– Originally written by Mark Lippett, CEO of XMOS –
Have you ever tried to hit a housefly with a rolled-up newspaper, only to find the fly long gone by the time your energetic swat reaches its original target? If you have, then you have some insight into the dilemma faced by traditional system-on-chip (SoC) product planners when they consider which products to build for the intelligent IoT.
For the first time in its history, semiconductor market growth is being driven by a collection of markets – collectively branded the IoT – with a diverse set of requirements, rather than a single homogeneous mega-market like PCs, or mobile phones. Whilst the size of the IoT opportunity is huge (even by mobile phone standards) each individual application is relatively small – like our metaphorical fly.
Even if an individual application represents a big enough market to justify the tens of millions of dollars to build a dedicated system on chip, it is unlikely that the market will be there when the device – our metaphorical rolled up newspaper – hits the market on its original trajectory, two to three years later.
The traditional SoC product planner is faced with the prospect of committing to a solution that will arrive too late at a market that was barely big enough in the first place. It is no wonder the semiconductor industry is beset with product planning paralysis and costly project restarts.
From ready-aim-fire to ready-fire-aim
Now imagine that we could continuously track the movements of our sprightly fly whilst our rolled-up newspaper was in flight. If product planners could adapt the behaviour of silicon at the speed with which the market is evolving, a hit is guaranteed.
Since the costs of developing silicon are so enormous, they drive a conservative approach to design which takes years, not months. We might imagine a future world in which new fabrication technologies make silicon design so quick and easy that we can economically address fragmented and evolving markets with custom silicon.
However, there are no signs of the seismic change required to make this happen – indeed IBS have reported that the costs of developing new semiconductor products continue to double on each advance in technology node. We must look elsewhere for flexibility – above the silicon process to the architecture.
For many years, some flexibility has been delivered by including software programmable processors within a System-on-chip (SoC). These processors are typically surrounded by a moat of fixed-function hardware that interface the device to the outside world.
Today, as they attempt to track market needs, these SoC architectures have evolved to include more than one processor (often of more than one type) alongside an array of hardware blocks to accommodate various combinations of interfaces to the outside world.
These heterogeneous SoCs are partially ‘soft’ (they go some way to delivering flexibility) but are rarely scalable – they remain limited by the performance of the embedded processors that are trapped behind a fixed selection of application-specific interfaces.
But what if those interfaces don’t fit the bill, or more processing is required? It appears that these architectural approaches are also of limited use in increasingly fragmented and volatile markets – our fly lives to fight another day, unlike our erstwhile product planner.
Making silicon soft
If traditional SoC architectural techniques do not offer a solution, what can we do? How can we make an affordable silicon platform where all the silicon is flexible – not just an isolated island in the middle? We need a platform where new solutions can be developed in weeks, not years, with the flexibility to deliver for multiple markets at once. Flymageddon.
In the 1990s a new class of semiconductors evolved, called Field Programmable Gate Arrays. They were an immediate hit – offering hardware engineers the ability to build chips whose behaviour was defined by a bit-stream that was loaded when the device was switched on.
In the following 30 years, the flexibility of FPGA drove higher and higher levels of integration, emerging most recently as heavy-iron AI processors in the data centre. However, whilst they demonstrated that flexibility can be achieved by architecture above the silicon process, FPGA was inaccessible to the embedded software community, and too expensive and power hungry for most IoT applications.
The answer to accessibility lies in software programmability. However, software cannot create flexibility where hardware does not support it – a new type of platform is required which has flexibility all the way to the very edges of the silicon. That software programmability must deliver on the four key classes of processing required by the IoT – namely control processing, DSP, AI and IO in a manner that is both flexible and scalable. This new kind of platform, which can be programmed using techniques that every embedded software engineer designing products for the IoT uses every day, can change the game.
Flexibility from edge to edge
To adapt a quote from Darwin – in this fast-moving world, it’s not the largest semiconductor companies that will survive and thrive, but those with technology that is most adaptable to change. Vendors must be able to adapt quickly to the emerging needs of the IoT across a multitude of markets, and affordably deliver the required performance no matter what the application. To be able to achieve this, flexibility is key at every stage of the device lifecycle.
To address this need, XMOS has pioneered a new intelligent IoT platform with flexibility and scalability in its DNA – xcore.ai is the only chip in the world that specifically meets the needs of the burgeoning intelligent IoT market, by combining control, DSP, AI and IO processing with built-in scalability to address a range of performance requirements. xcore.ai’s fast, flexible processing and neural network capabilities make the tiny chip capable of processing data and taking actions on-device within nanoseconds.
xcore.ai’s flexibility and scalability – accessible through industry standard embedded processing techniques – represents a breakthrough not only by opening up the possibility of new and exciting intelligent IoT applications, but also in realising the full potential of smart homes, smart healthcare, smart cities, industry 4.0, autonomous vehicles and other major technology markets.