz chip antenna. Alternative RF routing to edge pads on request.
    SIM
    FSA2567 dual-SIM switch. External 1.8 V SIM via edge pads; MFF2 eSIM footprint available, not populated by default.
    Power & energy management
    Power management
    Nordic nPM1300 · charging, power path, battery measurements for fuel-gauge software, ship and hibernate modes.
    USB supply (VBUS)
    4.0–5.5 V. This is not a 22 V module input: the nRF54 USB circuit limits the absolute maximum to 6.0 V.
    Battery (VBAT)
    2.3–4.45 V at the PMIC. Cellular transmission requires at least 3.0 V at the modem, including voltage drop; ≥3.2 V recommended for LTE.
    Charging
    Configurable 32–800 mA (PMIC range). Cell type, thermistor, temperature and thermal design determine the setting.
    Outputs
    VSYS, +1V8 and SIM_1V8. The 1.8 V buck has a total 200 mA budget, shared with the module itself.
    Low power
    Ship, hibernate, PSM and eDRX capabilities. Total module consumption and battery life have not yet been characterised.
    Sensors & interfaces
    Accelerometer
    ST LIS2DW12 · 3 axes · ±2 / 4 / 8 / 16 g · motion-detection interrupt.
    Real-time clock
    Micro Crystal RV-8803-C7 · temperature compensated · ±3 ppm per the component specification.
    Digital & analog I/O
    23 GPIO at 1.8 V, including 5 analog inputs, 7 clock-capable pins and 2 NFC pins. Not 3.3 V tolerant.
    Serial interfaces
    SPI, UART and I²C through appropriate GPIO and peripheral mapping. The internal I²C bus is not exposed at the edge.
    USB & NFC
    USB 2.0 high-speed D+/D−, USB-C CC1/CC2 and NFC-A tag on GPIO pins. Connector, antenna and ESD belong on the carrier.
    Debug & controls
    Separate SWD/reset for both SoCs. BUTTON, BAT_NTC and three 5 mA LED sink outputs. Requires a 1.8 V debug probe.
    Mechanical & integration
    Dimensions
    36.0 × 19.0 mm. 1.0 mm PCB per fabrication order; approximately 3.0 mm total height, excluding connected antenna cables.
    Mounting
    63 castellated edge pads on a 1.27 mm pitch. Components on the top side only; test pads on the bottom.
    PCB construction
    8-layer HDI, ENIG, black solder mask and white silkscreen.
    Temperature range
    −40 to +85 °C based on the components. Module validation remains required; charging range follows the selected battery.
    Standard & options
    Standard: LTE/GNSS u.FL and 2.4 GHz chip antenna. eSIM, RF pad configuration and RGB LED are build options to agree.
    Design status
    Hardware rev. 0.0.0 · preliminary specifications. No module certification; antennas, carrier and end product require verification and approval.

    Module power consumption, RF performance and temperature range still require characterisation. CE/RED and other approvals must be assessed for the final product.

    Design these in

    • All logic operates at 1.8 V. Use level shifters for 3.3 V peripherals.
    • Keep the antenna corner clear of copper and metal on the carrier.
    • The GNSS connection is for a passive antenna; an active antenna needs external bias.
    • Place ESD protection at external interfaces and respect the voltage and current budgets.
    CAS — Connect and Sense

    One family. More ways to sense.

    CAS means Connect and Sense: our development platform for connected products. From solar-powered tracking and vibration sensing to hardware for edge AI. We adapt the solution to your application.

    CAS Solar

    Solar-powered asset tracking.

    CAS Vibe

    Turn asset vibration into machine activity data for your platform.

    CAS Edge AI Module

    nRF54LM20B with Axon NPU. The hardware foundation for local AI in your product.

    You’re viewing this product

    What should your device recognize?

    A pattern, machine activity or an anomaly in sensor data. Tell us what you want to recognize, which measurements are available and where the device will operate. Together we scope the model, hardware and validation.

    Discuss your AI application