BrainBoard1500 puts AKD1500 AI on embedded boards

BrainChip and Neuromorphyx have launched BrainBoard1500, a compact developer and evaluation board built around BrainChip’s AKD1500 neuromorphic neural-network accelerator. Designed and manufactured by the Croatian company Neuromorphyx, the board puts the accelerator into an Arduino Nicla-compatible form factor intended to connect directly to familiar embedded hosts.

BrainBoard1500 measures 22.9mm by 22.9mm and provides SPI/QSPI host access, with I2C used for system control. Onboard model flash is 4MB as standard with an 8MB factory-configured option, while a two-channel current monitor allows developers to profile inference power and examine sleep and wake behaviour while models run on the device.

The hardware is supported by an open-source Arduino Nicla library and drivers. BrainChip and Neuromorphyx also describe a route to adapt the interface for Raspberry Pi, Seeed Studio/XIAO, OpenMV, Espressif, DFRobot, Adafruit, SparkFun, STMicroelectronics, and custom MCU or FPGA hosts, widening the number of existing sensor and controller platforms that can be used during evaluation.

That interface flexibility matters because a neural accelerator is only one part of an embedded inference chain. Sensor acquisition, preprocessing, model storage, host control, power management, and application software all influence latency and energy use, and a device that performs well on a larger development card may behave differently once placed beside the MCU, FPGA, or sensor stack intended for production.

The AKD1500 uses BrainChip’s event-based Akida architecture, which is designed to exploit sparse neural activity rather than processing every input as a dense stream. Neuromorphyx is already integrating the processor into its wider NeuroReflex hardware stack, combining event-based sensors, FPGA preprocessing, and local inference for robotics, defence, automotive, space, and industrial applications.

BrainBoard1500 separates the accelerator from that larger system so engineers can evaluate it under their own power, size, latency, and interface constraints. The direct SPI/QSPI path is intended for running spiking neural-network models on sensor data, while the onboard current monitor gives developers a way to compare workloads rather than relying solely on headline efficiency claims measured under a different system configuration.

Power profiling is especially useful for neuromorphic hardware because efficiency depends heavily on the model and input activity. Sparse event streams can reduce processing and data movement, but the result changes with sensor behaviour, network architecture, host activity, memory traffic, and the proportion of time the accelerator can remain in a low-power state. Measuring those factors on a small embedded board makes the evaluation closer to the eventual application.

The new board also broadens the hardware routes available around AKD1500. BrainChip already offers the processor in an M.2 format for systems that can accept a standard host expansion module; BrainBoard1500 targets smaller development environments where engineers may want to place the accelerator next to a sensor or microcontroller and establish whether the workload justifies a dedicated co-processor.

Evaluation hardware still leaves substantial production work unresolved. A final design must address EMC, power integrity, thermal behaviour, connector choice, environmental qualification, software maintenance, security, and the long-term availability of the accelerator and surrounding components. The Arduino-oriented path lowers the barrier to first use, but industrial adoption will depend on whether developers can reproduce measured latency and power benefits inside their own board and firmware constraints.

The onboard flash options are modest by accelerator-board standards, but that is consistent with a platform intended for compact models and local inference rather than large generative workloads. It also keeps the evaluation focused on the device’s intended edge role.

BrainBoard1500 will initially be sold through Neuromorphyx with its software library, drivers, datasheet, and supporting documentation. BrainChip is supplying the AKD1500 devices and technical enablement, while the companies plan broader availability through distribution, education, and approved ecosystem channels. For neuromorphic computing, compact evaluation hardware is not proof of a production market, but it is one of the practical steps required to move specialist accelerator silicon into ordinary embedded design workflows.