People ask me is there something wrong with the semiconductor industry if apparently good companies with good technologies such as SiliconBlue Technologies Inc. and PicoChip Ltd. are being sold for less than the venture capital invested in them over several years.
I don't think there is anything fundamentally wrong with the semiconductor industry – but it is in a process of resetting approaches to the market and technology, expectations of investors and transforming itself from something that was just hitching a ride down the escalator of Moore's Law in a race towards the bottom.
What had happened is that many protagonists had got hooked on Moore's Law as a way to get superior performance and reduced unit cost and therefore as a way of making sales by displacing their rivals. This made sense in the past and started in a time when chip companies all made similar building-block components. So, for those chip companies back in the day, the added value was not in the design but in the manufacturing expertise.
And indeed years ago the miniaturization that resulted from Moore's Law always seemed to bring economic success. The faster an integrated device manufacturer could get to the next manufacturing process node the more quickly a component could be designed-in through lower pricing and the faster a rival's component could be designed out. And it was also accompanied by exponential increases in the global electronics and semiconductor markets.
But now we are getting very close the atomic bottom of that escalator and the non-recurring engineering (NRE) costs of research and production are going up far more than can be routinely justified by the economic benefit of working at those leading-edge geometries. The few exceptions are the very high volume mass-produced devices such as memories, processors, FPGAs and some high volume system-chips.
And another flip around is that for IDMs and the fabless the added value is now more in the design than the manufacturing. Most companies are more or less able to get to the same CMOS process node and what makes a difference in sales is likely to be a superior design, or at least a design-win with a big system partner which guarantees the initial volume of chips.
The post is very informative. It is a pleasure reading it. I have also bookmarked you for checking out new posts.
MEMs integration into CMOS will be the next big opportunity. Stand-alone MEMS in MCM or other exotic packaging may be cost prohibitive unless there are no options for such integration for performance or complexity reasons. MEMs oscillators especially could be a dying business since CMOS based XTAL osc replacement is possible (e.g. IDT or Silicon Labs). The performance gap has narrowed significantly for such oscillators compared to XTAL.
The semiconductor industry is 50+ years old and is driven by theoretical knowledge and open sharing of it. The industry is still rising but that is happening in Taiwan, So. Korea. Both manufacturing and R&D expertise have now dispersed quite widely to enable much potential for profit in the developed West. They attract capital not just from their own but also from investors in the mature economies. Thus the inversion in investment - payout ratio mentioned by Peter re: silicon blue.
I thought that the migration to multicore CPUs already showed us that we were getting to the point of diminishing returns with Moore's Law, no? Clock speeds have been hovering around 3 GHz for quite some time now. At least 10 years, I'd say.
Another not-new trend is the ever increasing presence of uPs in all manner of consumer products. But even that trend started way back, with the initial introduction of single-chip computers. The uPs just get more and more pervasive.
But I don't see how this translates to the semiconductor industry rising again? I guess I'm saying, where is the discontinuity? Presumably, the few fabs which can produce the large scale integration chips economically will continue to see brisk demand, as their chips find their way into more products. Which has been happening for quite some time now. New ideas will certainly emerge, for even more clever ways of making better use of uPs, but it would take something more "disruptive" to reverse the trend in fabs?
I think "How the semiconductor industry will rise again" is a more fitting title for this article. Another path is to the dramatic increase the pervasiveness of ICs. For example EETimes has reported "The internet of things" earlier. By embedding a microprocessor and communication capability into everything from coffee makers to mattresses, the demand for semiconductor products will increase.
The industry is monopolized with a few giants waddling down a single path. None of them need any innovation.
On ther other hand, they present juicy targets for patent trolls. That is the only kind of start up that may do OK.
If I remember correctly, Silicon Blue's secret sauce was the optimization of its processor for low power (in HW, not SW), more than the lowest-node advantage. Still, Lattice did mention that the acquisition gave them (Lattice) a way down to 22nm that was missing with their own product line.
The industry has essentially experienced a split into two segments: more Moore and more than Moore...more Moore still exists, IC designs still move to 22nm, Intel, Arm, Broadcom etc but there are fewer and fewer players around as it gets more expensive to design...more than Moore requires putting some smarts into play, MEMs is only one example, there are many others, 0.35um process might be quite sufficient here, there is still some room for VC funding but the good, old days of late 90-ties will NOT return...Kris
I could have guessed that you might like my opinion on this. But it is more complex than digital bad; analog plus MEMS good.
We will still need digital, and big digital at that. But the economics have changed.
If we get on to beyond-CMOS, or 3-D processing, or quantum computing; the economics will change again.
But it always comes down to innovation and its cost, the functional value created and protectability.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.