Quark, Intel's new Pentium-based architecture, is aimed head on at ARM in applications such as the Internet of the Things (IoT), and there's been a lot of interest in quite what it means.
The fact that it uses ARM's AMBA bus interconnect underscores the vital importance of the ecosystem. It runs alongside a legacy serial bus to blocks such as the GPIO and real-time clock, separate from PCI Express and other serial interfaces.
The first instantiation of the core is used in the X1000 SoC. The specification, which was released at the end of last week, raises more questions than it answers for the SoC business, especially when you bring the IoT into the equation.
The way Intel has addressed the software ecosystem, with ports of Linux and VxWorks from Intel-owned Wind River and security from Intel-owned McAfee, highlights part of the challenge. Some SoC designers will welcome a ready-made software ecosystem, but this is primarily for Intel's customers buying the chip, rather than the core. Intel has said it will be a good long while before the IP is available on TSMC's technology.
Even then, there are key questions: How does this work as a multi-core device, both in homogeneous and heterogeneous systems? Exactly how the interfaces to graphics and security co-processors that need to be tightly coupled will work is not clear.
Having a synthesizable core helps with this. But creating an effective, multi-core IP solution for third-party SoC designers could take a signiifcant amount of work, and both ARM and Imagination Technologies are well ahead. Of course, a multi-core-enabled device (perhaps the X2000), could be in the roadmap.
Another issue is where this new architecture will actually compete. SoCs based on ARM's M0+ Flycatcher core will not run Linux, although they do hit the sub-50-cent price point for the IoT, including security engines and targeted peripherals.
With cache, wait states, legacy bus, and a larger area, Quark is unlikely to compete on area, price, and power. And with such price pressure, coupled with the memory and power issues, these are not going to be on the leading-edge 20nm and 14nm processes.
Atom is firmly aimed at the IoT gateway devices, and if Quark cannot get down to the silicon dust price point, it's not going to make a significant dent in the IoT market.
It seems Intel has a few large customers, including itself, lined up for Quark for wearable devices. But as exciting as it is to have a new architecture in the embedded SoC market, the opportunities for the wider market appear to be quite limited.
A smartwatch running Linux (which should really mean WindRiver Android) is interesting, but are we likely to take the hit of Android for a sensor controller in the IoT? That's unlikely, since the space between the ultra-low-cost sensor/controller and gateway is not really clear, while a heterogeneous multicore version will play well in low-cost smartphones and smart devices, alongside Intel's wireless IP.
This first part is an exploratory device with lots of options. The dedicated, optimized SoCs will come when Intel actually gets to focus on its end applications.
By Nick Flaherty www.flaherty.co.uk
No comments:
Post a Comment