Intel foundry takes a helping ARM

Article By : Peter Clarke

Intel Custom Foundry agrees to a full support package for ARM intellectual property on the upcoming 10nm FinFET manufacturing process.

Intel Custom Foundry has made a statement of intent that it is serious about competing with the likes of TSMC, Samsung and Globalfoundries by agreeing a full support package for ARM intellectual property on the upcoming 10nm FinFET manufacturing process.

The deal could also mark the start of the disaggregation of Intel and its eventual splitting up into separate manufacturing service and technology sales constituents, in other words the end of Intel as we have known it.

In the short term the deal is an acknowledgement that—at least in mobile—ARM's architecture rules the roost and it is what chip developers insist on for their system chips. Despite billions of dollars spent trying to get into mobile Intel Corp. has concurred and let its foundry manufacturing operation cut a deal.

The deal means that ARM's Artisan physical libraries—and as a result POP optimisations of its cores—will be available to third parties on Intel's 10nm FinFET process. Spreadtrum and LG Electronics are already customers of Intel's foundry. Intel has now accepted that although those companies may have tinkered with Atom system-chips to please Intel, their commercial architectural choice was ARM and Intel's foundry had to get with the program or risk losing customers.

The adoption of Artisan and POP is vital if Intel is to build its foundry customer base because the flexibility this engenders speeds the design of core implementations and SoCs, reduces time to tape-out and thereby reduces risk. It is part of the standard design flow set by ARM with foundries such as TSMC. The initial POP IP on Intel's 10nm FinFET process will be for two yet-to-be-announced Cortex-A processor cores designed for mobile computing applications in either big-little or stand-alone configurations.

Of course Intel's foundry has already been making ARM-based chips for the likes of Netronome and may soon do so for Altera, which of course is now a wholly-owned subsidiary of Intel's. However, these were essentially arms-length legacy engagements. What the latest agreement indicates is that Intel acknowledges it must engage with ARM if it is to be credible as a foundry manufacturer, even at the risk that it will undermine the position of its own processor architectures.

The risk

The risk to Intel is that this deal casts Intel as a whole, rather than just the foundry operation, as complementary to ARM. Marketing is a relatively blunt exercise and already it can be argued that ARM is seen as a leader in processor and physical design and Intel seen as a leader in the capital-intensive business of chip manufacturing at the leading edge.

That's kind of what Will Abbey, general manager of ARM's physical design group in San Jose, said in a blog, although he was careful to ascribe the manufacturing talent to Intel's Custom Foundry unit.

However, once these distinctions start getting made for an aggregated company there can be an almost inevitable flow of consequences. It sets a tide running.

It starts thus, with Intel having to continually strengthen the walls and distance between its foundry operation and its mainstream chip product business. It does this to provide potential foundry customers with assurance that their designs and commercial intelligence is safe with Intel Custom Foundry.

Having achieved a degree of separation the free-market arguments develop. If Intel Custom Foundry is allowed to service potential rivals to Intel and support competing architectures, then Intel's product divisions should be free to get chips manufactured where they can get the best deal, which may not be Intel's manufacturing operation. That is already happening to an extent with Intel going outside the company for some chip manufacturing.

And then come the financial arguments. Intel will find that with chip manufacturing and process development representing 90% of R&D cost but only 10% of sales revenue and the same percentages flipped round for chip design and product sales, shareholders will start to see value in splitting the business.

I am not saying that such a disaggregation of Intel is inevitable. Samsung is aggregated at a higher level in that it sells chips AND systems as well as providing chip foundry services. And it seems to be making a decent fist of all three, but it faces the same issues described above. In the other camp is TSMC whose chairman Morris Chang has always vowed TSMC will remain the specialist manufacturer and never compete with its customers.

One should not see Intel's agreement with ARM as a cause of disaggregation but yet another symptom that there is a tide running here.