Semiconductor blogging and perpetual motion: Part 1

A perpetual motion machine seems ideal.  It just goes and goes and goes.  Only one problem - friction.  So instead of going forever it will always lose energy and eventually grind to a halt.  This may happen slowly or quickly, but it will happen. The same is true for blogging.  Each time a story travels through another cycle a small amount is lost.  Like friction the bit lost may be small, almost unnoticeable, or it might be big, enough to change the meaning.   Semiconductor articles seem to fall into the latter camp as this area of knowledge is often elusive to even those tasked with writing about it.  
 
CHIP COMBAT: Samsung Building New Plant, Intel and ARM Take Notice! was the headline.  I guess it worked because I did take notice, though not in a good way.  It left the impression that Samsung is in competition with both Intel and ARM.  Intel yes, sort of, but ARM no.  Early on in this article there is a link to an earlier Reuters article covering the same topic.  While both had some odd statements one can definitely see the loss of content in the derivative article. Let’s wade into both, sort out a few points and move forward, considering a different angle on Samsung’s plans.
 
The basis for both articles is a statement that Samsung will be spending $2 billion to build a plant (fab) for logic chips.  Both further note that some of Samsung’s memory capacity will be converted to logic production. Big plans for sure, but shortly thereafter things go astray.  For starters, Reuters presents a list of “leading AP makers” that includes Samsung, Qualcomm, Texas Instruments and “mobile chip designer” ARM holdings.  The same companies are listed in the derivative article, but they are now “entrenched application processor makers”.  The Reuters article does at least describe ARM as a designer.  The problem, however, is that the wording infers ARM designs the whole chip.  It of course designs processing cores to be integrated in other designs. This might seem nit-picky, but the derivative piece simply puts ARM in the list of “entrenched application processor manufacturers”.  This time it is not quibbling it is just wrong.

The other point worth noting is the discussion around Application Processors (AP’s) and Central Processing Units (CPU’s), where CPU’s are presented as being associated with PCs.  Historically the logic chips associated with PCs were called microprocessors.  Both microprocessors and APs have CPUs as their “brains”.   In fact it is their line of CPU’s for which ARM is best known, and from which they derive the bulk of their revenue.  I should mention this discussion in the Reuters article is actually a quote from a financial analyst.  The whole thing is even more convoluted in the derivative piece.  

Finally, historically the term “Application Processor” describes the processor in a mobile phone that handles the software and graphics.  The other primary IC being the “baseband”, which handles communications.  Today Qualcomm is the undisputed leader in baseband processors.  They do also design APs but I would not necessarily call them “entrenched”.  I would have thought Nvidia has a bigger AP market share than Qualcomm.  I am not sure about the whole Windows 8 discussion, but I will not use up word count for this.  These may be minor, but Reuters calling TI fabless is a real gaff.  Yes TI has a relationship with TSMC, but I do not I think the Dallas Chamber of Commerce would call them fabless.

How do we move forward?

From the Reuters article:
Samsung's biggest challenge will be to get its mobile application chips into more devices than just its own and Apple's, particularly as the low-end smartphone market is growing at a much faster pace. "What Samsung needs to address first is to diversify its customer base and make standard chips for a variety of customers - as Qualcomm, Texas Instruments and Nvidia do," said HMC's Nho. Those firms' chips are found in products of LG Electronics, HTC and Motorola Mobility.
OK, so Samsung needs to move beyond Apple.  In engineering school there was often a min-max test.  In such a test you would put a very large and very small number into an equation to see how it behaves and see if it blows-up. Everyone seems to talk about diversification.  I would flip this around and push the equation the other way.  Apple apparently consumes over 50% of Samsung’s current logic wafer starts. That is a large number and one that would require a lot of diversification should Apple leave.  Somewhere in here might be a fatal blow to the rumours of Apple’s imminent departure to either TSMC or Intel.  Apple may decide to second source their expanding A(x) processor fab needs, but I would guess the bulk is sticking with Samsung.