Technology wins out over all
Intel, GloFo, and TSMC battle for position as the market moves to mobile
Big companies, big smart companies, have the ability to take a body block and keep going. Cash reserves, tons of engineers, and determination are their muscle. And like any company, they make false starts, and unsupportable claims. That delights the press, and short-siders, but doesn’t really add much to the ecosystem.
This week Intel revealed its latest power core design, Silvermont, fabricated on its 22nm FinFET process, a variant of the 22nm process it started to use last year for Ivy Bridge, but adapted for use in SOC applications. Intel decided in 2011 to sync up its process technology across all products, which meant the Atom family would make some giant leaps. Its Bay Trail platform, due later this year and targeted at Windows and Android tablets, will be one of the first SOCs to include this new core. Intel claims that Silvermont will deliver three times the performance of the prior 32nm Atom cores at the same power, or deliver the same performance as the earlier core at one-fifth the power.
Bay Trail, which is the platform name for the Valleyview SoCs, and contains Silvermont cores that will go in it, is the first instance of that happening. Intel will offer three versions of Valleyview processors: Bay Trail-T (3W), Bay Trail-M (4W – 6.5W) and Bay Trail-D (12W) for tablets, notebooks, and desktops, respectively.
Bay Trail is Intel’s next-generation tablet offering scheduled to be available in Holiday 2013 tablets. Intel’s “Merrifield” is scheduled to ship to customers by the end of this year, and Avoton and Rangeley are scheduled for the 2H of 2013.
What this means to the industry is Intel intends, with all its determination, to take a big hunk of the mobile SoC (or AP if you prefer) market. That means MediaTek, Qualcomm, and others have a serious, super powerful competitor to deal with.
And this is a formable competitor. Based on the numbers Intel gave out (which BTW are right in line with what the company promised over two years ago) the Silvermont core will be 2X better or more (depending on the parameter you like to look at) than Saltwell. In addition, Intel is going to use its own GPU design, based on the Haswell, and so that may reduce the cost of their chips a bit.
We won’t really know how the market will embrace the Bay Trail chips, but Intel’s claims of using less power to obtain the same results as the competition (and the inverse) are pretty compelling – however, the charts had little detail and no scales.
Intel can run Windows 8 or Android, so OEM customers have a choice (and may deploy products using each one, and let the consumer choose).
I said before, and still believe it is true, in this industry the best transistor wins. Going forward it will get even truer. Here’s Intel’s advantage and the pay-off for the decades of huge CapEx (while still delivering good to great profits). I think Intel is at least a year ahead of TSMC, and more than two of GF. Depending on your metric and company bias, you may want to debate that, so here are some of my proof-points.
Intel has been shipping high-volume Tri-Gate products at 22nm since 2011 and has shown samples of 14nm. TSMC said last year that its 20nm technology would begin volume production in 2013. At its investors meeting in April, TSMC said their 16nm FinFET would enter mass production within one year after 20nm ramp-up, and moved its 20nm process to risk production. The company is talking about 10nm showing up in 2015. Globalfoundries said at their Common Platform conference that they would have 20nm in 2013 and a quick follow up on 14nm in 2014, which is when the company will introduce their first FinFET transistor. Like TSMC they mentioned 10nm for late 2015.
Now not all transistors are built for mobile devices, but the size and growth of the market is so astounding everyone wants in. However, only two companies can built SoCs in their own fab, so all the fabless SoC builders have to go TSMC or GloFo (as Globalfoundries is being called, because the real name is just too long and uses too many mouth muscles to say).
Big volume shipper Apple is a key. It’s unlikely they will switch from their A-series ARM-based processors and adopt Intel’s processor. And confident rumors have it all of Apple’s SoC production will be done at TSMC in 2014, at TSMC’s 20nm Fab 14. Qualcomm, the biggest SoC shipper is already using TSMC. Obviously, those two heavyweights aren’t going to stand idly by and watch other firms outsell them with Intel’s superior chips. As a result, TSMC has revised its CapEx for 2013 to $9.5-10 billion (from the $9 billion set last year) to try and stay at pace with Intel.
Next entry: Nvidia winner in Q1, AMD flat, Intel down