Intel’s new Quark CPU core is one step towards a new, ARM-like foundry model

Intel’s new Quark CPU core is one step towards a new, ARM-like foundry model

A little over five years ago, Intel took the wraps off a new low-power CPU core, codenamed Silverthorne. That chip, branded as Atom, became the centerpiece of Intel’s low-power initiative, even if there’ve been a few missteps along the way. Now, Intel is aiming for an even lower power market with a new family of products — except this time, the company is bending previously ironclad rules about how it builds and licenses its own hardware, becoming a bit like its mobile archnemesis, ARM.

What’s smaller than an Atom? Meet Quark.

Wrong Quark.

That’s better.

Technical details on the Intel Quark family are still limited. We know the chip is x86-compatible and built on a 32nm process. Intel claims the Quark SoC will be 1/5 the size of the Atom SoC, and draw a tenth the power. Those are going to be challenging goals to hit and they suggest Quark is a truly embedded part aimed at markets Intel hasn’t previously deigned to enter. This isn’t a competitor to the Cortex-A family — it aims to compete with ARM’s Cortex-M series.

What makes Quark unique from a design level is that this is the first Intel chip that’s fully synthesizable and designed to integrate with third-party IP blocks. That implies that a customer can use Quark and hook it to custom I/O, graphics, storage, or WiFi/3G radios of their own choosing. For now, Intel intends to retain manufacturing control over the entire process, but Quark is apparently designed as a CPU design that other foundries could license and build long-term.

A new foundry strategy

There’s been talk for years that Intel might become a TSMC or GlobalFoundries competitor by throwing open its doors to all and sundry. This has always been unlikely given that Intel’s ace in the hole has always been its own foundry technology — a technology that it typically reserves for its own products. What the company is doing with Quark is leveraging its own IP in a way that lets it offer customizable hardware to potential customers without giving up control of either its processor IP or its own manufacturing capability.

While Intel has said that other foundries would eventually be able to build Quark chips in theory, this should be understood as a hypothetical “nothing standing in the way,” rather than a long-term licensing intention. Unlike Atom, which involves a great deal of hand-tuning and customized silicon, Quark is designed to be simpler and easier to produce. That means there are fewer roadblocks between the chip and long-term production if another foundry should seek a license. Any license Intel granted would undoubtedly be customer-specific and linked to a particular version of the core — there’s no sign that Intel is going to license the wider x86 model anytime soon — and to be fair, there’s no sign anyone else wants an x86 license.

Remember the early MIDs of 2007/2008, such as this Gigabyte M528? Probably not.

Will Quark be adopted? That’s hard to say. It’s easy to forget that Intel did a lot of heavy lifting around Atom, MID (Mobile Internet Device), and netbooks. While MIDs never took off, they definitely got the ball rolling as far as commissioning more mobile products and tablet-like designs. The company didn’t just put Atom out there as a design and invite folks to build something — it took an active part in creating the products. There’s no ready-made market for Quark, though it’s possible that Intel will integrate Quark cores as coprocessors in future designs.

What is clear is that Intel is going after a huge existing market in low-cost synthesizeable cores with entrenched competitors and customers that already enjoy a great deal of choice. MIPS and ARM are players in this space, alongside embedded manufacturers like Freescale.