XMOS SDK for connected AI devices

XMOS SDK for connected AI devices

New Products |
By Ally Winning

XMOS (Bristol, UK) has launched an all-new software development kit (SDK) for its chip launched earlier this year.

The SDK includes TensorFlowLite for Microcontroller development tools and provides standardised tools and resources that they need to create devices that absorb contextual data from their environment, infer meaning from that data, and translate the results into action.

Scripts, tools and libraries to convert TensorFlowLite for Microcontroller models into a format that targets accelerated operations on the platform, with libraries to support FreeRTOS operation on, providing a familiar, standard industry programming environment to work in.

The SDK includes examples showing a variety of operations based on bare-metal and FreeRTOS operation, including smart microphone sensing, as well as guides, example builds and execution walkthroughs, and access to XMOS’ open-source libraries of interfaces & signal processing algorithms

These tools will enable developers to rapidly deploy custom or off-the-shelf AI models using a standard framework alongside all of the control, communications, signal and I/O processing required to create a complete and secure application solution.

“Our AIoT SDK enables developers to create intelligent endpoint-AI solutions for a huge variety of applications,” said Mark Lippett, CEO of XMOS. “The flexibility of the architecture enables our customers to create truly differentiated solutions using standard embedded software techniques like TensorFlowLite for MCU in a fraction of the time required using traditional hardware approaches.”

“It’s great to see XMOS’s latest contributions to low-energy embedded machine learning and I’m pleased that TensorFlow Micro has been able to integrate,” said Pete Warden, Technical Lead for the TensorFlow Mobile team at Google. “This combination will enable a lot of exciting applications in the future.”

Early access to the XMOS AIoT SDK will be available on the GitHub open source platform, designed to be used in conjunction with the Explorer Kit, which is available on limited release via

Future releases will include other hardware platforms, targeting specific use case applications. This includes a smart home platform – a small form-factor reference design with additional Wi-Fi capability, designed to demonstrate the capabilities of voice at the edge of networks and due to be released in early 2021.

Next: the microcontroller for the AIoT

The microcontroller was launched back in February based around the original XMOS architecture and is fully programmable in ‘C’, with specific features such as DSP and machine learning accessible through optimised c-libraries. It has up to 128 pins of flexible IO (programmable in software) give access to a wide variety of interfaces and peripherals, along with an integrated USB 2.0 PHY and MIPI interface for collection and processing of data from a wide range of sensors.

It uses deep neural networks using binary values for activations and weights instead of full precision values, dramatically reducing execution time, which provides 2.6x to 4x more efficiency than 8bit AI mcrocontrollers. This gives a 32x improvement in AI performance over ARM’s Cortex-M cores, with 16x faster I/O processing and 15x digital signal processing performance says the company.

Related articles

Other articles on eeNews Europe  

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles