Running with the fox while hunting with the hounds

Everyone loves to tell people what they should do and when someone’s in trouble, the urge to give advice becomes acute.

Intel’s in trouble and it’s getting no shortage of advice on how to get out of it. The latest comes from IC Insights.

First, the diagnosis:

‘In 2013, Intel registered its third consecutive year of poor sales growth. In fact, the company logged a 2% decline in IC sales, eight points worse than the 6% growth rate displayed by the worldwide IC market. Moreover, its outlook for 2014 was for more of the same—relatively flat revenue growth in an IC market that IC Insights expects will increase 7%.’

Then the prescription:

Intel should open its shuttered Fab42 in Arizona to grab Apple’s estimated $3.4 billion foundry spend with Samsung, says IC Insights.

‘If Apple did agree to have Intel make its application processors at the 10nm node in Fab 42, it would gain access to the most leading edge processing technology in the world,’ says IC insights, ‘IC Insights believes that Intel is approximately one year ahead of both Samsung and TSMC in IC process technology – an advantage that Apple could parlay into a significant performance advantage in both the tablet PC and smartphone markets.’

Snag with prescription:

Although IC Insights may be correct in the one year advantage for Intel’s leading edge process, this begs the question: is Intel’s leading edge process optimised for mobile?

I think everyone knows it’s not – it’s optimised for performance.

Intel has always made its own mobile processors going back to X-Scale days and before, on processes behind the leading edge.

If Intel optimised its leading edge process for mobile, and made Apple’s processors on that leading edge process, it would make x86-based processors made on old processes look an even worse bet for the mobile market.

And if Intel put x86 mobile processors on leading edge processors it would destroy a business model set up to make $100+ die.

While if Intel ever put its server processors on less than its latest process it would lose that business to ARM-based processors.

Will Intel ever optimise it’s leading process for anything other than performance? I don’t think so.

It’s always been an IC industry shibboleth that you shouldn’t compete with your customers.
Can Intel run with the mobile x86 fox and hunt with the Apple hounds?



  1. Not that quickly Keiith, I was thinking of 10nm or more likely 8nm as a merging point.

    And remember the recent gains TSMC have announced in their 16FF process are down to their realisation of what Intel have done for their metal layers.

    But at the moment I accept the transistors are very different. I think whenever we go to III-V will be when the options close down a lot.

  2. “Is it just possible that we are all heading to a common single process ?”

    I don’t think so. TSMC 20nm (which is supposed to be the same backend as 16FF) and Intel 14nm are like chalk and cheese. I obviously can’t talk about details but there are fundamental differences.

  3. A reasonable summary David, but remember the speed of Intel processors peaked quite a few years ago when the die started overheating and so power consumption began to dominate over performance. Meanwhile TSMC are also shrinking towards a single process at ’16nm’ (i.e. about 19nm) as well..

    Is it just possible that we are all heading to a common single process ?

Leave a Reply

Your email address will not be published. Required fields are marked *