How xcore.ai puts power at the edge without blowing a budget
The cloud is a wonderful thing. It’s connected us, powered us and transformed us in ways that we never could’ve dreamed over the past two decades. There is barely a single aspect of our lives — be it personal or business — that hasn’t changed under the influence of the cloud.
But as we explained in our previous blog post, the cloud is holding back the next revolution in technology — especially when it comes to the artificial intelligence of things (AIoT). At XMOS, we set out to solve the issue of cloud limitations by launching xcore.ai – a crossover processor for the AIoT — which moves processing of data and decision-making to the edge and onto devices themselves.
Traditionally, moving the processing capabilities necessary for AI onto devices themselves would’ve been difficult because engineers would’ve had to embed expensive and power-hungry high-end CPUs on endpoint devices. xcore.ai, however manages to offer high performance without a hefty price tag — and it does so by addressing three key issues:
1. Removes third-party IP
Existing CPUs rely heavily on third-party hardware and software, which comes with associated licensing costs. xcore.ai uses new architectures built entirely by XMOS, removing the need for third-party intellectual property — keeping the production cost low.
2. Takes a new approach to processing
While some AIoT applications will not require the full heft of processing power that can be delivered by a server in a data centre, the requirements of delivering even basic AI and decision-making functionality are still exceptionally high. xcore.ai takes a different approach to processing with sophisticated algorithms picking up much of the heavy lifting — allowing the hardware itself to be relatively lightweight.
xcore.ai combines the performance and functionality of an application processor, with the ease of use, low-power and real-time operation of a microcontroller. xcore.ai also has native support for 32b, 8b and 1b (binarised) neural networks. A binarised network represents network data as either +1 or -1, which offers a 10x improvement in performance and memory density at the cost of a modest reduction in accuracy.
This combination of resolutions, and the ability to mix and match in a hybrid network, enables application developers to deliver machine learning optimised for a low memory and processing footprint.
3. Works in a multitude of applications
The AIoT market is not monolithic — the standards and requirements across the hundreds of markets, and thousands of market segments vary wildly. We all know the cost of delivering a new CPU. It is simply not realistic to expect vendors to produce thousands of variations of a platform to serve all of those different needs.
Therefore, we’ve deliberately designed xcore.ai with incredible flexibility in mind, enabling programmable trade-offs between compute classes (AI, DSP, control and IO) that can be defined by the product designer rather than us as chip vendor. This powerful device enables application developers to deploy every different class of processing workload on a single multicore crossover processor. It incorporates DSP and machine learning capability together with scalar, floating and fixed point and vector instructions to deliver efficient control.
Let’s be clear — we’re entering a new era. By 2025, Gartner predicts there will be 65 billion devices generating 180 zettabytes of data and a global spend of $3 trillion on AIoT. The issue, though, is that none of this is possible without reducing the cost of the AIoT for each individual organisation that wants to take advantage of it.
xcore.ai is, therefore, the key to unlocking this AIoT market once and for all, with the potential to underpin everything from the smart home to connected healthcare, the automotive industry, industry 4.0 and smart cities. We believe that it could herald a genuine intelligence revolution.
To find out more about xcore.ai and how it could help you to embed artificial intelligence into your devices, please visit http://xcore.ai/