The artificial intelligence of things (AIoT) is reaching a tipping point. Devices that combine the cognitive capabilities of artificial intelligence (AI) with the interconnectivity of the internet of things (IoT) are beginning to see real deployment cases, with some approaching mass market.
The momentum in this field has been driven by convenience. We’re increasingly accustomed to voice-based, touch-free interfaces that offer ease and speed of use, from the smart speaker to voice assistants on our phones.
A global pandemic that highlights the risks associated with physical contact has accelerated demand, despite its challenges to supply lines. We’re suddenly viewing touchless experiences in terms of our own security — from light switches to shared appliances, voice control sidesteps the biological risk of physical contact.
If we want to take advantage of this change of pace and develop a more sophisticated AIoT, we need to make it as easy as possible for creators to develop and hone devices, both in existing fields and for entirely new applications.
There are two challenges here, and they go hand-in-hand — cost and versatility.
Cost and consequence
The processors capable of acting as an AIoT-literate brain are generally very expensive. This either drives up the market price of devices or forces manufacturers to sell them at a loss — Amazon have admitted that their Echo device has been sold at cost or for a loss, but their overall revenue profile enables them to sacrifice profit to drive adoption; smaller manufacturers really suffer when revenue is put under pressure.
Much of the expense stems from processor design. The electronics marketplace has traditionally been dominated by a select number of verticals — digital cameras, mobile phones, PCs — that eat up the vast majority of semiconductor units, with manufacturers committing their resources accordingly.
This is increasingly no longer the case. With so many unique devices to cater for, and the shrinkage of traditional giants like digital cameras, manufacturers have compensated for market fragmentation with increased prices.
As a result, many processors simply don’t offer both versatility and cost-effective performance, which makes it difficult for engineers to accommodate creative license into an engineering bill of materials (EBOM). In an ideal world, the AIoT market needs a processor that is versatile enough for manufacturers to explore and maximise the capabilities of their own IP, without the restrictive cost.
The price of intelligence
The original xcore® processor was designed to deliver flexibility on three points: at a hardware behaviour level, at a control processing level, and in terms of digital processing. 2020 has seen the birth of the third-generation xcore processor, xcore.ai, which introduces a fourth element in AI processing.
This is academically interesting until you realise it’s in a device priced from just $1— it’s actually commercially viable. Such a low cost means that xcore.ai can succeed beyond current device classes, integrating into consumer and commercial electronics that are already in homes all over the world.
Price has been driven down in many different ways. xcore.ai dispenses with third-party IP to reduce licensing costs and integrates as many open source tools as possible. FreeRTOS and TensorFlow Lite are well-established and popular examples. Given the versatility of the processor, it’s also possible to order in larger batches, reducing EBOM cost.
As xcore.ai is competing for the same dollars in the EBOM as crossover processors, this price makes it a hugely compelling option. To take a comparable ‘field leader’ in the ARM Cortex M7, xcore.ai offers offer a 32x improvement in AI performance, 16x faster I/O processing, 15x digital signature processing performance and 21x 16-bit MACs.
In short, not only does xcore.ai outperform the devices that it competes for space with, but it does so at a price that makes it viable for commercial use. In a recent report, The Linley Group described xcore.ai as “much less expensive than competing chips… among the AI-chip vendors offering similar performance.”
The voice of reason
Convenience and COVID-19 — an unexpected duo — are driving us further and further towards voice-based, touch-free interfaces that deliver real utility without compromising on security. The AIoT devices that incorporate this technology, from the smart home to the smart city, can have a revolutionary impact on our day-to-day lives.
xcore.ai delivers the unique blend of flexibility needed to catalyse growth in the AIoT. It delivers a price/performance balance that offers the opportunity to push intelligence into machines that haven’t been able to realistically afford or access such levels of convenience and utility.
If we take full advantage of this, we can give devices a voice in a way that changes the way we interact with the world around us, making everyday tasks more fluid than ever before.
To find out more about the technical specifics of how XMOS is driving the AIoT industry, our CEO Mark Lippett recently presented our offering to Fierce AI. Watch his session in full here.