News

UK-Based XMOS Introduces AI Processor Designed Specifically for AIoT Applications

March 01, 2020 by Luke James

XMOS, a company that made headlines for developing the highest performing voice interface for under a dollar, unveils xcore.ai – the fast, flexible, and economical processor designed specifically for the AIoT market.

XMOS, a Bristol, England-based company, which received $15 million in funding in 2017, has spent three years developing its low-cost and efficient AI chip, dubbed xcore.ai – a processor for artificial intelligence processing in IoT products which combines this with digital signal processing (DSP), control, and I/O.

In a press release on February 12, XMOS unveiled xcore.ai – which it called a “disruptive crossover processor” – for the first time in a press release and an accompanying white paper. 

 

The Third Generation xcore.ai Architecture

The first-generation architecture was deployed into “hundreds” of applications bridging different I/O protocols and the second-generation boosted control and digital signal processing performance via a dual-issue pipeline. 

Now in its third generation, which was originally conceived to offer control processing that would allow engineers to design differentiated products, the latest xcore.ai is a crossover chip designed to deliver high-performance AI, DSP, control, and I/O in a single device with prices from just $1.

It is also capable of providing real-time inferencing and decisioning at the edge, in addition to signal control, processing, and communications which are usually handled by microcontrollers. 

“Traditionally this type of capability would be deployed either through a powerful applications processor or a microcontroller with additional components to accelerate key capabilities,” said an XMOS spokesperson. “However, the xcore.ai crossover processor is architected to deliver real-time inferencing and decisioning at the edge, as well as signal processing, control and communications.”

 

Xcore.ai graphic listing product features and capabilities.

A product graphic for the xcore.ai listing the major features and capabilities of the chip. Image used courtesy of XMOS

 

Other features of the third generation xcore.ai include:

  • 16 real-time logical cores, with support for scalar/float/vector instructions.

  • IO ports with nanosecond latency for real-time response.

  • Support for binarised 8-bit, 16-bit, and 32-bit neural network inference.

  • Multi-modal data capture and processing.

  • On-device inference of TensorFlow Lite.

  • Instruction set for DSP, machine learning, and cryptographic functions.

XMOS is also claiming an inference performance improvement when compared with an Arm Cortex-M7 in STM32M7; xcore.ai, running on 1xcore logical core at 160MHz offers an execution time figure of 5.236ms whereas the STM32M7 offers 35.676ms with the whole chip running at 600MHz. 

Inside the chip, you will find 1MB of RAM with up to 400Gb/s of bandwidth, 16 logical cores, and up to 128 pins of software-programmable I/O with low-latency interconnects. There is also an integrated USB 2.0 PHY and MIPI interface for data collection and processing. The RAM is split into two 512KB modules and each processor has a dual-issue execution unit that can execute instructions at twice the clock frequency.

 

Potential Applications

Product demonstrations are scheduled from June, however, XMOS provided us with an interesting use case in its press release.

“Imagine the humble smoke detector,” says the press release. “With xcore.ai embedded, a smoke detector could use radar and imaging to identify whether there are people in the affected building and, if so, determine how many and where they are located. Using voice interfaces, the detector could communicate with those inside, while vital sign detection could identify whether they are breathing. Put together, this builds an intelligent picture of the environment that can be fed straight to the emergency services, enabling an informed rescue operation, improving accuracy and speed of response.”

xcore.ai’s development has been part-funded by the European Union Horizon 2020 research and innovation scheme.