If you want to put in your diary a sobering thought for January 1st 2010 (a day when a sobering thought can be useful) it is this:
As you experience that familiar headache, dry mouth, and bloodshot eye, you will note you are in a world where the semiconductor market is worth somewhere between $206 billion and $212 billion.
Déjà vu? Well Yes, because when we woke up on January 1st 2000, we were in a year when the semiconductor industry would top $200 billion.
No growth for a decade. Why not?
Last month's Hot Chips conference at
Mark Horowitz, Stanford Professor and a leading electronics academic for 30 years said: "We're at an inflection point, you better believe it, and most of the world is in denial about it."
IBM Fellow, Brad McCredie, said: "We're done scaling. We've been playing tricks since 90 nanometers."
Tricks - like strained silicon, high k dielectrics, low K films, phase shift, double patterning etc etc have allowed the feature sizes to get smaller but they can't stop leakage.
And leakage means you can't crank up performance.
And 90 nanometer came into use around 2002 and 2003.
So, for most of this decade, the chip industry hasn't had its old, compelling, unique selling point - that the world has to buy its every new generation of its chips because the new generation chips are twice as fast as the old generation chips, use half the power, and have double the functionality.
Without that USP the chip industry has been struggling. For a start, it can't afford factories.
In 2000, 18 companies each spent over $1 billion on factories. This year only three companies will be doing that.
The list of companies on the way to fablessness gets longer and longer.
It could be the industry will stagnate or shrink until someone finds a new way of getting back onto the virtuous circle of smaller size-less power-more performance.
Materials and technologies that could get us there include graphane, graphene, nanowires, carbon nanotubes, small molecular organics, single electron, 'DNA origami', polymer and FinFETs.
Will one of those deliver a new manufacturing technology? No one knows.
But the venture capitalists aren't exactly rushing to back them.
And that's despite billions of dollars of public funding being put into developing those materials and technologies by governments from across the world.
The Holy Grail has remained stubbornly elusive for this decade.
Let's hope we'll have better luck in the next one.
This analysis assumes that the size of the market for semiconductors is related to the industry's ability to deliver increased value to customers. Unfortunately, it is not!
There are two basic factors that affect how much money you can extract from customers (i.e. market size): the first is the value you deliver, no-one will pay more than the product is worth to them, the second is what other people are charging for the same product (i.e. competition).
The problem with our industry is that the prices we achieve are almost entirely dependent on competition rather than value. We are delivering massive amounts of value to society and getting paid only a small fraction of that in return. Imagine there was only one company selling PCs, how much could they charge before it became uneconomic for a company to provide a PC for an office worker - thats the value of the PC. What the computer industry actually gets is about £250 maybe 20x less than the true value of the increased productivity.
The semiconductor market is shrinking in dollar terms because the industry has lost control of pricing and the industry has lost control of pricing because the executives at the largest chip companies have not figured out they are selling intellectual property and therefore they need to campaign for stronger and more international patent laws. Imagine if the executives in the music or film industry didn't realise they were dependent on copyright law to control pricing on their products.
If the industry wants to make more money they should use patent licenses as a way to control and inflate the price of chips. For example, suppose a chip requires technologies developed by companies A and B. Then instead of doing a free cross license - effectively valuing the intellectual property at zero - company A should charge a unit royalty a per chip and company B should charge a unit royalty b per chip. Then the minimum price of the chip would be a + b, just to cover the intellectual property - and any competitor in the market would need to pay at least a+b in patent license fees. Then when company C comes along with an improvement that everyone wants to use - it charges royalty c and the minimum chip price becomes a+b+c.
Suddenly, you have reduced scope for competition based on price - because a big fraction of the price is fixed by IP owners and you have prices that increase as the product improves: just like almost every other business on the planet.
