Intel Corp (INTC)

[wbmw]

The power reduction in the 32 nm Atoms is so substantial that
it is almost certain these are built in the SoC/low power variant
of the 32 nm process. This has traditionally lagged Intel's high
performance mainstream process of the same feature size by
at least a year.

I don't think the limitations were all in the back end. Moving from Menlow to Moorestown to Medfield for Intel involved a lot of "firsts".

Moorestown involved both a new Lincroft CPU as well as a new PCH, both of which were power optimized, compared to the 3-chip Menlow implementation. It was the first to implement a PMIC power management architecture, as well as the S0ix power states, which allowed 50mW operating points - and led to the 50x reduction in standby power compared to Menlow. Intel also built in video decode/encode to handle 1080p content, as well as a number of other features.

Medfield took this a big step forward, and the improvements are numerous, so I thought I'd put it in list format:

- First product on 32nm-LP process
- First fully integrated SOC, which required all new physical layer optimizations on I/O
- Brand new, low power SOC I/O (eMMC, MIPI-DSI, MIPI-CSI, etc.)
- Package-on-Package (PoP) LP-DDR2 memory support
- Si Hive ISP successfully ported and optimized on an Intel process
- New audio/video codecs supported
- New security engine integrated
- New Imagination GPU integrated
- All new certification for 3G wireless standards, gov't regulations, etc.

To think that Intel could have done all of the Moorestown optimizations, and all of the Medfield optimizations in one fell swoop, and be on time with the 32nm ramp, is a huge stretch of the imagination.

Of course, the legitimate argument for Medfield was in partnering and putting all the eggs in the Nokia basket, which of course backfired and delayed the introduction of Medfield to the market by 6-9 months. That was the disappointing part, since Medfield would have been very competitive in mid-2011 when it was planned, and a mid-2012 refresh with CT+ would been best-in-class this holiday season, given the current phone chips in the market.

But it's kind of silly to rehash that mistake, as if Intel management hasn't learned and changed their tactics. In fact, it really pushed Intel to embrace Android and get a number of design wins with more diverse companies.

Medfield managed to get a paltry 0.2% of the phone SOC market in Q2'12 - before any major designs were shipping. I think it will be interesting to see how this number changes in Q3 or Q4 (not sure if it's updated quarterly or semi-annually), since I fully expect a major improvement, given the number of phones shipping, as well as the great news on how popular the Motorola RAZR-i is turning out to be.