Further examples and illustrations

This brings us to another comparison we alluded to earlier: the Nokia N93 and the Samsung P920. Both are aimed at quite different markets, the former focuses most on its camera and gaming capabilities, while the latter was a product made specifically for the world cup in May 2006, and sported DVB-H functionality. So, the focus is different, and the apparent technology is also quite different - but the approaches taken still bear some real resemblance.

The N93 is powered by the TI OMAP2420 SoC, while the Samsung P920 uses the NVIDIA GoForce 5500 and an unknown ARM processor, also obviously manufactured by Samsung. The OMAPs have high-end ARM processors, but do not use them much if at all for things like video decoding; instead, they use on-chip TI-designed DSPs, apparently with enhanced instruction sets for imaging and video. The GoForce 5500 seems to be more highly featured (although that's debatable), but it also doesn't include any ARM core, which implies a lower practical level of integration in the eyes of handheld manufacturers.

But excluding the lack of an ARM core, the GoForce 5500's internals aren't that different from OMAP's - in theory, at least. The 5500 seems based on a number of Tensilica instruction-enhanced DSPs, unlike its predecessors which look to have been much heavier on fixed-function blocks for video decoding. Tensilica only publicizes their IP's importance for the audio processor of the chip, but it does look like NVIDIA's approach for video was roughly similar, although with maybe more handmade blocks, leveraging their earlier R&D in that area.

Thus, at a finer level, both companies had relatively different technologies at their disposal, but the approaches taken are still quite similar when looking at the bigger picture; general-purpose DSPs are taken and extended for certain target purposes in order to improve both performance-per-mm2 and power efficiency. Other things, such as 3D and JPEG processing, are completely independent and mostly fixed-function.

It's easy to see that this kind of design is much more high-budget than that of smaller competitors, and also possibly a lot more efficient (per-watt, and possibly also per-mm2) than just gluing some ARM cores together and writing some custom software. And the results seem to show: the GoForce 5500 takes only 20 milliwatts for MP3 decoding, while PortalPlayer's last-generation chips used in the 5G iPod can take up to 40 milliwatts for the same task. This could be slightly exaggerated, however, because the latter is manufactured on a slightly older process.

To illustrate the software advantage doing part of the work on programmable cores has, you don't have to go much further than the spec sheet of the 5500; it has fairly impressive specifications for H.264, yet rather average ones for WMV. This would likely be explained by the lack of special-purpose acceleration for that codec. But at least, they get basic and likely "good enough" support in there that way - otherwise, they might have had nothing.

Conclusion

So that's it. This should have given you a small taste of some of the various dynamics involved in the design and sale of handheld chipsets in general, as well as how these factors both differ and resemble those of other semiconductor-based businesses. Hopefully you've enjoyed the read, and gained some insight that'll let you consider this industry segment from a new perspective.

The coming years are very likely to be some of the most interesting ones in the history of the handheld industry. A much wider variety of features and capabilities are about to go mainstream, but we are not anywhere near the point where competing products no longer offer significant differentiation. And that's good for the industry, because it means there's plenty of value to add for those that can deliver innovative feature-sets or higher efficiency. Those are some very exciting times ahead.

Comments

If you want to comment on the industry as a whole, drop by our handheld discussion forum, and there's always the comments thread for this very article here.