Well, Tom that would be one way of getting prices up. It seems to me that the semiconductor industry's products are genetically programmed to commoditise because of The Learning Curve and Moore's Law, which is why a constant stream of new products is another way to keep prices up. The best way to keep prices up is to hold down capacity, but that rarely happens. So maybe yours is the best way
I agree - commoditisation is programmed into the industry's DNA because Moore's law has allowed us to be lazy with respect to extracting value. But when Moore's law stops working we need to look out for ourselves by making customers pay more. Patent law is specifically intended to prevent commoditisation of inventive products.
Hence the escalating volume of litigation in the chip industry, the emergence of the patent troll, and the enrichment of a great many lawyers. Are all these, Tom, a worthwhile price to pay for higher ASPs?
Absolutely! The selling price is the most critical thing.
I agree there are too many lawyers making too much money. That is one thing the industry should campaign about - trying to make asserting patents faster and cheaper. Right now we have a system where big companies and lawyers are motivated to keep doubling down and spinning out litigation in the expectation that smaller companies will run out of money before they receive any damages.
One thing that should be done is to drop the rules that only allow lawyers from a particular country to act in the patent office of that country. Then we could get the Indian outsourcers competing agianst the lawyers for patent preparation business.
The 'troll' epithet is just a term big companies have invented for small companies whose inventions they have stolen. Many inventions are only a small part of a complete solution and it is not realistic to expect them to be 'practiced', the only realistic mechanism for realising value is licensing.
Tom I couldn't agree with you more about the need to make patents easier, quicker and cheaper both to file and to protect. I'm just not optimistic about the will of governments to implement it.
Tom, yours suggestion is absurd.
I will take an random industry as example and make an exagerated example to it became clearer. Like cd package industry. CD packages are an dirty cheap product, isn't? We could all pay more for then, considering that an average music CD is like $15,00. What would them be w/o the package?
Well, now imagine that there are only 3 companies making it in the world. Then they have this great idea of each one of them making one inovation, like an pink balls in different parts of the package, and charge a royality 1 dollar per unit from each other. Then, the cd package, normaly sold for 10 cents for example, can now be sold for 2,10 dolars, right?
WRONG!
If each company produces 10 cd packages, then each company will receive 200 dollars of royalities, and have to pay 200 dollars of royalities to others. Then, if there is any concurrency at all, one of the companies will try to sell it's cd pacages for 2 dollars, to grab an bigger market share, as it can with an enourmous lucrativity. This will continue until we reach 11 cents per cd package. The 1 cent more will not be profit, but an added cost for inneficiency and the lawyers because this stupid system. ;)
If all the companies can agree to not compete with each other, to sustain the $2,10 price, then they could have fixed this price w/o all this patent nonsense. But now we have another problem. I suggest you to search for "Competition Law" in Wikipedia. Fortunately we have it, because monopolies are an Bad Thing (TM). Intelectual monopoly (specifically copyright law & patent law) included, most of the time.
Sorry if my english hurt your eyes. But hey, I advocate Esperanto, so don't put the blame on me.
In my post above, I assumed that the royalities profit was equally distributed between the companies. You may argue that if that is not true, then the selling price will rise proportionaly to the inequality of this distribution if every company join in this plot. Still, not the simple add-up price that Tom said. BUT (!) that is if everyone acts as Tom discribed. If an segment of the industry makes an patent-pool, then they will be able to sell their products at an lower price, grabbing market share from the other segment of the industry using the Tom's strategy.
Making the following steps in mind again, as we did earlier, adding up some imperfections, the end result will be more or less exactly what we see today. Not the abomination of the cd package industry that I described earlier, with the 20 fold increase in price that Tom predicted. Check-mate. :)
In competitive capitalism as we know, things are not sold by their use-value, but by their exchange-value, that is determined by the socially necessary labor time to make it avaiable for you (as defined by Marx). That is why, you pay little for your water, but a lot for an car. It seems that "the market" don't think that there is an need for an bigger semiconductor industry, or that it needs to have more profit right now.