SMALL FORM FACTOR FORUM
SFF Still Riding Mobile Coattails
COLIN MCCRACKEN & PAUL ROSENFELD
Page 1 of 1
Intel’s recently announced single chip processor known under the code name Tunnel Creek, represents a huge leap forward with respect to serving the needs of the small form factor embedded x86 market. By repartitioning the processor / core logic elements into a single chip “processor” with built-in memory and video controllers, and dumping all I/O into a PCIe-linked companion chip, Intel has enabled smaller and more application-focused system designs than ever before. Does this mean that Intel finally understands the requirements of this market and is starting to design parts to meet these requirements? In order to fully answer this question, it’s important to remember where we came from.
In the late 80s and early 90s, embedded x86 designers struggled to squeeze off-the-shelf 8086, 286 and 386 processors and core logic into SFF boards such as EBX and PC/104. When Intel partnered with an embedded board company to craft an “embedded” version of the 386 (remember the 386sxi?), we all hoped that Intel was returning to its embedded roots from the pre-PC days. Alas, it was not to be. Throughout the next decade, Intel offered embedded customers standard parts designed for desktops and laptops. The one concession to embedded customers (and it was a big one) was the creation of an ‘embedded roadmap’, specifying which of the standard parts would have an extended five- or seven-year life cycle. Intel’s reluctance to deal with size or low power issues was seized upon by Cyrix / National Semiconductor / AMD with their Geode parts and by VIA Technologies with their Eden family to gain significant market share in the SFF embedded space.
In this decade, the emergence of the laptop as the dominant computing platform, and the proliferation of other mobile devices, have shined the spotlight on requirements such as small design footprint and low power consumption that dovetail nicely with embedded requirements. Hence, Intel’s desire to grow market share in the mobile space has resulted in components that match more closely with the needs of the SFF market. However, other SFF requirements (such as I/O expansion strategies) remained in conflict. But more and more, the mobile market requirements and SFF embedded requirements began to merge.
As a result of this convergence, Intel’s penetration of the embedded space with the Core Duo and Core 2 Duo product lines initiated a resurgence of Intel processor / chipset products in embedded applications. But it was the Atom processor introduced in 2008 and built for the netbook market that captured the imagination of embedded designers. This was in spite of the fact that Atom and its accompanying chipsets made virtually no concessions to the special needs of embedded applications. Atom’s success in the embedded market was the result of the further convergence of requirements for mobile (netbook) and embedded applications, not the result of any active support from Intel.
Now we have Tunnel Creek. It represents another order of magnitude improvement over the original Atom parts in terms of meeting requirements of the SFF embedded market. Tunnel Creek repartitions the traditional processor / northbridge / southbridge architecture by lumping the memory controller and video controller into the processor chip itself, and uses a standard PCI Express interface to connect to an application-specific companion chip. No proprietary FSB here! This reopens the companion chip market to third parties in a new and powerful way. Plus, a Tunnel Creek design uses some 50% less board space than earlier Atom “Menlow” solutions.
But before we welcome Intel back to the SFF embedded market, we need to look at what companion chips (IOH) Intel will offer with Tunnel Creek. And we need to ask whether any of these companion chips will really help off-the-shelf SFF board manufacturers provide an evolutionary path for the hundreds of thousands of existing applications From the potential samples displayed, the answer appears to be “not much.” The general purpose I/O expansion problem remains—as sticky as ever. And yes, there is a solution if you throw enough chips and dollars at it.
So what could Intel do to really step up to the embedded plate? A great first step would be to begin to participate actively in some of the industry standards organizations that deal with issues facing SFF embedded boards. A second would be a set of reference designs in common SFF SBC (not COM + carrier) form factors to speed the adoption process by SFF SBC board makers. And a third would be to bring a Tunnel Creek companion chip to market that includes both leading-edge and legacy I/O.