AB2 AML100 development board
13/07/2023
Aspinity has created a machine learning core that utilises completely analogue circuitry to accomplish machine learning tasks. It is called analogML and, due to its analogue nature, it offers users an ultra-low-power solution for machine learning processing.
The use of analogue circuits to process sensor data allows the more power-hungry circuits, such as digital signal processors and analogue-to-digital converters, to remain asleep until relevant data is detected and a wake-up signal is received. This can reduce a system’s power consumption from milliwatts to microwatts, which in applications such as edge machine learning can have a significant impact on the device’s overall battery life.
The first chip by Aspinity that features the analogML core is the AML100. Natively processing analogue data, it will consume less than 20 µA. It can support up to four analogue sensors and is also field programmable. Furthermore, it is currently offered in a
7 mm × 7 mm 48-pin quad-flat no-leads (QFN) package or users can pick up one of three development boards that Aspinity offers. The most recent of these boards is the AB2 AML100 application board.
The AB2 AML100 application board is offered in the familiar Arduino form factor. As a result, it can support the many sensor interface and shield boards currently available on the market. Additionally, it features compatibility with Renesas’s
Quick-Connect Internet of Things (IoT) platform. This enables simple prototyping with the currently available hardware and software blocks from the Quick-Connect Internet of Things ecosystem. For example, sensor and peripheral boards are available that can be daisy-chained to one another, allowing for extendable hardware. Furthermore, full access to the code and libraries for interfacing to hardware is available through Renesas’s most common integrated development environments and software workflows.
Chris Allexandre, Senior Vice President at Renesas, explained: “System designers are continuously searching for faster ways to develop new and innovative Internet of Things products. Our collaboration with Aspinity allows designers to add new features to their next-generation power-constrained products without compromising the performance of
microcontroller units (MCUs). We are excited to continue to work with Aspinity on providing ultra-low-power solutions for the myriad of always-on applications that require both high performance and an extended battery life.”
Various demonstrations have been provided by Aspinity showcasing the potential applications for the AML100. These include voice detection, vibration monitoring, acoustic event detection and glass breaking. The glass breaking demonstration included the AB2 AML100 application board interfacing with the Quick-Connect Internet of Things platform’s EK-RA6M3 evaluation kit. The system was designed to detect glass breaking utilising sensors. The entire system consumed less than 45 µA of current when the AML100 was used for sensor listening and processing while the Renesas microcontroller unit was in deep sleep mode. Overall, the microcontroller unit stays in deep sleep mode for 99.9% of the time while the AML100 monitors sensors, allowing significant battery life savings to be achieved.
