The cellular modem: an in-house technology that's increasingly golden

I realize that I was somewhat critical of Nvidia's ARM-based SOC product line in my previous writeup. To be clear, my comments predominantly reflected the past history and current status of the various Tegra families. The bulk of the existing announced Tegra 2 design wins, for example, as well as those for the more recent Tegra 3 (aka Kal-El), comprise Android-based tablets that to date have dramatically under-performed their iPad counterparts containing Apple-labeled SOCs. Arguably, the responsibility for this Android malaise predominantly lies with ecosystem players other than Nvidia; Google prematurely rushed tablet-only Android 3.0 'Honeycomb' to market, for example, neither it or its 3.x successors ever achieved open-source status, and the Android 4.0 'Ice Cream Sandwich' successor has been slow to ramp into the market. And the tablet OEMs haven't done a good job of differentiating their offerings from the iPad, specifically with features that matter to potential purchasers.

Nvidia isn't blameless, however. As well-known Google software engineer Tim Bray alluded in an early-2011 blog post ("Tegra 2 is wonderful as long as your application has the characteristics of Doom"), the Tegra architecture, specifically its proprietary GPU scheme, isn't particularly easy to efficiently program. The reason why Tegra 2 remained restricted to high-end phones was, I suspect, fundamentally due to Nvidia's inability to quickly proliferate the family into multiple (specifically lower cost) price-versus-feature set variants. And it remains to be seen how successful the ARM-compiled variant of Microsoft's upcoming Windows 8 operating system will be. Frankly, on the latter point, I'm skeptical, particularly considering how battery-friendly and performance-robust Intel's latest Atom-based x86 competitors are, both in their current 32 nm- and upcoming 22 nm-fabricated forms.

But Nvidia has an 'ace in the hole' that may still ensure its long-term success in the mobile communications and computing space, potentially leapfrogging currently more successful competitors in the process. It's the cellular baseband, which processes both data and voice traffic, and a technology that Nvidia obtained via the acquisition of Icera announced last summer. Remember how, in my prior writeup, I discussed how Intel obtained graphics processor technology via a Real3D partnership and subsequent patent purchase (once Real3D closed its doors)? How AMD obtained its graphics technology via the acquisition of ATI Technologies? And how, in both cases, the standalone graphics processor has eventually ended up integrated on the CPU? I believe that for future cellular handsets and cellular-inclusive tablets, baseband integration capability within the SOC will be equally crucial to semiconductor supplier success. And it's an area where Nvidia is now in a comparatively strong position versus key competitors.

Who else has in-house cellular baseband expertise? Qualcomm, of course; the company was a cellular technology and product developer first and foremost, before it ever expanded into the ARM realm via an architecture (i.e. instruction set) license. Broadcom, too, although its ARM-based SOCs are not particularly leading-edge. And Intel as well, ironically, via its recent acquisition of Infineon. Although the Infineon wireless assets purchase gave Intel not only cellular baseband products, expertise and patents but also access once again to an ARM architecture license, I personally doubt that we'll ever see any more ARM-based SOCs from the company, which seems single-mindedly determined to drive x86 into currently ARM-dominated portable device spaces. But Marvell, another leading ARM SOC developer (and the recipient of Intel's former ARM architecture license via its mid-2006 acquisition of Intel's XScale assets), does not to the best of my knowledge have any in-house cellular baseband capabilities. Nor does Apple or Samsung, as far as I'm aware.

Nor does Texas Instruments, although it used to...TI decided in 2008 to phase out of the cellular baseband business, although it remains a notable ARM SOC supplier via its OMAP product line. And nor does Freescale (any more, at least), either, for the same fundamental reason; the company began selling its cellular chipset business, both in product and patent portfolio forms, several years ago. At last year's Freescale Technology Forum, a company spokesperson strove to reassure me that the i.MX ARM applications processor product line was not tangibly hampered in the near term, thanks to the availability of discrete baseband offerings from other companies. However, he flatly said, once integrated cellular baseband functionality became a SOC requirement, a sooner-or-later inevitable outcome which Qualcomm in particular is aggressively pushing toward, it would lead to market opportunities in which Freescale would no longer be able to effectively participate. Although the cellular baseband spin-offs and shut-downs may have been short-term fiscally attractive for both TI and Freescale, I wonder if both companies are now seeing their past decisions as "penny wise, pound foolish" moves.